Hi to all

We are migrating our IDS 7.30 on Linux RH 7.2 (working fine but with strange but documented network error on same heavy connection) with the 9.30_UC1 version.

We have now big problem of performance: import operation are too much delayed operation on the new version (5 times more!!!).

We replicated the same onconfig and kernel parameters (our machine is a 1Gb RAM):

echo 250 32000 100 128 > /proc/sys/kernel/sem
echo 536870912 > /proc/sys/kernel/shmmax

Another issue: we are trying to use the CSDK 2.30 on Win32 to connect to IDS 9.30 (and ODBC Informix 3.30); is it possible or I have to upgrade to CSDK 2.70?

Thanks for any support.


karlo

You might want to check your release/machine notes for the correct kernel parameters and other issues in $INFORMIXDIR/release.
About your CSDK. The latest version is 2.80. And yes, it is possible.

Yes!

Maybe the problem is related to the libpthread.so!

During the IDS 9.3 installation on RH 7.2 I've not changed the link to the redistribuited version of this lib: $INFORMIXDIR/lib/libifxpthread.so.

But on RH 7.2 I have glib 2.2.4, and the patched lib seems to be related to glib 2.1! (the MACHINE SPECIFIC NOTES said This product was built on RedHat Linux 6.2 and is targeted for Linux Kernel 2.2.14-5 or higher and glibc 2.1.3-15 or higher.).

Is it right to change the libpthread.so also on RH 7.2?

Thanks again.

karlo

I've checked more deeply my libifxpthread.so situation, and I'm already using the Informix patched version (based on glibc-linuxthreads-2.1.3).

Maybe using this version under RH 7.2 is bad... any experience?

karlo

If you don't encounter any problems, I think you should leave it that way. Check your message log(s).

I'm agree with you!

Our actual problem is the very long time for import operations compared to IDS 7.3 environment and the same hardware/SO.

The only difference on the new linux-box is that we add a second HDD EIDE and IDS 9.3 has now chunck on both the devices; could it be a problem?

The online transactional operations with 25 users seem to be ok as response time.


karlo

Check the io queue of the devices with onstat -g ioq
Check the maxlen column. It should not exceed 25. Otherwise your disks can't follow up.

Look at this report: maybe we have some kind of problem with i/o!

We are also tunning the logical log size: the online.log return an overflow!

Thanks again for the help! We are not so much skilled on Informix (our target db is Oracle).

karlo

-----------------------------------------------------------------------------
[informix@infolinux2 etc]$ onstat -g ioq

Informix Dynamic Server Version 9.30.UC1 -- On-Line -- Up 03:09:03 -- 68044 Kbytes

AIO I/O queues:
q name/id len maxlen totalops dskread dskwrite dskcopy
adt 0 0 0 0 0 0 0
msc 0 0 1 1823 0 0 0
aio 0 0 1 33830 422 33336 0
pio 0 0 1 434 0 434 0
lio 0 0 1 409 0 409 0
gfd 3 0 2 485 367 118 0
gfd 4 0 151 52654 25660 26994 0
gfd 5 0 25 34379 34163 216 0
gfd 6 0 23 61950 61710 240 0
gfd 7 0 151 135199 45784 89415 0
gfd 8 0 151 1013723 56130 957593 0
gfd 9 0 150 199004 6246 192758 0
gfd 10 0 58 527 246 281 0
gfd 11 0 1 61 61 0 0
gfd 12 0 1 2 2 0 0
gfd 13 0 1 1 1 0 0
gfd 14 0 1 33 33 0 0
gfd 15 0 1 1 1 0 0
gfd 16 0 1 1 1 0 0
gfd 17 0 1 1 1 0 0
gfd 18 0 1 1 1 0 0
gfd 19 0 3 3 0 3 0

It's a sad thing to hear that your target DB is Oracle. It's sad that marketing wins over technology!

How many AIO VP's did you configure?
onstat -g sch shows you an overview of all Virtual Processors.
These Virtual Processors make sure the IO gets done. Linux does not support Kernel IO, otherwise your IO could run on the main CPU VP.

Now the I/O can be monitored with several onstat commands. Check onstat -- |grep IO

onstat -g iov (Check io per second column)
onstat -g iof (io per chunk, similar with onstat -D)
onstat -g iog (some global info about IO)

here is absolutely something wrong about your data/IO spreading.
gfd 4 0 151 52654 25660 26994 0
gfd 5 0 25 34379 34163 216 0
gfd 6 0 23 61950 61710 240 0
gfd 7 0 151 135199 45784 89415 0
gfd 8 0 151 1013723 56130 957593 0
gfd 9 0 150 199004 6246 192758 0
gfd 10 0 58 527 246 281 0


Meaning chunk 2,3,4,5,6,7,8
Are overloaded. Check with
onstat -d to see what the chunk name is and chech if they reside on the same disk.
The other chunks are almost doing nothing!

Did you use raw devices or cooked files to store your data?

Finally we are understanding something more about Informix! Thank you a lot!!!!

[by the way: Oracle is only a "marketing toy"? You are the first Informix defender I know!]

On the 7.3 we not configured AIO VP's and we had 30 (maybe a default?)!

On 9.3 we use a standard onconfig file and the AIO VP's was 2! Now we putted 30 to have a similar situation as the old one!

It's not clear to me how to tune this VP (also for other classes).. I've to study deeply.


Regarding data/IO spreading (our chunck are "coocked file", not raw), how is it possible to balance and "defrag" chunckes usage?


karlo

There is no default value for the AIO VP's.
You can check it in yr old instance (7.3) with onstat -c |grep AIO
A new parameter is introduced in the 9.3 version:
VPCLASS
With this parameter you can specify the VP class and the number of VP's you want to define.
Raw devices are A LOT faster.
There is no way of balancing IO manually over chunks. You can only fragment tables/indexes over dbspaces.

How did you create a raw disk on redhat 7.x, im trying to make one as a character device, but cant find much about it??
If you could give me an example, it would be great

thanks any way
Rene