Question on tuning

Posted in 2017

Summary

Larry reported that a Java record-processing job ran 80-100 transactions/hour on an old Sun V490 but only 25-35/hour on a "newer" T5240 (UltraSPARC T2+ 1.4GHz) with identical IDS 11.50.FC5 onconfig and kernel settings, with CPUs 95-98% idle on the new box. Responders questioned whether the T2+ is really faster (slow cores, many threads) and suggested hunting the bottleneck: check I/O and KAIO (onstat -g ioq, iostat), sysmaster.syssesprof, onstat -u/-g ath/-g ses for sessions waiting in netnorm (i.e. on the Java client), onstat -g glo per-CPU use, -p, -C, -g ckp, update statistics, SET EXPLAIN plans, SQLTRACE/syssqltrace for lock waits, row vs page locking, and testing a non-database program. No resolution is recorded in the thread.

Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.

Topics: Performance & Tuning, Server Administration, Platform-Specific Issues, Java & JDBC Development

LARRY SORENSEN — Tue, 02 May 2017 03:37:56 — source: IIUG Forums & Mailing Lists

Solaris 10 IDS 11.50.FC5 Workgroup SPARC 64 bit I have two servers, one older server and one newer server. The newer server is obviously more powerful. I have an involved process that processes user records on the server. The process uses mostly Java to connect to the database to process records. The Java application runs on the database server itself. The onconfig files are identical on both servers with the exception of the SERVERNUM. I have set up projects on both servers specifying max-shm-memory, max-file-descriptor, etc. that are identical for the kernel parameters. On the old server, we could depend upon 80-100 record transactions to be processed per hour. During that time the server CPUs would remain 15%-25% idle (so almost to capacity). On the new server, we are seeing 25-35 record transactions being processed per hour and the server CPUs remain 95%-98% idle. What are some tuning parameters and things we can look at and tweak to help the new server reach its potential, or at least process as fast as the older server? Larry

↪ replying to LARRY SORENSEN

Frank Langelage — Tue, 02 May 2017 07:11:06 — source: IIUG Forums & Mailing Lists

On 02.05.17 05:37, LARRY SORENSEN wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer server is > obviously more powerful. > Which server models you're using?

↪ replying to LARRY SORENSEN

Fernando Nunes — Tue, 02 May 2017 10:29:16 — source: IIUG Forums & Mailing Lists

Larry, Unfortunately tuning is not that simple... Typically and in abstract we could consider performance to depend on three factors: - Memory - CPU - I/O Apparently CPU became more idle and I assume the memory is equal or higher and the use is the same. So, in a blind shot I'd say you may have an I/O problem. But keep in mind that I'm just trying to simplify what I begun to say it's not "that simple". So, you really need to do more monitoring and identify where's the bottleneck. Regards. On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --001a113f19a2d16bb5054e880247

↪ replying to Fernando Nunes

david@smooth1.co.uk — Tue, 02 May 2017 10:43:50 — source: IIUG Forums & Mailing Lists

Check sysmaster.syssesprof, are the sessions doing a lot more logical or physical I/O on the new machine? Check onstat -g ioq are both machines using KAIO (kio) or not KAIO (aio)? Have you run the usual update statistics on the new machine? Check with "SET EXPLAIN ON" - are query plans different on the new machine? Also you can enable SQLTRACE with high on the new machine and check the syssqltrace,syssqltrace_iter,syssqltrace_hvar tables to see where the time is being spent. This includes lock wait time in syssqltrace.sql_lockwaits and syssqltrace.sql_lockwttime Regards, David. > On 02 May 2017 at 11:29 Fernando Nunes <domusonline@gmail.com> wrote: > > > Larry, > Unfortunately tuning is not that simple... Typically and in abstract we > could consider performance to depend on three factors: > - Memory > - CPU > - I/O > > Apparently CPU became more idle and I assume the memory is equal or higher > and the use is the same. > So, in a blind shot I'd say you may have an I/O problem. But keep in mind > that I'm just trying to simplify what I begun to say it's not "that simple". > So, you really need to do more monitoring and identify where's the > bottleneck. > > Regards. > > On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > > > Solaris 10 > > > > IDS 11.50.FC5 Workgroup > > > > SPARC 64 bit > > > > I have two servers, one older server and one newer server. The newer > > server is > > obviously more powerful. > > > > I have an involved process that processes user records on the server. The > > process uses mostly Java to connect to the database to process records. The > > Java application runs on the database server itself. > > > > The onconfig files are identical on both servers with the exception of the > > SERVERNUM. > > > > I have set up projects on both servers specifying max-shm-memory, > > max-file-descriptor, etc. that are identical for the kernel parameters. > > > > On the old server, we could depend upon 80-100 record transactions to be > > processed per hour. During that time the server CPUs would remain 15%-25% > > idle > > (so almost to capacity). > > > > On the new server, we are seeing 25-35 record transactions being processed > > per > > hour and the server CPUs remain 95%-98% idle. > > > > What are some tuning parameters and things we can look at and tweak to help > > the new server reach its potential, or at least process as fast as the > > older > > server? > > > > Larry > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently... > > --001a113f19a2d16bb5054e880247 > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. >

↪ replying to LARRY SORENSEN

Art Kagel — Tue, 02 May 2017 10:47:08 — source: IIUG Forums & Mailing Lists

You THINK the new system is more powerful, but with Sun systems it is a definite possibility that it is not. What are the models and processor types for the two systems? Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --f403043a7ae8045e86054e88417d

↪ replying to Frank Langelage

LARRY SORENSEN — Tue, 02 May 2017 12:04:58 — source: IIUG Forums & Mailing Lists

The new server is a T5240. I am trying to get the model of the older server, but it is offline at the moment. I will send it soon. Larry ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Frank Langelage <frank@lafr.de> Sent: Tuesday, May 2, 2017 1:11 AM To: ids@iiug.org Subject: Re: Question on tuning [39070] On 02.05.17 05:37, LARRY SORENSEN wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer server is > obviously more powerful. > Which server models you're using? ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to Fernando Nunes

LARRY SORENSEN — Tue, 02 May 2017 12:06:28 — source: IIUG Forums & Mailing Lists

What are some of the commands that I can use to monitor on a regular basis? ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Fernando Nunes <domusonline@gmail.com> Sent: Tuesday, May 2, 2017 4:29 AM To: ids@iiug.org Subject: Re: Question on tuning [39071] Larry, Unfortunately tuning is not that simple... Typically and in abstract we could consider performance to depend on three factors: - Memory - CPU - I/O Apparently CPU became more idle and I assume the memory is equal or higher and the use is the same. So, in a blind shot I'd say you may have an I/O problem. But keep in mind that I'm just trying to simplify what I begun to say it's not "that simple". So, you really need to do more monitoring and identify where's the bottleneck. Regards. On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38/S220- s75/fnunes.jpg]<http://informix-technology.blogspot.com/>; Informix Technology<http://informix-technology.blogspot.com/>; informix-technology.blogspot.com IBM and HCL entering a strategic partnership to jointly develop and market the IBM Informix family of products (original version here: http://informix-technology ... My email works... but I don't check it frequently... --001a113f19a2d16bb5054e880247 ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to Art Kagel

LARRY SORENSEN — Tue, 02 May 2017 12:22:03 — source: IIUG Forums & Mailing Lists

The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ The older server is a V490, not sure the processors ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel <art.kagel@gmail.com> Sent: Tuesday, May 2, 2017 4:47 AM To: ids@iiug.org Subject: Re: Question on tuning [39073] You THINK the new system is more powerful, but with Sun systems it is a definite possibility that it is not. What are the models and processor types for the two systems? Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com<http://www.askdbmgt.com>; ASK Database Management - Home<http://www.askdbmgt.com/>; www.askdbmgt.com Database Consulting and Support ... This is the site for Art S. Kagel's consultancy. The soaring majesty and beauty in the image above hides the complex ecology and ... Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --f403043a7ae8045e86054e88417d ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to LARRY SORENSEN

Andreas Legner — Tue, 02 May 2017 12:27:32 — source: IIUG Forums & Mailing Lists

I'd really first try to establish whether the time is lost on Informix=20 server side, or whether it's the java application and/or its communication = to the server and the server spends a lot of its time waiting on the=20 client. -> you'd monitor the java application's server session -> if finding=20 that in "cond wait netnorm" a lot, i.e. waiting on the client application, = it's not the server causing the slowness. -> look at thread states (onstat -g ath, or thread part of onstat -g ses=20 <sess=5Fid>). Stating that cpu consumption is lower on new box, would this mean none of=20 the cpus is heavily used? You could have a situation where what you (your = application) is doing doesn't occupy many cpus, but those which it does=20 occupy are heavily used. -> cpu consumption in onstat -g glo. Andreas From: "LARRY SORENSEN" <LSORENSEN25@msn.com> To: ids@iiug.org Date: 02.05.2017 14:08 Subject: Re: Question on tuning [39075] Sent by: ids-bounces@iiug.org What are some of the commands that I can use to monitor on a regular=20 basis?=20 =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F= =5F=5F=5F=5F=5F=5F=5F=20 From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Fernando=20 Nunes=20 <domusonline@gmail.com>=20 Sent: Tuesday, May 2, 2017 4:29 AM=20 To: ids@iiug.org=20 Subject: Re: Question on tuning [39071]=20 Larry,=20 Unfortunately tuning is not that simple... Typically and in abstract we=20 could consider performance to depend on three factors:=20 - Memory=20 - CPU=20 - I/O=20 Apparently CPU became more idle and I assume the memory is equal or higher = and the use is the same.=20 So, in a blind shot I'd say you may have an I/O problem. But keep in mind=20 that I'm just trying to simplify what I begun to say it's not "that=20 simple".=20 So, you really need to do more monitoring and identify where's the=20 bottleneck.=20 Regards.=20 On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com>=20 wrote:=20 > Solaris 10=20 >=20 > IDS 11.50.FC5 Workgroup=20 >=20 > SPARC 64 bit=20 >=20 > I have two servers, one older server and one newer server. The newer=20 > server is=20 > obviously more powerful.=20 >=20 > I have an involved process that processes user records on the server.=20 The=20 > process uses mostly Java to connect to the database to process records.=20 The=20 > Java application runs on the database server itself.=20 >=20 > The onconfig files are identical on both servers with the exception of=20 the=20 > SERVERNUM.=20 >=20 > I have set up projects on both servers specifying max-shm-memory,=20 > max-file-descriptor, etc. that are identical for the kernel parameters.=20 >=20 > On the old server, we could depend upon 80-100 record transactions to be = > processed per hour. During that time the server CPUs would remain=20 15%-25%=20 > idle=20 > (so almost to capacity).=20 >=20 > On the new server, we are seeing 25-35 record transactions being=20 processed=20 > per=20 > hour and the server CPUs remain 95%-98% idle.=20 >=20 > What are some tuning parameters and things we can look at and tweak to=20 help=20 > the new server reach its potential, or at least process as fast as the=20 > older=20 > server?=20 >=20 > Larry=20 >=20 >=20 > ************************************************************=20 > *******************=20 > Forum Note: Use "Reply" to post a response in the discussion forum.=20 >=20 >=20 --=20 Fernando Nunes=20 Portugal=20 http://informix-technology.blogspot.com=20 [ http://4.bp.blogspot.com/=5FowXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38= /S220-s75/fnunes.jpg ]<http://informix-technology.blogspot.com/>=20 Informix Technology<http://informix-technology.blogspot.com/>=20 informix-technology.blogspot.com=20 IBM and HCL entering a strategic partnership to jointly develop and market = the=20 IBM Informix family of products (original version here:=20 http://informix-technology ...=20 My email works... but I don't check it frequently...=20 --001a113f19a2d16bb5054e880247=20 ***************************************************************************= ****=20 Forum Note: Use "Reply" to post a response in the discussion forum.=20 ***************************************************************************= ****=20 Forum Note: Use "Reply" to post a response in the discussion forum.=20

↪ replying to LARRY SORENSEN

Marcus Haarmann — Tue, 02 May 2017 12:34:02 — source: IIUG Forums & Mailing Lists

Hi, Some hints: first thing is to look at the database itself. How was the transfer to the new machine done ? Do you just link to the same DBSpaces on a SAN ? Or was there a kind of backup/restore process or unload/load ? Did you upgrade to a newer version ? What about your statistics ? Are the chunks created the same way as on the old machine (e.g. did you work with raw chunks on the old machine and now you might have cooked files ?) In case you activated a newer IDS release: did you run "update statistics low drop distributions" ? Are your tables/indexes on dbSpaces with same page size as before (if not: do you have a bufferpool for the new page size ?) Monitor the checkpoints using onstat -g ckp Check the Caching using onstat -p Are the indexes balanced or does the background btree scanner very active (check onstat -C output) Next look at the queries: You should check the query behaviour by using "set explain on" (which can be done dynamically using onmode -Y <sid> 1|2) This will produce a detailed output about the query plans. Look for sequential scans where you do not expect sequential access (statistics are wrong or an index is maybe missing or a foreign key constraint is missing). Look at the machine: Check your I/O using iostat (or whatever the command is on Solaris) What partitions (chunks) are busy during checkpoints or when the program is running ? Do you have read or write activity to search for ? Check the disk access using dd command Marcus Haarmann Von: "LARRY SORENSEN" <LSORENSEN25@msn.com> An: "ids" <ids@iiug.org> Gesendet: Dienstag, 2. Mai 2017 14:06:37 Betreff: Re: Question on tuning [39075] What are some of the commands that I can use to monitor on a regular basis? ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Fernando Nunes <domusonline@gmail.com> Sent: Tuesday, May 2, 2017 4:29 AM To: ids@iiug.org Subject: Re: Question on tuning [39071] Larry, Unfortunately tuning is not that simple... Typically and in abstract we could consider performance to depend on three factors: - Memory - CPU - I/O Apparently CPU became more idle and I assume the memory is equal or higher and the use is the same. So, in a blind shot I'd say you may have an I/O problem. But keep in mind that I'm just trying to simplify what I begun to say it's not "that simple". So, you really need to do more monitoring and identify where's the bottleneck. Regards. On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38/S220- s75/fnunes.jpg]<http://informix-technology.blogspot.com/>; Informix Technology<http://informix-technology.blogspot.com/>; informix-technology.blogspot.com IBM and HCL entering a strategic partnership to jointly develop and market the IBM Informix family of products (original version here: http://informix-technology ... My email works... but I don't check it frequently... --001a113f19a2d16bb5054e880247 ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to Marcus Haarmann

LARRY SORENSEN — Tue, 02 May 2017 12:49:48 — source: IIUG Forums & Mailing Lists

The server versions are identical. To move the data over, the space was set up identically, RAW, and we did an ontape -s -L STDIO -F and copied it directly to the new server.. Update statistics has been run on both the databases in the instance. ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Marcus Haarmann <marcus.haarmann@midoco.de> Sent: Tuesday, May 2, 2017 6:34 AM To: ids@iiug.org Subject: Re: Question on tuning [39078] Hi, Some hints: first thing is to look at the database itself. How was the transfer to the new machine done ? Do you just link to the same DBSpaces on a SAN ? Or was there a kind of backup/restore process or unload/load ? Did you upgrade to a newer version ? What about your statistics ? Are the chunks created the same way as on the old machine (e.g. did you work with raw chunks on the old machine and now you might have cooked files ?) In case you activated a newer IDS release: did you run "update statistics low drop distributions" ? Are your tables/indexes on dbSpaces with same page size as before (if not: do you have a bufferpool for the new page size ?) Monitor the checkpoints using onstat -g ckp Check the Caching using onstat -p Are the indexes balanced or does the background btree scanner very active (check onstat -C output) Next look at the queries: You should check the query behaviour by using "set explain on" (which can be done dynamically using onmode -Y <sid> 1|2) This will produce a detailed output about the query plans. Look for sequential scans where you do not expect sequential access (statistics are wrong or an index is maybe missing or a foreign key constraint is missing). Look at the machine: Check your I/O using iostat (or whatever the command is on Solaris) What partitions (chunks) are busy during checkpoints or when the program is running ? Do you have read or write activity to search for ? Check the disk access using dd command Marcus Haarmann Von: "LARRY SORENSEN" <LSORENSEN25@msn.com> An: "ids" <ids@iiug.org> Gesendet: Dienstag, 2. Mai 2017 14:06:37 Betreff: Re: Question on tuning [39075] What are some of the commands that I can use to monitor on a regular basis? ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Fernando Nunes <domusonline@gmail.com> Sent: Tuesday, May 2, 2017 4:29 AM To: ids@iiug.org Subject: Re: Question on tuning [39071] Larry, Unfortunately tuning is not that simple... Typically and in abstract we could consider performance to depend on three factors: - Memory - CPU - I/O Apparently CPU became more idle and I assume the memory is equal or higher and the use is the same. So, in a blind shot I'd say you may have an I/O problem. But keep in mind that I'm just trying to simplify what I begun to say it's not "that simple". So, you really need to do more monitoring and identify where's the bottleneck. Regards. On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Solaris 10 > > IDS 11.50.FC5 Workgroup > > SPARC 64 bit > > I have two servers, one older server and one newer server. The newer > server is > obviously more powerful. > > I have an involved process that processes user records on the server. The > process uses mostly Java to connect to the database to process records. The > Java application runs on the database server itself. > > The onconfig files are identical on both servers with the exception of the > SERVERNUM. > > I have set up projects on both servers specifying max-shm-memory, > max-file-descriptor, etc. that are identical for the kernel parameters. > > On the old server, we could depend upon 80-100 record transactions to be > processed per hour. During that time the server CPUs would remain 15%-25% > idle > (so almost to capacity). > > On the new server, we are seeing 25-35 record transactions being processed > per > hour and the server CPUs remain 95%-98% idle. > > What are some tuning parameters and things we can look at and tweak to help > the new server reach its potential, or at least process as fast as the > older > server? > > Larry > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38/S220- s75/fnunes.jpg]<http://informix-technology.blogspot.com/>; Informix Technology<http://informix-technology.blogspot.com/>; informix-technology.blogspot.com IBM and HCL entering a strategic partnership to jointly develop and market the IBM Informix family of products (original version here: http://informix-technology ... [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38/S220- s75/fnunes.jpg]<http://informix-technology.blogspot.com/>; Informix Technology<http://informix-technology.blogspot.com/>; [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/AAAAAAAAABc/AlV-RTx0M38/S220- s75/fnunes.jpg]<http://informix-technology.blogspot.com/>; Informix Technology<http://informix-technology.blogspot.com/>; informix-technology.blogspot.com IBM and HCL entering a strategic partnership to jointly develop and market the IBM Informix family of products (original version here: http://informix-technology ... informix-technology.blogspot.com IBM and HCL entering a strategic partnership to jointly develop and market the IBM Informix family of products (original version here: http://informix-technology ... My email works... but I don't check it frequently... --001a113f19a2d16bb5054e880247 ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to LARRY SORENSEN

Fernando Nunes — Tue, 02 May 2017 13:29:38 — source: IIUG Forums & Mailing Lists

Someone already suggested some queries on sysmaster. Apart from that, some very basic process is to check an onstat -u of the session(s) you have dealing with the transactions. You could also (if the sessions are permanent or if they stay there long enough) try to activate the sqltrace facility for a session and try to capture the queries it run. Meanwhile I forgot another important factor: Network In my defense I could advocate it's just another form of I/O, but we usually associate I/O with hard disk / storage, so let's consider network a different factor. By any chance did you previously had the application and database on the same server and now they've been split? Two other possibilities to consider: - Some queries are doing full scans or using bad indexes (check the statistics, but a bug is also an option) - Locks (someone also mentioned this): Check that your tables are using row level locking and not page. Check that your sessions are not waiting on locks for too long Regards. On Tue, May 2, 2017 at 1:06 PM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > What are some of the commands that I can use to monitor on a regular basis? > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Fernando > Nunes > <domusonline@gmail.com> > Sent: Tuesday, May 2, 2017 4:29 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39071] > > Larry, > Unfortunately tuning is not that simple... Typically and in abstract we > could consider performance to depend on three factors: > - Memory > - CPU > - I/O > > Apparently CPU became more idle and I assume the memory is equal or higher > and the use is the same. > So, in a blind shot I'd say you may have an I/O problem. But keep in mind > that I'm just trying to simplify what I begun to say it's not "that > simple". > So, you really need to do more monitoring and identify where's the > bottleneck. > > Regards. > > On Tue, May 2, 2017 at 4:37 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > wrote: > > > Solaris 10 > > > > IDS 11.50.FC5 Workgroup > > > > SPARC 64 bit > > > > I have two servers, one older server and one newer server. The newer > > server is > > obviously more powerful. > > > > I have an involved process that processes user records on the server. The > > process uses mostly Java to connect to the database to process records. > The > > Java application runs on the database server itself. > > > > The onconfig files are identical on both servers with the exception of > the > > SERVERNUM. > > > > I have set up projects on both servers specifying max-shm-memory, > > max-file-descriptor, etc. that are identical for the kernel parameters. > > > > On the old server, we could depend upon 80-100 record transactions to be > > processed per hour. During that time the server CPUs would remain 15%-25% > > idle > > (so almost to capacity). > > > > On the new server, we are seeing 25-35 record transactions being > processed > > per > > hour and the server CPUs remain 95%-98% idle. > > > > What are some tuning parameters and things we can look at and tweak to > help > > the new server reach its potential, or at least process as fast as the > > older > > server? > > > > Larry > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > > > [http://4.bp.blogspot.com/_owXf8TIBUXI/S2bpGijdAWI/ > AAAAAAAAABc/AlV-RTx0M38/S220-s75/fnunes.jpg]<http:// > informix-technology.blogspot.com/> > > Informix Technology<http://informix-technology.blogspot.com/>; > informix-technology.blogspot.com > IBM and HCL entering a strategic partnership to jointly develop and market > the > IBM Informix family of products (original version here: > http://informix-technology ... > > My email works... but I don't check it frequently... > > --001a113f19a2d16bb5054e880247 > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --001a11c15102a3a63c054e8a8775

↪ replying to LARRY SORENSEN

Madison Pruet — Tue, 02 May 2017 13:45:29 — source: IIUG Forums & Mailing Lists

Why not test a program outside of the database? You need to verify that the problem really involves Informix. Madison Pruet Retired and Loving it On Monday, May 1, 2017 10:38 PM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: Solaris 10 IDS 11.50.FC5 Workgroup SPARC 64 bit I have two servers, one older server and one newer server. The newer server is obviously more powerful. I have an involved process that processes user records on the server. The process uses mostly Java to connect to the database to process records. The Java application runs on the database server itself. The onconfig files are identical on both servers with the exception of the SERVERNUM. I have set up projects on both servers specifying max-shm-memory, max-file-descriptor, etc. that are identical for the kernel parameters. On the old server, we could depend upon 80-100 record transactions to be processed per hour. During that time the server CPUs would remain 15%-25% idle (so almost to capacity). On the new server, we are seeing 25-35 record transactions being processed per hour and the server CPUs remain 95%-98% idle. What are some tuning parameters and things we can look at and tweak to help the new server reach its potential, or at least process as fast as the older server? Larry ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to Madison Pruet

Marcus Haarmann — Tue, 02 May 2017 13:50:35 — source: IIUG Forums & Mailing Lists

Hi, in case I understood the request correctly, the Java App runs on the same machine but not within the IDS process, correct ? In this case, IDS should not make any difference between a tcp connection to localhost or any connected host in the LAN. Marcus Haarmann Von: "Madison Pruet" <madison_pruet@yahoo.com> An: "ids" <ids@iiug.org> Gesendet: Dienstag, 2. Mai 2017 15:45:29 Betreff: Re: Question on tuning [39081] Why not test a program outside of the database? You need to verify that the problem really involves Informix. Madison Pruet Retired and Loving it On Monday, May 1, 2017 10:38 PM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: Solaris 10 IDS 11.50.FC5 Workgroup SPARC 64 bit I have two servers, one older server and one newer server. The newer server is obviously more powerful. I have an involved process that processes user records on the server. The process uses mostly Java to connect to the database to process records. The Java application runs on the database server itself. The onconfig files are identical on both servers with the exception of the SERVERNUM. I have set up projects on both servers specifying max-shm-memory, max-file-descriptor, etc. that are identical for the kernel parameters. On the old server, we could depend upon 80-100 record transactions to be processed per hour. During that time the server CPUs would remain 15%-25% idle (so almost to capacity). On the new server, we are seeing 25-35 record transactions being processed per hour and the server CPUs remain 95%-98% idle. What are some tuning parameters and things we can look at and tweak to help the new server reach its potential, or at least process as fast as the older server? Larry ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to LARRY SORENSEN

Art Kagel — Tue, 02 May 2017 14:28:51 — source: IIUG Forums & Mailing Lists

Larry: That's what I was afraid of. The T5240 is a dual socket system sporting UltraSparc T2+ processors. These T-class processors were designed in the last days of Sun Microsystems with the specific purpose of running Web apps and Java programs quickly. These highly threaded processes were choking modern multipath SMP processors like the UltraSparc M-class processors because their threads executed only a few hundred instructions before passing control to another thread. This behavior caused performance problems because long instruction queues, caches, translation pre-fetch buffers, and multi-path predictive circuits had to be flushed of unneeded instructions causing many-cycle delays every few hundred cycles. To offset that problem the Niagra processors (T1) were developed without these "unnecessary" optmizations. The T2 & T2+ processors took the design even further in the same direction. For web apps and Java programs, and even some heavily threaded C++ applications, the strategy worked. The T* processors shone out in Java based web service benchmarks. However, larger more traditional processes - like database servers - suffered extreme performance degradation on these processors as opposed to how they ran on the earlier M-class. Informix is highly threaded, however, because it was developed before the four competing thread library versions resolved themselves to a single winner in the late '90s, Informix uses its own threading library and its own thread scheduling. To the OS and to the processors each oninit process behaves as a massive single threaded application. Informix on T1 through T3 is a technology mismatch. It wasn't until the T4 & T5 processors, developed by Oracle after it acquired Sun, as they tried to get their own unthreaded database servers to perform on the hardware they now built and sold under their name, that the missing pieces of a modern optimized processor were restored to the T-class. Only the T5 is marginally successful in this way. Your old system is outperforming your new one because the old system, which is probably an M15 system, is MUCH more powerful than the new one is. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 8:04 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > The new server is a T5240. I am trying to get the model of the older > server, > but it is offline at the moment. I will send it soon. > > Larry > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Frank > Langelage > <frank@lafr.de> > Sent: Tuesday, May 2, 2017 1:11 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39070] > > On 02.05.17 05:37, LARRY SORENSEN wrote: > > Solaris 10 > > > > IDS 11.50.FC5 Workgroup > > > > SPARC 64 bit > > > > I have two servers, one older server and one newer server. The newer > server > is > > obviously more powerful. > > > Which server models you're using? > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --001a1145e9389d692e054e8b5b0d

↪ replying to Art Kagel

LARRY SORENSEN — Tue, 02 May 2017 14:35:17 — source: IIUG Forums & Mailing Lists

Thank you for the unfortunate news. Is there anything that can be done to at least offset the issue a little? Larry ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel <art.kagel@gmail.com> Sent: Tuesday, May 2, 2017 8:28 AM To: ids@iiug.org Subject: Re: Question on tuning [39084] Larry: That's what I was afraid of. The T5240 is a dual socket system sporting UltraSparc T2+ processors. These T-class processors were designed in the last days of Sun Microsystems with the specific purpose of running Web apps and Java programs quickly. These highly threaded processes were choking modern multipath SMP processors like the UltraSparc M-class processors because their threads executed only a few hundred instructions before passing control to another thread. This behavior caused performance problems because long instruction queues, caches, translation pre-fetch buffers, and multi-path predictive circuits had to be flushed of unneeded instructions causing many-cycle delays every few hundred cycles. To offset that problem the Niagra processors (T1) were developed without these "unnecessary" optmizations. The T2 & T2+ processors took the design even further in the same direction. For web apps and Java programs, and even some heavily threaded C++ applications, the strategy worked. The T* processors shone out in Java based web service benchmarks. However, larger more traditional processes - like database servers - suffered extreme performance degradation on these processors as opposed to how they ran on the earlier M-class. Informix is highly threaded, however, because it was developed before the four competing thread library versions resolved themselves to a single winner in the late '90s, Informix uses its own threading library and its own thread scheduling. To the OS and to the processors each oninit process behaves as a massive single threaded application. Informix on T1 through T3 is a technology mismatch. It wasn't until the T4 & T5 processors, developed by Oracle after it acquired Sun, as they tried to get their own unthreaded database servers to perform on the hardware they now built and sold under their name, that the missing pieces of a modern optimized processor were restored to the T-class. Only the T5 is marginally successful in this way. Your old system is outperforming your new one because the old system, which is probably an M15 system, is MUCH more powerful than the new one is. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com<http://www.askdbmgt.com>; ASK Database Management - Home<http://www.askdbmgt.com/>; www.askdbmgt.com Database Consulting and Support ... This is the site for Art S. Kagel's consultancy. The soaring majesty and beauty in the image above hides the complex ecology and ... Blog: http://informix-myview.blogspot.com/ Informix - My view<http://informix-myview.blogspot.com/>; informix-myview.blogspot.com A place to share my views and ideas which may be of interest to the Informix user Community. Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 8:04 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > The new server is a T5240. I am trying to get the model of the older > server, > but it is offline at the moment. I will send it soon. > > Larry > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Frank > Langelage > <frank@lafr.de> > Sent: Tuesday, May 2, 2017 1:11 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39070] > > On 02.05.17 05:37, LARRY SORENSEN wrote: > > Solaris 10 > > > > IDS 11.50.FC5 Workgroup > > > > SPARC 64 bit > > > > I have two servers, one older server and one newer server. The newer > server > is > > obviously more powerful. > > > Which server models you're using? > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --001a1145e9389d692e054e8b5b0d ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.

↪ replying to LARRY SORENSEN

Art Kagel — Tue, 02 May 2017 14:39:19 — source: IIUG Forums & Mailing Lists

The V490 Sunfire has: Dual Core Sun UltraSPARC IV 1.35 GHz A much better processor for running databases. See my previous post. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 8:22 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ > > The older server is a V490, not sure the processors > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel > <art.kagel@gmail.com> > Sent: Tuesday, May 2, 2017 4:47 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39073] > > You THINK the new system is more powerful, but with Sun systems it is a > definite possibility that it is not. > > What are the models and processor types for the two systems? > > Art > > Art S. Kagel, President and Principal Consultant > ASK Database Management > www.askdbmgt.com<http://www.askdbmgt.com>; > > ASK Database Management - Home<http://www.askdbmgt.com/>; > www.askdbmgt.com > Database Consulting and Support ... This is the site for Art S. Kagel's > consultancy. The soaring majesty and beauty in the image above hides the > complex ecology and ... > > Blog: http://informix-myview.blogspot.com/ > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and do not reflect on the IIUG, nor any other organization with which I am > associated either explicitly, implicitly, or by inference. Neither do > those opinions reflect those of other individuals affiliated with any > entity with which I am affiliated nor those of the entities themselves. > > On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> > wrote: > > > Solaris 10 > > > > IDS 11.50.FC5 Workgroup > > > > SPARC 64 bit > > > > I have two servers, one older server and one newer server. The newer > > server is > > obviously more powerful. > > > > I have an involved process that processes user records on the server. The > > process uses mostly Java to connect to the database to process records. > The > > Java application runs on the database server itself. > > > > The onconfig files are identical on both servers with the exception of > the > > SERVERNUM. > > > > I have set up projects on both servers specifying max-shm-memory, > > max-file-descriptor, etc. that are identical for the kernel parameters. > > > > On the old server, we could depend upon 80-100 record transactions to be > > processed per hour. During that time the server CPUs would remain 15%-25% > > idle > > (so almost to capacity). > > > > On the new server, we are seeing 25-35 record transactions being > processed > > per > > hour and the server CPUs remain 95%-98% idle. > > > > What are some tuning parameters and things we can look at and tweak to > help > > the new server reach its potential, or at least process as fast as the > > older > > server? > > > > Larry > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > --f403043a7ae8045e86054e88417d > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --f403045ff6820f39be054e8b811b

↪ replying to LARRY SORENSEN

Art Kagel — Tue, 02 May 2017 14:53:25 — source: IIUG Forums & Mailing Lists

SSD drives and LOTS of buffer cache are the main things that can offset the processing deficit. I would also try overclocking the CPU VPs. Run 4 or 5 per core to get better overall throughput, but since the server is competing with the Java apps for cycles, that may not help. Maybe move the applications to another box to free up cycles for the server to use. One thing that's probably exacerbating the problem is that your Java apps are running better on the new platform so they are putting more stress on the database server and leaving fewer cycles free for the oninits to use. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 10:35 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Thank you for the unfortunate news. Is there anything that can be done to > at > least offset the issue a little? > > Larry > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel > <art.kagel@gmail.com> > Sent: Tuesday, May 2, 2017 8:28 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39084] > > Larry: > > That's what I was afraid of. The T5240 is a dual socket system sporting > UltraSparc T2+ processors. These T-class processors were designed in the > last days of Sun Microsystems with the specific purpose of running Web apps > and Java programs quickly. These highly threaded processes were choking > modern multipath SMP processors like the UltraSparc M-class processors > because their threads executed only a few hundred instructions before > passing control to another thread. This behavior caused performance > problems because long instruction queues, caches, translation pre-fetch > buffers, and multi-path predictive circuits had to be flushed of unneeded > instructions causing many-cycle delays every few hundred cycles. > > To offset that problem the Niagra processors (T1) were developed without > these "unnecessary" optmizations. The T2 & T2+ processors took the design > even further in the same direction. For web apps and Java programs, and > even some heavily threaded C++ applications, the strategy worked. The T* > processors shone out in Java based web service benchmarks. However, larger > more traditional processes - like database servers - suffered extreme > performance degradation on these processors as opposed to how they ran on > the earlier M-class. Informix is highly threaded, however, because it was > developed before the four competing thread library versions resolved > themselves to a single winner in the late '90s, Informix uses its own > threading library and its own thread scheduling. To the OS and to the > processors each oninit process behaves as a massive single threaded > application. Informix on T1 through T3 is a technology mismatch. > > It wasn't until the T4 & T5 processors, developed by Oracle after it > acquired Sun, as they tried to get their own unthreaded database servers to > perform on the hardware they now built and sold under their name, that the > missing pieces of a modern optimized processor were restored to the > T-class. Only the T5 is marginally successful in this way. > > Your old system is outperforming your new one because the old system, which > is probably an M15 system, is MUCH more powerful than the new one is. > > Art > > Art S. Kagel, President and Principal Consultant > ASK Database Management > www.askdbmgt.com<http://www.askdbmgt.com>; > ASK Database Management - Home<http://www.askdbmgt.com/>; > www.askdbmgt.com > Database Consulting and Support ... This is the site for Art S. Kagel's > consultancy. The soaring majesty and beauty in the image above hides the > complex ecology and ... > > Blog: http://informix-myview.blogspot.com/ > Informix - My view<http://informix-myview.blogspot.com/>; > informix-myview.blogspot.com > A place to share my views and ideas which may be of interest to the > Informix > user Community. > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and do not reflect on the IIUG, nor any other organization with which I am > associated either explicitly, implicitly, or by inference. Neither do > those opinions reflect those of other individuals affiliated with any > entity with which I am affiliated nor those of the entities themselves. > > On Tue, May 2, 2017 at 8:04 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > wrote: > > > The new server is a T5240. I am trying to get the model of the older > > server, > > but it is offline at the moment. I will send it soon. > > > > Larry > > > > ________________________________ > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Frank > > Langelage > > <frank@lafr.de> > > Sent: Tuesday, May 2, 2017 1:11 AM > > To: ids@iiug.org > > Subject: Re: Question on tuning [39070] > > > > On 02.05.17 05:37, LARRY SORENSEN wrote: > > > Solaris 10 > > > > > > IDS 11.50.FC5 Workgroup > > > > > > SPARC 64 bit > > > > > > I have two servers, one older server and one newer server. The newer > > server > > is > > > obviously more powerful. > > > > > Which server models you're using? > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > --001a1145e9389d692e054e8b5b0d > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --94eb2c13edd6d2ce58054e8bb087

↪ replying to Art Kagel

Fernando Nunes — Tue, 02 May 2017 15:06:45 — source: IIUG Forums & Mailing Lists

Does your explanation match the 95% idle the OP mentions? On Tue, May 2, 2017 at 3:39 PM, Art Kagel <art.kagel@gmail.com> wrote: > The V490 Sunfire has: > > Dual Core Sun UltraSPARC IV 1.35 GHz > > A much better processor for running databases. See my previous post. > > Art > > Art S. Kagel, President and Principal Consultant > ASK Database Management > www.askdbmgt.com > > Blog: http://informix-myview.blogspot.com/ > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and do not reflect on the IIUG, nor any other organization with which I am > associated either explicitly, implicitly, or by inference. Neither do > those opinions reflect those of other individuals affiliated with any > entity with which I am affiliated nor those of the entities themselves. > > On Tue, May 2, 2017 at 8:22 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > wrote: > > > The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ > > > > The older server is a V490, not sure the processors > > > > ________________________________ > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel > > <art.kagel@gmail.com> > > Sent: Tuesday, May 2, 2017 4:47 AM > > To: ids@iiug.org > > Subject: Re: Question on tuning [39073] > > > > You THINK the new system is more powerful, but with Sun systems it is a > > definite possibility that it is not. > > > > What are the models and processor types for the two systems? > > > > Art > > > > Art S. Kagel, President and Principal Consultant > > ASK Database Management > > www.askdbmgt.com<http://www.askdbmgt.com>; > > > > ASK Database Management - Home<http://www.askdbmgt.com/>; > > www.askdbmgt.com > > Database Consulting and Support ... This is the site for Art S. Kagel's > > consultancy. The soaring majesty and beauty in the image above hides the > > complex ecology and ... > > > > Blog: http://informix-myview.blogspot.com/ > > > > Disclaimer: Please keep in mind that my own opinions are my own opinions > > and do not reflect on the IIUG, nor any other organization with which I > am > > associated either explicitly, implicitly, or by inference. Neither do > > those opinions reflect those of other individuals affiliated with any > > entity with which I am affiliated nor those of the entities themselves. > > > > On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> > > wrote: > > > > > Solaris 10 > > > > > > IDS 11.50.FC5 Workgroup > > > > > > SPARC 64 bit > > > > > > I have two servers, one older server and one newer server. The newer > > > server is > > > obviously more powerful. > > > > > > I have an involved process that processes user records on the server. > The > > > process uses mostly Java to connect to the database to process records. > > The > > > Java application runs on the database server itself. > > > > > > The onconfig files are identical on both servers with the exception of > > the > > > SERVERNUM. > > > > > > I have set up projects on both servers specifying max-shm-memory, > > > max-file-descriptor, etc. that are identical for the kernel parameters. > > > > > > On the old server, we could depend upon 80-100 record transactions to > be > > > processed per hour. During that time the server CPUs would remain > 15%-25% > > > idle > > > (so almost to capacity). > > > > > > On the new server, we are seeing 25-35 record transactions being > > processed > > > per > > > hour and the server CPUs remain 95%-98% idle. > > > > > > What are some tuning parameters and things we can look at and tweak to > > help > > > the new server reach its potential, or at least process as fast as the > > > older > > > server? > > > > > > Larry > > > > > > > > > ************************************************************ > > > ******************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > --f403043a7ae8045e86054e88417d > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > --f403045ff6820f39be054e8b811b > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently... --001a113f19a20aeb9a054e8bdf15

↪ replying to Fernando Nunes

Art Kagel — Tue, 02 May 2017 15:14:04 — source: IIUG Forums & Mailing Lists

Yes, I believe that it does. Informix cannot take good advantage of the processors. The behavior is consistent with what I have seen at other T1, T2, & T3 sites. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 11:06 AM, Fernando Nunes <domusonline@gmail.com> wrote: > Does your explanation match the 95% idle the OP mentions? > > On Tue, May 2, 2017 at 3:39 PM, Art Kagel <art.kagel@gmail.com> wrote: > > > The V490 Sunfire has: > > > > Dual Core Sun UltraSPARC IV 1.35 GHz > > > > A much better processor for running databases. See my previous post. > > > > Art > > > > Art S. Kagel, President and Principal Consultant > > ASK Database Management > > www.askdbmgt.com > > > > Blog: http://informix-myview.blogspot.com/ > > > > Disclaimer: Please keep in mind that my own opinions are my own opinions > > and do not reflect on the IIUG, nor any other organization with which I > am > > associated either explicitly, implicitly, or by inference. Neither do > > those opinions reflect those of other individuals affiliated with any > > entity with which I am affiliated nor those of the entities themselves. > > > > On Tue, May 2, 2017 at 8:22 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > > wrote: > > > > > The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ > > > > > > The older server is a V490, not sure the processors > > > > > > ________________________________ > > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art > Kagel > > > <art.kagel@gmail.com> > > > Sent: Tuesday, May 2, 2017 4:47 AM > > > To: ids@iiug.org > > > Subject: Re: Question on tuning [39073] > > > > > > You THINK the new system is more powerful, but with Sun systems it is a > > > definite possibility that it is not. > > > > > > What are the models and processor types for the two systems? > > > > > > Art > > > > > > Art S. Kagel, President and Principal Consultant > > > ASK Database Management > > > www.askdbmgt.com<http://www.askdbmgt.com>; > > > > > > ASK Database Management - Home<http://www.askdbmgt.com/>; > > > www.askdbmgt.com > > > Database Consulting and Support ... This is the site for Art S. Kagel's > > > consultancy. The soaring majesty and beauty in the image above hides > the > > > complex ecology and ... > > > > > > Blog: http://informix-myview.blogspot.com/ > > > > > > Disclaimer: Please keep in mind that my own opinions are my own > opinions > > > and do not reflect on the IIUG, nor any other organization with which I > > am > > > associated either explicitly, implicitly, or by inference. Neither do > > > those opinions reflect those of other individuals affiliated with any > > > entity with which I am affiliated nor those of the entities themselves. > > > > > > On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> > > > wrote: > > > > > > > Solaris 10 > > > > > > > > IDS 11.50.FC5 Workgroup > > > > > > > > SPARC 64 bit > > > > > > > > I have two servers, one older server and one newer server. The newer > > > > server is > > > > obviously more powerful. > > > > > > > > I have an involved process that processes user records on the server. > > The > > > > process uses mostly Java to connect to the database to process > records. > > > The > > > > Java application runs on the database server itself. > > > > > > > > The onconfig files are identical on both servers with the exception > of > > > the > > > > SERVERNUM. > > > > > > > > I have set up projects on both servers specifying max-shm-memory, > > > > max-file-descriptor, etc. that are identical for the kernel > parameters. > > > > > > > > On the old server, we could depend upon 80-100 record transactions to > > be > > > > processed per hour. During that time the server CPUs would remain > > 15%-25% > > > > idle > > > > (so almost to capacity). > > > > > > > > On the new server, we are seeing 25-35 record transactions being > > > processed > > > > per > > > > hour and the server CPUs remain 95%-98% idle. > > > > > > > > What are some tuning parameters and things we can look at and tweak > to > > > help > > > > the new server reach its potential, or at least process as fast as > the > > > > older > > > > server? > > > > > > > > Larry > > > > > > > > > > > > ************************************************************ > > > > ******************* > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > > > > > --f403043a7ae8045e86054e88417d > > > > > > > > > ************************************************************ > > > ******************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > ************************************************************ > > > ******************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > --f403045ff6820f39be054e8b811b > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently... > > --001a113f19a20aeb9a054e8bdf15 > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --f403045d172ae8b7df054e8bf6f5

↪ replying to Art Kagel

LARRY SORENSEN — Tue, 02 May 2017 15:18:17 — source: IIUG Forums & Mailing Lists

The thing that is the most irritating is that the server looks like it is sitting there idle. ________________________________ From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel <art.kagel@gmail.com> Sent: Tuesday, May 2, 2017 8:53 AM To: ids@iiug.org Subject: Re: Question on tuning [39089] SSD drives and LOTS of buffer cache are the main things that can offset the processing deficit. I would also try overclocking the CPU VPs. Run 4 or 5 per core to get better overall throughput, but since the server is competing with the Java apps for cycles, that may not help. Maybe move the applications to another box to free up cycles for the server to use. One thing that's probably exacerbating the problem is that your Java apps are running better on the new platform so they are putting more stress on the database server and leaving fewer cycles free for the oninits to use. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com<http://www.askdbmgt.com>; ASK Database Management - Home<http://www.askdbmgt.com/>; www.askdbmgt.com Database Consulting and Support ... This is the site for Art S. Kagel's consultancy. The soaring majesty and beauty in the image above hides the complex ecology and ... Blog: http://informix-myview.blogspot.com/ Informix - My view<http://informix-myview.blogspot.com/>; informix-myview.blogspot.com A place to share my views and ideas which may be of interest to the Informix user Community. Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 10:35 AM, LARRY SORENSEN <LSORENSEN25@msn.com> wrote: > Thank you for the unfortunate news. Is there anything that can be done to > at > least offset the issue a little? > > Larry > > ________________________________ > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art Kagel > <art.kagel@gmail.com> > Sent: Tuesday, May 2, 2017 8:28 AM > To: ids@iiug.org > Subject: Re: Question on tuning [39084] > > Larry: > > That's what I was afraid of. The T5240 is a dual socket system sporting > UltraSparc T2+ processors. These T-class processors were designed in the > last days of Sun Microsystems with the specific purpose of running Web apps > and Java programs quickly. These highly threaded processes were choking > modern multipath SMP processors like the UltraSparc M-class processors > because their threads executed only a few hundred instructions before > passing control to another thread. This behavior caused performance > problems because long instruction queues, caches, translation pre-fetch > buffers, and multi-path predictive circuits had to be flushed of unneeded > instructions causing many-cycle delays every few hundred cycles. > > To offset that problem the Niagra processors (T1) were developed without > these "unnecessary" optmizations. The T2 & T2+ processors took the design > even further in the same direction. For web apps and Java programs, and > even some heavily threaded C++ applications, the strategy worked. The T* > processors shone out in Java based web service benchmarks. However, larger > more traditional processes - like database servers - suffered extreme > performance degradation on these processors as opposed to how they ran on > the earlier M-class. Informix is highly threaded, however, because it was > developed before the four competing thread library versions resolved > themselves to a single winner in the late '90s, Informix uses its own > threading library and its own thread scheduling. To the OS and to the > processors each oninit process behaves as a massive single threaded > application. Informix on T1 through T3 is a technology mismatch. > > It wasn't until the T4 & T5 processors, developed by Oracle after it > acquired Sun, as they tried to get their own unthreaded database servers to > perform on the hardware they now built and sold under their name, that the > missing pieces of a modern optimized processor were restored to the > T-class. Only the T5 is marginally successful in this way. > > Your old system is outperforming your new one because the old system, which > is probably an M15 system, is MUCH more powerful than the new one is. > > Art > > Art S. Kagel, President and Principal Consultant > ASK Database Management > www.askdbmgt.com<http://www.askdbmgt.com>; > ASK Database Management - Home<http://www.askdbmgt.com/>; > www.askdbmgt.com<http://www.askdbmgt.com>; > Database Consulting and Support ... This is the site for Art S. Kagel's > consultancy. The soaring majesty and beauty in the image above hides the > complex ecology and ... > > Blog: http://informix-myview.blogspot.com/ > Informix - My view<http://informix-myview.blogspot.com/>; > informix-myview.blogspot.com > A place to share my views and ideas which may be of interest to the > Informix > user Community. > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and do not reflect on the IIUG, nor any other organization with which I am > associated either explicitly, implicitly, or by inference. Neither do > those opinions reflect those of other individuals affiliated with any > entity with which I am affiliated nor those of the entities themselves. > > On Tue, May 2, 2017 at 8:04 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > wrote: > > > The new server is a T5240. I am trying to get the model of the older > > server, > > but it is offline at the moment. I will send it soon. > > > > Larry > > > > ________________________________ > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Frank > > Langelage > > <frank@lafr.de> > > Sent: Tuesday, May 2, 2017 1:11 AM > > To: ids@iiug.org > > Subject: Re: Question on tuning [39070] > > > > On 02.05.17 05:37, LARRY SORENSEN wrote: > > > Solaris 10 > > > > > > IDS 11.50.FC5 Workgroup > > > > > > SPARC 64 bit > > > > > > I have two servers, one older server and one newer server. The newer > > server > > is > > > obviously more powerful. > > > > > Which server models you're using? > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > --001a1145e9389d692e054e8b5b0d > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the disc

↪ replying to Art Kagel

Marcus Haarmann — Tue, 02 May 2017 15:30:56 — source: IIUG Forums & Mailing Lists

If that is the case and you are suffering from a machine architecture problem, try setting NUMCPUVPS (or newer: VPCLASS cpu) to a number which is like double of your available processor cores. That is what Art meant by overclocking the CPU. Also, you could try CPU binding (with VPCLASS cpu aff=x-y but I don't know how this behaves if you have a num= which is bigger than the real processors) to not interfere with the JAVA application. You could try to bind the java processes to a set of cores and the informix ones to a different set to prevent interfering context switches. (see pbind/psrset man page, I would assume this is existing (was some years ago)) Marcus Haarmann Von: "Art Kagel" <art.kagel@gmail.com> An: "ids" <ids@iiug.org> Gesendet: Dienstag, 2. Mai 2017 17:14:14 Betreff: Re: Question on tuning [39095] Yes, I believe that it does. Informix cannot take good advantage of the processors. The behavior is consistent with what I have seen at other T1, T2, & T3 sites. Art Art S. Kagel, President and Principal Consultant ASK Database Management www.askdbmgt.com Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Tue, May 2, 2017 at 11:06 AM, Fernando Nunes <domusonline@gmail.com> wrote: > Does your explanation match the 95% idle the OP mentions? > > On Tue, May 2, 2017 at 3:39 PM, Art Kagel <art.kagel@gmail.com> wrote: > > > The V490 Sunfire has: > > > > Dual Core Sun UltraSPARC IV 1.35 GHz > > > > A much better processor for running databases. See my previous post. > > > > Art > > > > Art S. Kagel, President and Principal Consultant > > ASK Database Management > > www.askdbmgt.com > > > > Blog: http://informix-myview.blogspot.com/ > > > > Disclaimer: Please keep in mind that my own opinions are my own opinions > > and do not reflect on the IIUG, nor any other organization with which I > am > > associated either explicitly, implicitly, or by inference. Neither do > > those opinions reflect those of other individuals affiliated with any > > entity with which I am affiliated nor those of the entities themselves. > > > > On Tue, May 2, 2017 at 8:22 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > > wrote: > > > > > The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ > > > > > > The older server is a V490, not sure the processors > > > > > > ________________________________ > > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art > Kagel > > > <art.kagel@gmail.com> > > > Sent: Tuesday, May 2, 2017 4:47 AM > > > To: ids@iiug.org > > > Subject: Re: Question on tuning [39073] > > > > > > You THINK the new system is more powerful, but with Sun systems it is a > > > definite possibility that it is not. > > > > > > What are the models and processor types for the two systems? > > > > > > Art > > > > > > Art S. Kagel, President and Principal Consultant > > > ASK Database Management > > > www.askdbmgt.com<http://www.askdbmgt.com>; > > > > > > ASK Database Management - Home<http://www.askdbmgt.com/>; > > > www.askdbmgt.com > > > Database Consulting and Support ... This is the site for Art S. Kagel's > > > consultancy. The soaring majesty and beauty in the image above hides > the > > > complex ecology and ... > > > > > > Blog: http://informix-myview.blogspot.com/ > > > > > > Disclaimer: Please keep in mind that my own opinions are my own > opinions > > > and do not reflect on the IIUG, nor any other organization with which I > > am > > > associated either explicitly, implicitly, or by inference. Neither do > > > those opinions reflect those of other individuals affiliated with any > > > entity with which I am affiliated nor those of the entities themselves. > > > > > > On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> > > > wrote: > > > > > > > Solaris 10 > > > > > > > > IDS 11.50.FC5 Workgroup > > > > > > > > SPARC 64 bit > > > > > > > > I have two servers, one older server and one newer server. The newer > > > > server is > > > > obviously more powerful. > > > > > > > > I have an involved process that processes user records on the server. > > The > > > > process uses mostly Java to connect to the database to process > records. > > > The > > > > Java application runs on the database server itself. > > > > > > > > The onconfig files are identical on both servers with the exception > of > > > the > > > > SERVERNUM. > > > > > > > > I have set up projects on both servers specifying max-shm-memory, > > > > max-file-descriptor, etc. that are identical for the kernel > parameters. > > > > > > > > On the old server, we could depend upon 80-100 record transactions to > > be > > > > processed per hour. During that time the server CPUs would remain > > 15%-25% > > > > idle > > > > (so almost to capacity). > > > > > > > > On the new server, we are seeing 25-35 record transactions being > > > processed > > > > per > > > > hour and the server CPUs remain 95%-98% idle. > > > > > > > > What are some tuning parameters and things we can look at and tweak > to > > > help > > > > the new server reach its potential, or at least process as fast as > the > > > > older > > > > server? > > > > > > > > Larry > > > > > > > > > > > > ************************************************************ > > > > ******************* > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > > > > > --f403043a7ae8045e86054e88417d > > > > > > > > > ************************************************************ > > > ******************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > ************************************************************ > > > ******************* > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > --f403045ff6820f39be054e8b811b > > > > > > ************************************************************ > > ******************* > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > -- > Fernando Nunes > Portugal > > http://informix-technology.blogspot.com > My email works... but I don't check it frequently... > > --001a113f19a20aeb9a054e8bdf15 > > > ************************************************************ > ******************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --f403045d172ae8b7df054e8bf6f5 ******************************************************************************* Forum Note: Use "Reply" to post

↪ replying to Marcus Haarmann

david@smooth1.co.uk — Tue, 02 May 2017 16:16:36 — source: IIUG Forums & Mailing Lists

Hi, 11.7 also supports aff=<cpu number>,<cpu number>,... I wonder what would happen if you used say aff=1,2,3,1,2,3 Possible it would bind each cpu vp in turn to the given cpn number. I am not sure if there is a check for duplicate cpu numbers... Regards, David, > On 02 May 2017 at 16:30 Marcus Haarmann <marcus.haarmann@midoco.de> wrote: > > > If that is the case and you are suffering from a machine architecture problem, > try setting NUMCPUVPS (or newer: VPCLASS cpu) > to a number which is like double of your available processor cores. > That is what Art meant by overclocking the CPU. > > Also, you could try CPU binding (with VPCLASS cpu aff=x-y but I don't know how > this behaves if you have a num= which is bigger > than the real processors) to not interfere with the JAVA application. > You could try to bind the java processes to a set of cores and the informix > ones to a different set > to prevent interfering context switches. > (see pbind/psrset man page, I would assume this is existing (was some years > ago)) > > Marcus Haarmann > > Von: "Art Kagel" <art.kagel@gmail.com> > An: "ids" <ids@iiug.org> > Gesendet: Dienstag, 2. Mai 2017 17:14:14 > Betreff: Re: Question on tuning [39095] > > Yes, I believe that it does. Informix cannot take good advantage of the > processors. The behavior is consistent with what I have seen at other T1, > T2, & T3 sites. > > Art > > Art S. Kagel, President and Principal Consultant > ASK Database Management > www.askdbmgt.com > > Blog: http://informix-myview.blogspot.com/ > > Disclaimer: Please keep in mind that my own opinions are my own opinions > and do not reflect on the IIUG, nor any other organization with which I am > associated either explicitly, implicitly, or by inference. Neither do > those opinions reflect those of other individuals affiliated with any > entity with which I am affiliated nor those of the entities themselves. > > On Tue, May 2, 2017 at 11:06 AM, Fernando Nunes <domusonline@gmail.com> > wrote: > > > Does your explanation match the 95% idle the OP mentions? > > > > On Tue, May 2, 2017 at 3:39 PM, Art Kagel <art.kagel@gmail.com> wrote: > > > > > The V490 Sunfire has: > > > > > > Dual Core Sun UltraSPARC IV 1.35 GHz > > > > > > A much better processor for running databases. See my previous post. > > > > > > Art > > > > > > Art S. Kagel, President and Principal Consultant > > > ASK Database Management > > > www.askdbmgt.com > > > > > > Blog: http://informix-myview.blogspot.com/ > > > > > > Disclaimer: Please keep in mind that my own opinions are my own opinions > > > and do not reflect on the IIUG, nor any other organization with which I > > am > > > associated either explicitly, implicitly, or by inference. Neither do > > > those opinions reflect those of other individuals affiliated with any > > > entity with which I am affiliated nor those of the entities themselves. > > > > > > On Tue, May 2, 2017 at 8:22 AM, LARRY SORENSEN <LSORENSEN25@msn.com> > > > wrote: > > > > > > > The new server has T5240 - 1415MHz SUNW,UltraSPARC-T2+ > > > > > > > > The older server is a V490, not sure the processors > > > > > > > > ________________________________ > > > > From: ids-bounces@iiug.org <ids-bounces@iiug.org> on behalf of Art > > Kagel > > > > <art.kagel@gmail.com> > > > > Sent: Tuesday, May 2, 2017 4:47 AM > > > > To: ids@iiug.org > > > > Subject: Re: Question on tuning [39073] > > > > > > > > You THINK the new system is more powerful, but with Sun systems it is a > > > > definite possibility that it is not. > > > > > > > > What are the models and processor types for the two systems? > > > > > > > > Art > > > > > > > > Art S. Kagel, President and Principal Consultant > > > > ASK Database Management > > > > www.askdbmgt.com<http://www.askdbmgt.com>; > > > > > > > > ASK Database Management - Home<http://www.askdbmgt.com/>; > > > > www.askdbmgt.com > > > > Database Consulting and Support ... This is the site for Art S. Kagel's > > > > consultancy. The soaring majesty and beauty in the image above hides > > the > > > > complex ecology and ... > > > > > > > > Blog: http://informix-myview.blogspot.com/ > > > > > > > > Disclaimer: Please keep in mind that my own opinions are my own > > opinions > > > > and do not reflect on the IIUG, nor any other organization with which I > > > am > > > > associated either explicitly, implicitly, or by inference. Neither do > > > > those opinions reflect those of other individuals affiliated with any > > > > entity with which I am affiliated nor those of the entities themselves. > > > > > > > > On Mon, May 1, 2017 at 11:37 PM, LARRY SORENSEN <LSORENSEN25@msn.com> > > > > wrote: > > > > > > > > > Solaris 10 > > > > > > > > > > IDS 11.50.FC5 Workgroup > > > > > > > > > > SPARC 64 bit > > > > > > > > > > I have two servers, one older server and one newer server. The newer > > > > > server is > > > > > obviously more powerful. > > > > > > > > > > I have an involved process that processes user records on the server. > > > The > > > > > process uses mostly Java to connect to the database to process > > records. > > > > The > > > > > Java application runs on the database server itself. > > > > > > > > > > The onconfig files are identical on both servers with the exception > > of > > > > the > > > > > SERVERNUM. > > > > > > > > > > I have set up projects on both servers specifying max-shm-memory, > > > > > max-file-descriptor, etc. that are identical for the kernel > > parameters. > > > > > > > > > > On the old server, we could depend upon 80-100 record transactions to > > > be > > > > > processed per hour. During that time the server CPUs would remain > > > 15%-25% > > > > > idle > > > > > (so almost to capacity). > > > > > > > > > > On the new server, we are seeing 25-35 record transactions being > > > > processed > > > > > per > > > > > hour and the server CPUs remain 95%-98% idle. > > > > > > > > > > What are some tuning parameters and things we can look at and tweak > > to > > > > help > > > > > the new server reach its potential, or at least process as fast as > > the > > > > > older > > > > > server? > > > > > > > > > > Larry > > > > > > > > > > > > > > > ************************************************************ > > > > > ******************* > > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > > > > > > > > > --f403043a7ae8045e86054e88417d > > > > > > > > > > > > ************************************************************ > > > > ******************* > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > > > ************************************************************ > > > > ******************* > > > > Forum Note: Use "Reply" to post a response in the discussion forum. > > > > > > > > > > > > > > --f403045ff6820f39be054e8b811b > > > >

Related threads

Glossary

onconfig