Slow network with HP ProCurve 28xx L3 Switch Family

hi there,

I do have a client where we have brand new hp procurve switches

Network Hardware:
main cluster comander: 2848 (10/100/1000)
member 2824 (10/100/1000)
member 2650 (10/100)
member 2650 (10/100)
member 2626 (10/100)

network is kath6 with all ports GigabitOK on the network protocol form the electrician.

I am running the most recent firmware on the swtches.

Computer Hardware:
32x I-macs 10/100 (autosensing)
12x G5 10/100/1000 (autosensing)
12x IBM Laptops 10/100
2x IBM Servers i do think 1000

all ports are auto sensing also the ports on the computers.

we tested network performace with Helios LanTest (perferences for gigabit network):
writing 6.4mb per sec
reading 36 mb per sec

we tried to isolate the problem and just connect the server and 5 g5 to one 2828 (10/100/1000) and took a brand new one (with standard config)
still the slow write performace.

then we took a cheep level1 gigabit swich and run the same test.
guess what
writing 31mb per sec
reading 33 mb per sec

crazy now?
shouldn't a level3 switch form hp with the standard configuration do a better job?
does anybody has an explanation for that?

thanx christof

Posted on May 20, 2005 7:26 PM

Reply
8 replies

May 21, 2005 6:27 AM in response to reco

"main cluster comander" ??? don't know anyting about that.
I only know I did get good performance with a HP 2824 - "plain" configured.

Level 3 means routing so depending on your setup maybe the switches are wasting time routing packets the wrong way (through a 100Mbit segment/gw)?

I know of one BIG installation that had terrible performance because of routing and "bad firmware" problems in the switches.

You are running windows 2000/2003 server on the IBM servers?
Maybe you should look into configuring your Macs to have delayed ack off.
Take look at what Broadband optimizer does. You can also set your own startup parameters in 10.3 (and up?) with a /etc/sysctl.conf file.

Jun 21, 2005 3:06 PM in response to reco

FWIW we had LOTS of trouble with network home folders in a lab where all eMacs were on a procurve 2650. That 2650 was connected through another 2650 to a procurve 5300 series switch. (all gigabit.) We discovered two problems: outdated firmware, AND, we had to turn OFF the LACP setting on ALL ports to which macs were connected. Now, we have zero problems in that lab.

Sep 9, 2005 11:39 AM in response to reco

Yes this works well with 2500 series too. Here are the commands (quoting Scooter at the HP discussion board)
"
Log into the switch over console or telnet. Then in the CLI type the following commands:

2650# config t
2650(config)# no interface all lacp
2650(config)# write mem
2650(config)# show lacp

The response from the switch should be "no LACP ports found".

Kind regards,

SCOOTE"

Here's the link to that thread: http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=711325

Wish I found it sooner.

Sep 9, 2005 11:54 AM in response to Jonathan Roberts1

hi jonathan,

also disable qos passthrough if you have 100mbit ports.

hp is saying if you connect for example one 2848 (gbit) with a 2650(100mbit with 1 gbit ports) even if they are connected via gbit on both swiches you run into delays and packages loss problems. which makes the network really slow.

with
lacp off
qos passthrough off

I have my best experience 🙂

christof

Sep 9, 2005 12:46 PM in response to Jonathan Roberts1

hi jonathan,

I do have a mixed environement.
the speed troubles i had on win and mac.

hp tech in generell said package loss could happen if you mix 28xx and 26xx devices (even switches connected with 1gbit link) and have qos passthrough on.

the way he explained to me:
if you have a 100mbit client taking to a gbit server the buffer cahes on the 1000gbit port the server is attached are to big.

buttom line from support:
qos passthrough is for gbit switched networks only. and there you should see a better perfomance then.

I didnt ask further cause since i turned it off i had expected performaces.
a G5 writing to a Win2003 server scsi raid 78mb read and write per sec.
tested with helios lan test (www.helios.de)

christof

Sep 20, 2005 7:53 AM in response to reco

Hi Christof,

Thanks for your help. I tracked down the release notes on this. They confirm the info that the hp tech gave you, EXCEPT, they say that qos-passthrough is a SOLUTION to the problem of 10/100/1000 mismatches on the 2800. It is off by default. I will test when I have an opportunity. I was wondering whether you meant that you turned it ON, not OFF.

QoS Pass-Through Mode on the Series
2800 Switches
Release I.07.52 introduced a new command to enhance the performance of
line-rate traffic transfers through the Series 2800 switches. This feature should only be used in environments where Quality of Service (QoS) is not of major importance, but where lossless data transfers are key. This command essentially disables any discrimination of QoS queues for traffic, consolidating packet buffer memory to provide line-rate flows with no loss of data.
General Operation
The port buffering design for the Series 2800 switches has been optimized forgigabit-to-gigabit traffic flows. For this reason, some flows from Gigabit-to-100Base or even 100Base-to-10Base may not perform as well as would be expected. The QoS Pass-Through mode enhancement can provide a significant performance improvement for high-bandwidth traffic flows through the 2800 switches, particularly when running traffic flows from 1000Base to either
100Base or 10Base connections.

This thread has been closed by the system or the community team. You may vote for any posts you find helpful, or search the Community for additional answers.

Slow network with HP ProCurve 28xx L3 Switch Family

Welcome to Apple Support Community
A forum where Apple customers help each other with their products. Get started with your Apple Account.