Chatr on HPUX and Informix
After upgrading from IDS 7.24.UC5 to 7.30.UC7 on HP-UX 10.20 (Baan), the poster saw steadily degrading performance and growing shared memory, and got "shmat: shared memory base address illegal / can't create virtual segment" errors when raising SHMVIRTSIZE past ~1GB and trying chatr -M. Replies explained HP-UX 32-bit 1GB quadrants limit a single segment (suggesting SINGLESEG_OFF=1 to split resident/virtual segments, and that chatr is only needed above ~1.75GB), pointed at a buffer-management bug fixed in 7.30.UC7XK, warned about poor performance once a fourth segment is added, and suspected a memory leak. No confirmation of a fix from the poster is recorded.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Howdy Folks,
We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We
are running on HPUX 10.20, Application is Baan.
Since we have performed this upgrade, we are expierencing the database
gradually becoming slower and slower. Load averages on the top display are
getting higher over the period of the day (around 5) which is pretty high
for our system. The database seems to require more than 900 meg of memory
as it seems to dynamically allocate more segments. B4 the upgrade we were
using below the 900meg mark. I have been advised to alter the Informix
binaries with "chatr -M" I have tested this on a test system but I am
getting error messages on startup of DB. I am wondering if I have missed
something....
Has anyone already set up / using this on a HPUX system, using more than 1
gb of shared memory (7.30 Uc7).
Would you please advise the process you used / followed?? What issues did
you come across implementing chatr ??
TIA
Regards
Suze.
Please refer to the attached copy of my online.log
Fri Nov 5 14:02:19 1999
14:02:19 Informix Dynamic Server Stopped.
16:00:52 Segment locked: addr=0x40000000, size=956596224
16:00:52 The number of configured CPU poll threads exceeds 'NUMCPUVPS'. NETTYPE 'baantcp' poll threads started on NET VPs.
Sun Nov 7 16:00:52 1999
16:00:52 Event alarms enabled. ALARMPROG =
'/opt/informix/etc/ifx_alarm.sh'
16:00:58 DR: DRAUTO is 0 (Off)
16:00:58 Requested shared memory segment size rounded from 588KB to 592KB
16:00:59 Informix Dynamic Server Version 7.30.UC7 Software Serial NumberAAC#J578294
16:01:00 Informix Dynamic Server Initialized -- Shared Memory Initialized.
16:01:00 Physical Recovery Started.
16:01:00 Physical Recovery Complete: 0 Pages Restored.
16:01:00 Logical Recovery Started.
16:01:03 Logical Recovery Complete. 0 Committed, 0 Rolled Back, 0 Open, 0 Bad Locks
16:01:03 Onconfig parameter LTAPEDEV modified from /dev/rmt/0m to
/dev/null.
16:01:03 Dataskip is now OFF for all dbspaces
16:01:03 HPUX Version B.10.20 -> Using flag/select style KAIO
16:01:04 On-Line Mode
16:01:04 Checkpoint Completed: duration was 1 seconds.
16:01:54 Informix Dynamic Server Stopped.
I altered the onconfig for SHMVERT to be approx 1200 meg and started the
intance. Error messages below.
16:05:00 shmat: [EINVAL][22]: shared memory base address illegal
16:05:00 using 0x60000000, needs 0xc1483000
16:05:00 Could not create single shared memory segment with resident andnon-resident par
titions. Proceeding to create 2 shared memory segments instead.
16:05:03 Segment locked: addr=0x40000000, size=117719040
16:05:03 shmat: [EINVAL][22]: shared memory base address illegal
16:05:03 using 0x67044000, needs 0x80000000
16:05:03 mt_shm_init: can't create virtual segment
uxfgrn4:/opt/informix>
Susan Elliott
______________________
Host Team Leader
Information Services Group
Concrete Group
* (09) 520 8608
* susane@fclcis.co.nz
UPDATE STATISTICS?
Is your online.log available?
Neil
Susan Elliott (ISG) wrote in message <80304n$e3c$1@news.xmission.com>...
>
>
>Howdy Folks,
>
>We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We
>are running on HPUX 10.20, Application is Baan.
>
>Since we have performed this upgrade, we are expierencing the database
>gradually becoming slower and slower. Load averages on the top display are
>getting higher over the period of the day (around 5) which is pretty high
>for our system. The database seems to require more than 900 meg of memory
>as it seems to dynamically allocate more segments. B4 the upgrade we were
>using below the 900meg mark. I have been advised to alter the Informix
>binaries with "chatr -M" I have tested this on a test system but I am
>getting error messages on startup of DB. I am wondering if I have missed
>something....
>
>Has anyone already set up / using this on a HPUX system, using more than 1
>gb of shared memory (7.30 Uc7).
>
>Would you please advise the process you used / followed?? What issues did
>you come across implementing chatr ??
>
>
>TIA
>
>Regards
>Suze.
>
>
>Please refer to the attached copy of my online.log
>
>
>Fri Nov 5 14:02:19 1999
>
>14:02:19 Informix Dynamic Server Stopped.
>16:00:52 Segment locked: addr=0x40000000, size=956596224
>16:00:52 The number of configured CPU poll threads exceeds 'NUMCPUVPS'.> NETTYPE 'baantcp' poll threads started on NET VPs.
>
>
>Sun Nov 7 16:00:52 1999
>
>16:00:52 Event alarms enabled. ALARMPROG =
>'/opt/informix/etc/ifx_alarm.sh'
>16:00:58 DR: DRAUTO is 0 (Off)
>16:00:58 Requested shared memory segment size rounded from 588KB to 592KB
>16:00:59 Informix Dynamic Server Version 7.30.UC7 Software Serial Number>AAC#J578294
>16:01:00 Informix Dynamic Server Initialized -- Shared Memory Initialized.
>16:01:00 Physical Recovery Started.
>16:01:00 Physical Recovery Complete: 0 Pages Restored.
>16:01:00 Logical Recovery Started.
>16:01:03 Logical Recovery Complete.> 0 Committed, 0 Rolled Back, 0 Open, 0 Bad Locks
>
>16:01:03 Onconfig parameter LTAPEDEV modified from /dev/rmt/0m to
>/dev/null.
>16:01:03 Dataskip is now OFF for all dbspaces
>16:01:03 HPUX Version B.10.20 -> Using flag/select style KAIO
>16:01:04 On-Line Mode
>16:01:04 Checkpoint Completed: duration was 1 seconds.
>16:01:54 Informix Dynamic Server Stopped.>
>I altered the onconfig for SHMVERT to be approx 1200 meg and started the
>intance. Error messages below.
>
>16:05:00 shmat: [EINVAL][22]: shared memory base address illegal
>
>16:05:00 using 0x60000000, needs 0xc1483000
>
>16:05:00 Could not create single shared memory segment with resident and>non-resident par
>
>titions. Proceeding to create 2 shared memory segments instead.
>16:05:03 Segment locked: addr=0x40000000, size=117719040
>16:05:03 shmat: [EINVAL][22]: shared memory base address illegal
>
>16:05:03 using 0x67044000, needs 0x80000000
>
>16:05:03 mt_shm_init: can't create virtual segment>
>uxfgrn4:/opt/informix>
>
>
>Susan Elliott
>______________________
>Host Team Leader
>Information Services Group
>Concrete Group
>* (09) 520 8608
>* susane@fclcis.co.nz
>
>
>
There's a nasty buffer management bug that afflicts releases prior to 7.30.UC7XK. The system works real hard, but nothing gets done very fast. The only real solution is to upgrade to a later release that doesn't have the problem. It drove a lot of SAP admins nuts for a few months until a solution became available. Also, in a 32 bit HP environment, no single segment can be bigger than 1GB, so don't bother trying. I doubt chatr or shmvirtsize have anything to do with your performance problems. Greg "Susan Elliott (ISG)" wrote: > Howdy Folks, > > We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We > are running on HPUX 10.20, Application is Baan. > > Since we have performed this upgrade, we are expierencing the database > gradually becoming slower and slower. Load averages on the top display are > getting higher over the period of the day (around 5) which is pretty high
Hi Susan ,
Can you tell me what version of Baan IV it is ? (i.e b3 or c3)
We have two instance on the same server but we do not use more than 1 G per
instanc(BUFFERS & SHMVIRTSIZE) with approx max 250 users between both
instances. Using BaaN IV b3 & C3. Although we connect all users as remote
users from other servers.
We have never had to use chatr under informix.
Cheers
Pankaj Patel
Susan Elliott (ISG) wrote in message <80304n$e3c$1@news.xmission.com>...
>
>
>Howdy Folks,
>
>We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We
>are running on HPUX 10.20, Application is Baan.
>
>Since we have performed this upgrade, we are expierencing the database
>gradually becoming slower and slower. Load averages on the top display are
>getting higher over the period of the day (around 5) which is pretty high
>for our system. The database seems to require more than 900 meg of memory
>as it seems to dynamically allocate more segments. B4 the upgrade we were
>using below the 900meg mark. I have been advised to alter the Informix
>binaries with "chatr -M" I have tested this on a test system but I am
>getting error messages on startup of DB. I am wondering if I have missed
>something....
>
>Has anyone already set up / using this on a HPUX system, using more than 1
>gb of shared memory (7.30 Uc7).
>
>Would you please advise the process you used / followed?? What issues did
>you come across implementing chatr ??
>
>
>TIA
>
>Regards
>Suze.
>
>
>Please refer to the attached copy of my online.log
>
>
>Fri Nov 5 14:02:19 1999
>
>14:02:19 Informix Dynamic Server Stopped.
>16:00:52 Segment locked: addr=0x40000000, size=956596224
>16:00:52 The number of configured CPU poll threads exceeds 'NUMCPUVPS'.> NETTYPE 'baantcp' poll threads started on NET VPs.
>
>
>Sun Nov 7 16:00:52 1999
>
>16:00:52 Event alarms enabled. ALARMPROG =
>'/opt/informix/etc/ifx_alarm.sh'
>16:00:58 DR: DRAUTO is 0 (Off)
>16:00:58 Requested shared memory segment size rounded from 588KB to 592KB
>16:00:59 Informix Dynamic Server Version 7.30.UC7 Software Serial Number>AAC#J578294
>16:01:00 Informix Dynamic Server Initialized -- Shared Memory Initialized.
>16:01:00 Physical Recovery Started.
>16:01:00 Physical Recovery Complete: 0 Pages Restored.
>16:01:00 Logical Recovery Started.
>16:01:03 Logical Recovery Complete.> 0 Committed, 0 Rolled Back, 0 Open, 0 Bad Locks
>
>16:01:03 Onconfig parameter LTAPEDEV modified from /dev/rmt/0m to
>/dev/null.
>16:01:03 Dataskip is now OFF for all dbspaces
>16:01:03 HPUX Version B.10.20 -> Using flag/select style KAIO
>16:01:04 On-Line Mode
>16:01:04 Checkpoint Completed: duration was 1 seconds.
>16:01:54 Informix Dynamic Server Stopped.>
>I altered the onconfig for SHMVERT to be approx 1200 meg and started the
>intance. Error messages below.
>
>16:05:00 shmat: [EINVAL][22]: shared memory base address illegal
>
>16:05:00 using 0x60000000, needs 0xc1483000
>
>16:05:00 Could not create single shared memory segment with resident and>non-resident par
>
>titions. Proceeding to create 2 shared memory segments instead.
>16:05:03 Segment locked: addr=0x40000000, size=117719040
>16:05:03 shmat: [EINVAL][22]: shared memory base address illegal
>
>16:05:03 using 0x67044000, needs 0x80000000
>
>16:05:03 mt_shm_init: can't create virtual segment>
>uxfgrn4:/opt/informix>
>
>
>Susan Elliott
>______________________
>Host Team Leader
>Information Services Group
>Concrete Group
>* (09) 520 8608
>* susane@fclcis.co.nz
>
>
>
"Susan Elliott (ISG)" wrote:
> Howdy Folks,
>
> We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We
> are running on HPUX 10.20, Application is Baan.
>
> Since we have performed this upgrade, we are expierencing the database
> gradually becoming slower and slower. Load averages on the top display are
> getting higher over the period of the day (around 5) which is pretty high
> for our system. The database seems to require more than 900 meg of memory
> as it seems to dynamically allocate more segments. B4 the upgrade we were
> using below the 900meg mark. I have been advised to alter the Informix
> binaries with "chatr -M" I have tested this on a test system but I am
> getting error messages on startup of DB. I am wondering if I have missed
> something....
>
> Has anyone already set up / using this on a HPUX system, using more than 1
> gb of shared memory (7.30 Uc7).
>
> Would you please advise the process you used / followed?? What issues did
> you come across implementing chatr ??
>
> TIA
>
> Regards
> Suze.
>
> Please refer to the attached copy of my online.log
>
> Fri Nov 5 14:02:19 1999
>
> 14:02:19 Informix Dynamic Server Stopped.
> 16:00:52 Segment locked: addr=0x40000000, size=956596224
> 16:00:52 The number of configured CPU poll threads exceeds 'NUMCPUVPS'.> NETTYPE 'baantcp' poll threads started on NET VPs.
>
> Sun Nov 7 16:00:52 1999
>
> 16:00:52 Event alarms enabled. ALARMPROG =
> '/opt/informix/etc/ifx_alarm.sh'
> 16:00:58 DR: DRAUTO is 0 (Off)
> 16:00:58 Requested shared memory segment size rounded from 588KB to 592KB
> 16:00:59 Informix Dynamic Server Version 7.30.UC7 Software Serial Number> AAC#J578294
> 16:01:00 Informix Dynamic Server Initialized -- Shared Memory Initialized.
> 16:01:00 Physical Recovery Started.
> 16:01:00 Physical Recovery Complete: 0 Pages Restored.
> 16:01:00 Logical Recovery Started.
> 16:01:03 Logical Recovery Complete.> 0 Committed, 0 Rolled Back, 0 Open, 0 Bad Locks
>
> 16:01:03 Onconfig parameter LTAPEDEV modified from /dev/rmt/0m to
> /dev/null.
> 16:01:03 Dataskip is now OFF for all dbspaces
> 16:01:03 HPUX Version B.10.20 -> Using flag/select style KAIO
> 16:01:04 On-Line Mode
> 16:01:04 Checkpoint Completed: duration was 1 seconds.
> 16:01:54 Informix Dynamic Server Stopped.>
> I altered the onconfig for SHMVERT to be approx 1200 meg and started the
> intance. Error messages below.
>
> 16:05:00 shmat: [EINVAL][22]: shared memory base address illegal
>
> 16:05:00 using 0x60000000, needs 0xc1483000
>
> 16:05:00 Could not create single shared memory segment with resident and> non-resident par
>
> titions. Proceeding to create 2 shared memory segments instead.
> 16:05:03 Segment locked: addr=0x40000000, size=117719040
> 16:05:03 shmat: [EINVAL][22]: shared memory base address illegal
>
> 16:05:03 using 0x67044000, needs 0x80000000
>
> 16:05:03 mt_shm_init: can't create virtual segment>
> uxfgrn4:/opt/informix>
>
> Susan Elliott
> ______________________
> Host Team Leader
> Information Services Group
> Concrete Group
> * (09) 520 8608
> * susane@fclcis.co.nz
There was a problem of memory allocation in some older versions of Informix,
where as rather than freeing and allocating memory when required, Informix
just allocated new memory, as a result the processes used to grow and become
slower and eventually die out ( Fatigue). The best solution was to reboot the
system from time to time. I was under the impression that this bug was fixed
long ago.
PS: This bug was present both in 4GL Compiler and DBMS.
--
Compliments of QueriX
--------------------------------------------------------------------------------------------------
QueriX 4GL Compilers are Informix 4GL Compatible. Some features are:
True Windows GUI Clients.
ActiveX support.
HTML Report Generation.
More rigorous error handling.
Connection to all versions of Informix 4GL with no need to change compiler.
Connection to other RDBMS such as Oracle.
For more details visit: http://www.querix.com/
---------------------------------------------------------------------------------------------------
Hello Susan,
Your question has two parts.
About chatr:
You don't have to use chatr if you aren't going to give
Informix MORE than 1.75Gb. In other case you should work with
chatr AND shm_magic.
About memory:
You crossed with HP shared memory mechanism. On HP 10.20 (32 bit)
the process address space is divided into four quadrants. On a
32 bit system, these quadrants are each 1 GB. The size of the
quadrants
limits the amount of address space available to processes and a
single shared
memory segment cannot cross quadrant boundaries.
Informix tries to allocate 900Mb shared memory segment in third
quadrant. It's impossible
for third quadrant, because system shared libraries are in the
quadrant too. You can divide your
Virtual and Resident segment (use SINGLESEG_OFF=1 environment
variable for oninit)
and placed the segments in third and fourth quadrants.
Also your problem looks like you have memory leak. The better way is
investigation which application
cause the leak and why. After that you'll be able to open a case at
Informix Tech Support ;-) or
resolved the problem yourself.
Good luck,
Eugene Nechayev
In article <80304n$e3c$1@news.xmission.com>,
"Susan Elliott (ISG)" <SusanE@fclcis.co.nz> wrote:
>
>
> Howdy Folks,
>
> We have just recently upgraded our database from 7.24uc5 to 7.30 uc7.
We
> are running on HPUX 10.20, Application is Baan.
>
> Since we have performed this upgrade, we are expierencing the database
> gradually becoming slower and slower. Load averages on the top
display are
> getting higher over the period of the day (around 5) which is pretty
high
> for our system. The database seems to require more than 900 meg of
memory
> as it seems to dynamically allocate more segments. B4 the upgrade we
were
> using below the 900meg mark. I have been advised to alter the
Informix
> binaries with "chatr -M" I have tested this on a test system but I am
> getting error messages on startup of DB. I am wondering if I have
missed
> something....
>
> Has anyone already set up / using this on a HPUX system, using more
than 1
> gb of shared memory (7.30 Uc7).
>
> Would you please advise the process you used / followed?? What issues
did
> you come across implementing chatr ??
>
> TIA
>
> Regards
> Suze.
>
Sent via Deja.com http://www.deja.com/
Before you buy.
Hi! I have heard that there is a specific problem with virtual shared memory allocation in IDS systems when you have HP-UX 10.XX on a PA-RISC. For example, Joe Lumpley writes in his "Informix DBA survival guide", page 271, as follows: "One higly implementation specific caveat comes in systems running on the HP/UX 10.XX on PA-RISC processors. HP systems have some problems with algorithms for adding shared memory segments. These systems -- should never be allowed to grow beyond three shared memory segments. Once the forth segment is added performance goes into toilet." He suggests to adjust SHMVIRTSIZE big enough and then disallow any dynamic growth of the shared memory beyond first three segments. For more info you'd better read the book as I do not have much experience on this. Recards Katja, Finland In article <1EF0AC9A9B84C78F.22E0A05243808CD0.E0E4FAFD1E00A107@lp.airnews.net>, Greg <greg@fastlane.net> wrote: > There's a nasty buffer management bug that afflicts releases prior to > 7.30.UC7XK. The system works real hard, but nothing gets done very fast. The > only real solution is to upgrade to a later release that doesn't have the > problem. It drove a lot of SAP admins nuts for a few months until a solution > became available. > > Also, in a 32 bit HP environment, no single segment can be bigger than 1GB, so > don't bother trying. > > I doubt chatr or shmvirtsize have anything to do with your performance > problems. > > Greg > > "Susan Elliott (ISG)" wrote: > > > Howdy Folks, > > > > We have just recently upgraded our database from 7.24uc5 to 7.30 uc7. We > > are running on HPUX 10.20, Application is Baan. > > > > Since we have performed this upgrade, we are expierencing the database > > gradually becoming slower and slower. Load averages on the top display are > > getting higher over the period of the day (around 5) which is pretty high > > Sent via Deja.com http://www.deja.com/ Before you buy.