ProCruve LACP to Linux

[jamesr-qm]

Hi all,
I am wondering if you can help me with a problem, we have a ProCrurve Switch (ProCurve J4899B Switch 2650) Running firmware H.08.83 and am trying to get a LACP Port Trunk working between this and a Linux server. At the moment I have the trunk up and running but I cannot however see an increased performance, a single port seems to be preferred without the second port being used even when the first is maxed out to its full speed.

The Linux side of the trunk is running in LACP mode and there seem to be no errors reported by each side. I am wondering if there is some option that may have been missed to utilize both links fully rather than a fail-over system.

If I can provide any more information please let me know what you require.

[Matt Hobbs]

Trunking works based on Source and Destination mac-address pairs. So to see the benefit, you need to be using multiple hosts. If there are only 2 hosts, only one of the links will be used. The pairs are randomly selected, eventually with enough hosts you should see fairly even traffic distribution.

[rick jones]

On the Linux side you would want to select a packet scheduling mode that included the TCP/UDP port numbers if you wanted to see an increase in performance between two hosts - and you would have to run multiple, concurrent connections between those hosts.

I keep meaning to look to see if any switches have similar capabilities but haven't gotten around to it :(

If you need a single connection/flow/stream to go faster than trunking isn't the way to go - you need to upgrade to a faster link - eg from 100BT to 1G or from 1G to 10G.

[jamesr-qm]

Unfortunately I am using single flows, I have found that leaving the switch with the trunk define and linux using the mode=0 (balance-rr) instead of mode=4 (LACP) is providing me with a 200Mb single flow link. This does seem to risk getting Out Of Order frames more often however. It does seem to be working for me at the present time.

[rick jones]

Just remember that no reputable trunking software offers RR :) (Ok, I only know that HP-UX Auto Port Aggregation doesn't but it sounded better that way :)

You are definitely scating on thin ice, and I suspect that if you were to do a bit of netperf benchmarking you would find that the service demand for the RR stuff at 200Mbit/s is rather higher per KB transferred than a 100 Mbit/s stream through some other scheduling algorithm.

There are some reasonably inexpensive GbE switches in the ProCurve line these days, and GbE NICs aren't all that expensive now, so do consider an upgrade to gigabit ASAP

[jamesr-qm]

Unfortunately its not a question of cost of the switch its a case that there is no easy access to get gbit links to this room at the moment as its fairly out of the way when it comes to location. It is being looked into by the relevant people but that will take time, this was a stop-gap measure. I will continue to monitor the performance of this link but everything seems fairly OK for my uses but I wouldn't recommend it to others lets say.