Delay in commands returning
Answered: red (solid confidence) — Several plausible diagnostic avenues are suggested (truss, onstat -g spi, auditing, GLS/msg file opens, general startup-time comparison) but the thread ends without anyone identifying the actual cause or confirming a fix.
Advisory only.
Posted in 2009
Neil Truby saw about a one-second delay for any onstat command on a Sun M9000 running IDS 10.00.FC9, versus milliseconds on a v890 and Linux; shared memory made no difference. david suggested onstat -z with onstat -g spi and tracing with truss -d. Fernando Nunes asked about auditing. Andrew Clarke suggested comparing process startup and library attach costs. Unresolved.
Topics: Networking & sqlhosts Configuration, Platform-Specific Issues
Delay before resonding to commands.
On a certain server type (Sun M9000), and maybe others, I get about a
second's delay issuing any onstat command ... even onstat -m. See below,
took about one second.
On any other server - a Sun v890, a Linux server, it takes a few
millseconds.
I've tried a shared-memory rather than TCP/IP connection. No difference.
It doesn't seem to be causing any issue, I'm just curious ...
Thanks
Neil
*****************************************************************************
-bash-3.00$ time onstat -p
IBM Informix Dynamic Server Version 10.00.FC9 -- On-Line -- Up
00:02:41 -- 25968640 Kbytes
Profile
dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits
%cached
7256 7738 161766 95.51 141 141 32
0.00
isamtot open start read write rewrite delete
commit rollbk
165785 8971 16332 54645 0 0 0
0 0
gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs
0 0 0 0 0 0 0
ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes
0 0 0 29.99 48.62 1 4
bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress
seqscans
0 0 127852 0 0 0 8
5
ixda-RA idx-RA da-RA RA-pgsused lchwaits
0 0 184 184 54368
real 0m1.027s
user 0m0.006s
sys 0m0.014s
On 21 Oct, 19:26, "Neil Truby" <neil.tr...@ardenta.com> wrote:
> Delay before resonding to commands.
>
> On a certain server type (Sun M9000), and maybe others, I get about a
> second's delay issuing any onstat command ... even onstat -m. See below,
> took about one second.
>
> On any other server - a Sun v890, a Linux server, it takes a few
> millseconds.
>
> I've tried a shared-memory rather than TCP/IP connection. No difference.
>
> It doesn't seem to be causing any issue, I'm just curious ...
>
> Thanks
> Neil
> *****************************************************************************
> -bash-3.00$ time onstat -p
>
> IBM Informix Dynamic Server Version 10.00.FC9 -- On-Line -- Up
> 00:02:41 -- 25968640 Kbytes>
> Profile
> dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits
> %cached
> 7256 7738 161766 95.51 141 141 32
> 0.00
>
> isamtot open start read write rewrite delete
> commit rollbk
> 165785 8971 16332 54645 0 0 0
> 0 0
>
> gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs
> 0 0 0 0 0 0 0
>
> ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes
> 0 0 0 29.99 48.62 1 4
>
> bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress
> seqscans
> 0 0 127852 0 0 0 8
> 5
>
> ixda-RA idx-RA da-RA RA-pgsused lchwaits
> 0 0 184 184 54368
>
> real 0m1.027s
> user 0m0.006s
> sys 0m0.014s
Well onstat just attaches to shared memory so the connection type is
irrelevant.
What about "time onstat -" how long does that take??
Before hand do onstat -z then onstat -g spi afterwards, what latches
are taken?
Try truss -d onstat - where is the time taken?
onstat, at least for version 7 seems to open and close a lot of files
under $INFORMIXDIR/gls and $INFORMIXDIR/msg
Neil Truby wrote:
> Delay before resonding to commands.
>
> On a certain server type (Sun M9000), and maybe others, I get about a
> second's delay issuing any onstat command ... even onstat -m. See
> below, took about one second.
>
> On any other server - a Sun v890, a Linux server, it takes a few
> millseconds.
>
> I've tried a shared-memory rather than TCP/IP connection. No difference.
>
> It doesn't seem to be causing any issue, I'm just curious ...
>
> Thanks
> Neil
> *****************************************************************************
>
> -bash-3.00$ time onstat -p
>
> IBM Informix Dynamic Server Version 10.00.FC9 -- On-Line -- Up
> 00:02:41 -- 25968640 Kbytes>
> Profile
> dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits
> %cached
> 7256 7738 161766 95.51 141 141 32 0.00
>
> isamtot open start read write rewrite delete
> commit rollbk
> 165785 8971 16332 54645 0 0 0
> 0 0
>
> gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs
> 0 0 0 0 0 0 0
>
> ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes
> 0 0 0 29.99 48.62 1 4
>
> bufwaits lokwaits lockreqs deadlks dltouts ckpwaits
> compress seqscans
> 0 0 127852 0 0 0 8 5
>
> ixda-RA idx-RA da-RA RA-pgsused lchwaits
> 0 0 184 184 54368
>
>
> real 0m1.027s
> user 0m0.006s
> sys 0m0.014s
>
Do you have auditing turned on? I assume not, but just in case...?
Regards.
what about startup times of simple things like vi? Is it a machine difference?
Longer $PATH or $LD_LIBRARY_PATH with more or less to wade through? OS
differences in the the implementation of "sticky" executables (ie retaining
the executable in memory on the off-chance it will be called again) or maybe
on one platform it's more expensive to attach to shared libraries; onstat
probably attaches to several while vi probably doesn't...
There's a process tracing tool that shows the system calls although the name
escapes me. You could probably use that to know which part of program startup
is taking the most time.
Related threads
- Informix 9 to 10
- filetoblob error with /dev/null
- Re: NVL changing data type of a datetime?
- oninit -r