On Thu, Apr 14, 2011 at 9:11 AM, Max <[email protected]> wrote: > > Terry, > I do not confess to be anywhere near an expert, but layer2 is, as far > as I know based on a hash of the MAC addresses > > http://www.linuxfoundation.org/collaborate/workgroups/networking/bonding > > So how well things are balanced depends on the how many machines are active > to/from a server, what their MAC addresses are, and how these MAC addresses > hash. If you only have a few hosts active then you may end up with only > one interface being used - things will only really even out if you have > 'enough hosts' all doing approximately the same amount of work against the > server, all at the same time. > > [Also be aware that using 3 interfaces may not work well with some > algorithms/ > routers ... I am told that for Cisco's it is best to use 1, 2, 4, or 8 > interfaces > from the server, otherwise you may never get a balance coming into you > server] > >> ... >> Transmit Hash Policy: layer2 (0) >> ... >> >> I think the reason why I see one interface dominating RX and another >> dominating TX is due to the xmit_hash_policy but there are three hosts >> that use this particular server for network traffic. That's 3 >> different physical mac addresses. The layer2 algorithm should be fine >> in that situation I would think. What am I missing? Would I just be >> better off with balance-rr? >> > Max > _______________________________________________
Thanks for the replies. Ok, good. At least I have something I can try. From what I am seeing, there are two hash algorithms, are there more?: layer2 layer3+4 I will try layer3+4 and report back what I see. I didn't think about the odd number of interfaces. The switching stack is brocade. Since I can't move to 10G immediately, I am forced to try this other xmit policy. If that doesn't work, I will scale back down to two interfaces on each host and add more hosts. The application I am using is truly load balanced active/active/active so it will use all hosts pretty evenly. I may open a case with brocade too to see what they have to say about it. _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
