1810g-24 latency, lag, pain

[tractor]

Hi,
I am an owner of a 1810g-24 switch. I am having a significant latency problem that I can't figure out.

I use a PXE boot server to image PC's remotely. I put it on a test bed with a cheapo linksys switch and have been able to deploy and upload images over the network in about 20 minutes.
Moving toward using it in a working environment, I connected it to this 1810g-24 switch, and times went from 20 minutes to 5 hours.

Just to make sure I wasn't introducing other variables, I grabbed that same cheap linksys 4 port switch and swapped the cables over to it and now I'm back at 20 minutes. I've also tried using the same cables and ran them to our Cisco catalyst and the times are slightly less even.

Is there a setting in the HP switch that would cause this kind of symptom? There are no other workstations on the switch. Just my PXE server and my test mule workstation.

[rick jones]

I'm not familiar with the 1800s, but that magnitude of performance change smells like packet loss and retries. You might try packet tracing on your server and see what you see there during the PXE boots. If I recall correctly, the timeouts for TFTP retries are quite long in comparison to typical RTTs.

[tractor]

Thanks. I recommended this switch to my boss (a Cisco fan), so I'm feeling the heat to get this working. :)

[rick jones]

For reasons unrelated to this, an 1810g-24 happens to have crossed my path. I also happened to have a Z400 workstation and 8510w laptop running Ubuntu Maverick and a private network between them, with a tftp server enabled on the Z400. I was able to tftp a 13711983 byte file in roughly 2 seconds:

raj@raj-8510w:~$ tftp 192.168.1.2
tftp> get K_15_03_0007.swi
Received 13799190 bytes in 2.0 seconds
tftp> get K_15_03_0007.swi
Received 13799190 bytes in 2.0 seconds
tftp> get K_15_03_0007.swi
Received 13799190 bytes in 2.0 seconds
tftp> get K_15_03_0007.swi
Received 13799190 bytes in 1.9 seconds
tftp> get K_15_03_0007.swi
Received 13799190 bytes in 1.9 seconds

and it was like that for the various combinations of ports I was using.

BTW, here are the salient performance specs from the quickspecs:

http://h18000.www1.hp.com/products/quickspecs/13447_div/13447_div.HTML#Technical%20Specifications

[Mohammed Faiz]

Have you checked that the ports are negotiating to the correct speed? I believe the 1800 series are web-interface only and I'm not that familiar with it but there should be somewhere that you can check.
It may be worth just forcing the speed to 100-full even if it looks ok, just as a test.
Also, are you using multicast with your PXE server? If so you may want to enable IGMP (snooping) on the relevant VLAN.

[tractor]

I got on the line with HP support. They said they have heard of this issue before, and had a newer firmware version which hadn't been released to the public yet. Ver. 2.2

I applied the update. You need to update to 1.20 first which I had to get from them also.

The update did not solve the problem.

I turned 'flow control' on and the switch suddenly sped up immensely. I was under the impression that turning flow control on might slow things down because of the heavy traffic that would occur in my situation.

Although the speed is now acceptable, the imaging process does pause periodically for a few seconds. This must be the flow control kicking in, because this doesn't happen on other switches I've used.

Thanks

[rick jones]

Were you able to get any packet traces of what was going-on?

[Jeff Brownell]

With flow control enabled, how long does it take to load an image on the 1800 now? (20 minutes for checpo Cisco)