HP3500 losing configuration and poor ping response

[Matt Hobbs]

You can add this one to the list as well:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1040132

Is it only routed traffic that is seeing the problem? I'm sure it will be, just want to make sure.

Do you think you could reproduce this in a lab setup? If I can get my hands on a 3500/5400 next week I'll see I can punish it enough to see it.

[Drew_38]

Mike;

Very interesting comments thank-you. glad (if glad is the word) that it's not just me who is experiencing this. This is why it's tricky for me to reproduce it in a test environment as with no load the 3500 performs perfectly ok.

Matt;

Again thanks for spotting the other similar thread. Yes it is only traffic which is routed away from the 3500 which is experiencing slow ping reponses. Having said that I only have one subnet which has any users on it so pretty much everything they connect to is off the subnet.

I'll report this problem to HP then now I think I can construct a job to them which kinda makes sense.

I suspect basically it'll be something they need to sort out in the next software version. Wonder when that might be?

Drew

[doubleH]

Hi Guys,

I've worked with Matt before on some other performance issues and he has helped out large. I am having the same sort of issues Drew is describing. I am just starting to migrate on to new HP gear and am not having very good response times in some cases either. I have 2 5400's in the core connected to each other with a 4gig trunk with a 3500 edge device (not in use yet). All are running K.11.33. There have been no spanning tree configuration changes in 14 days.

I don't mean to hijack this thread, but would like to explain my current issue. If you would like I will post to a new thread...

Server 1, HP ML530, W2K Sp4, HP NC7771 Copper Gig Nic, both nic and switch port running 1000/Full. Attached to Core 2, A7

Workstation 1, HP with onboard Broadcom Gig Copper card, WinXP SP2 (no firewall), both nic and switch port are set to auto. Attached to Core 2, D10

If i load iperf on the server and run from the workstation I get 10mbit/sec which is ridiculous. Server is not under much load. If I then take that same workstation connection and plug it into Core 1 the speeds jump up to 250-300mbit/sec. It doesn't make sense to me why I get better speeds going between the 2 cores compared to being on the same core and getting down right nasty speeds. If I run the same test from a W2K workstation on my old network running at 100/Full (forced) I get 85mbit/sec which what I would expect.

[doubleH]

Just thought i'd post back. It's 8:30pm EST and everyone has gone home. When running the iperf tool from a workstation on Core 2 (same as server) I am now getting 200+mbit/sec whereas before I was getting 10mbit/sec. Could it be another device that is causing the slow down? I still have better results on Core 1.

Thanks

[doubleH]

i did more testing and posted the results to my original thread...

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1036281

[doubleH]

i've had this problem with my 5406's and 3500 for a few months now. today we are working on the theory the there is an incompatibility with these switches and PCM+. go ahead and either shutdown the device that is running PCM+ or if you can't just stop the 3 procurve services and you may see your issue go away.

[BOVC]

Hi,

Matth Hobbs posted a link to a post that I initially started about performance issues on the 5400zl.

We actually solved the problem, but not the best way :)

It turned out, that when we set up a computer with PCM+ (newest version), after a couple of hours the main switch (which is doing all the routing) exprienced exceptionally poor performance (but only on routed traffic). We pulled the plug on the management server, and restarted the switch - and gone were the problems.

Some days later (and Update 6, instead of Update 5 for PCM) we tried again - with the exact same result!

Again we turned of PCM+, and everything has been working fine ever since.

I called HP about this, but since we were having no problems anymore, I don't think the tech guy took it too seriously. Anyway, I told him that because of this, we would not buy PCM+ (which we had thought of before this).

Rasmus

[doubleH · Accepted solution]

my issue has been escalated to the level 3 engineers for a few weeks now. yesterday they were able to reproduce the issue in their own lab like a few of us are experiencing. they are looking at this as a bug in PCM and should be fixed.

in the mean time i have PCM running, but just remove traffic monitoring from all your 5400/3500 and the issue goes away. they mentioned something with sFlow is the issue. at least this way you can still run PCM, just not the traffic monitoring.

[Drew_38]

Thanks all for your recent responses. Can this really be right that it is PCM triggering the problem? How incredible. I really am losing faith in HP and their ability to deliver reliable network equipment. Moreover the (European) engineers who I have been dealing with when trying to report this problem are so way off the mark it's laughable. Last communication we had from them they were trying to blame 3Com switches elsewhere on our LAN (when we don't have any).

Anyhow I'll try and turn off PCM traffic monitoring and see if the problem goes away.

Obviously it isn't actually a problem with PCM, it's more how the 35xx and 54xx series respond to data collection requests from PCM. I love PCM it's great so I'd not be put off buying that.

Drew

[doubleH]

i know it sounds weird, but i've been banging my head against the wall for 2 months now. this actually makes sense for what i was seeing and i can reproduce the problem on my end no problem.

if stopping the traffic monitoring deosn't work just shutdown the pcm machine or just stop the 3 procurve services.