Network problem...

[Sreekanth Makam]

Hi folks,
Iam running around 100 users(through loadrunner). My application is informix 4gl. There are many system call. Allways informix is like that only...

Now glanceplus is showing 100% network bottelneck. I kept my loadrunner machine in same lan where ny server is there. Allmost it is directly connected.

What are all the resons may be for this network bottelneck.

With out glance plus..That is through command prompt ..how to check this bottelneck..

Please explain me parameters that i want to see regarding network..by that i can guess what may be the wrong..

Please help out this netwrok issue.

*** problem is a problem untill u feel that is a problem ***

[SAHA]

Did you tried lanadmin command or interactive to check the statistics of lan interface card ?

Thanks,

[Bharat Katkar]

hi,

1. Check NIC status using lanadmin
2. See your NIC's are configured in FULL duplex / 100 Mbps mode.
3. Look out for any faulty NIC in the network in case it is producing unnecessary broadcast.
4. Check for faulty patch cords as well.

Hope that helps.
Regards,

[Tim D Fulford]

Hi

1 - What type of network cards you are using? "lanscan" & "ioscan -fknClan"
2 - what speeds & duplex are they set to are they set to (lanadmin will tell you if you do not know)
3 - glance has fairly hardwired limits based on packets per second. The last time I looked they were VERY low (it is probably versio dependat, so C.02.4 will be lower than say C03.70!!) so the bottleneck may not actually exist.
4 - What packet rates & bandwidth are you getting (glance -l).

If you can answer the above then it may be possible to help more objectively

Regards

Tim

[Tim D Fulford]

Hi again..

I previously answered a question of yours on Hgh CPU utilsation. my outstanding questions are still
1 - please supply "onstat -g glo"
2 - please supply "onstat -c" or ONCONFIG file

I suspect the network & high cpu could be linked to poor Informix configuration... It's only a hunch though...

Tim

[Sreekanth Makam]

Hi tim,
Please find the attachment for details of onconfig and onstat -g glo output..

[Tim D Fulford]

Hi

Hmm as I suspected. Informix is being knobbled... badly... very badly... very very badly

1 - The number of cpu-vps is set to 36, you only have a 9- physical CPU's in your partition (from previous post) so this is way too high, it will cause massive ammounts of contension & probably high system calls.
2 - You have specifically configured 6 soc-vps poll threads. The 36 cpu-vps are doing slightly in total more work than these 6 socket vp-s... VERY bad. I've put them onto the cpu-vps for efficiency. If you have to do network loading then this is your only choice, but if you can use the shared memory protocol, this would be better again (ipcshm) and change the NETTYPE's appropriately
3 - You have 12 Logical log files of just under 10MB each. This is not enough. My instance is much smaller than yours and I have some 300 16MB logfiles. Consider having say 200-400 16MB logs in a seperate DBspace
4 - Your root physical (& probably logical) log DBspace are all in the root DBspace. this will not help
5 - You have configured processor affinity. your trying to deal with 36 cpu-vp. these may well have locked you out after you reached 9 (the number of physical CPUs in your system).
6 - You have configured some 3 million or 6GB memory as buffers and 640 LRU queues. But inly 128 cleaners. Do you really need 6GB shared memory!! I will assume you do need the full 6GB. I think that the number of LRU's could reduce to 128. OR set LRU & CLEANERS to about 256. It is a good idea to keep LRU=CLEANERS
7 LRU_MIN & MAX. The setting here will do lots of flushing between checkpoi9nts to reduce the checkpoint timmings. This is fine but your checkpoint interval is 1200, so 20 mins. I suggest changing CHPTINTVL 300 and relly reduce the ammount of data being flused at checkpoint time.
8 - Lastly. If you TURN OFF processor affinity you could have NUMCPUVPS = number of physical cpus in system as no provcessor will affinitise itself to the monarch processor (CPU0). You could then increase the number of poll threads by 1

I would additionally reccomend that you ask IBM-Informix to help out or get some Informix training/tuning experience.

I hope the above improves you processing.

Thanks for the posting the outputs it was interesting... please do say if the new ONCONFIG helps..

Regards

Tim
Tim