This is the output that you see with your test running? Only one hash bucket (143) has significant load, so there's no effective balancing.
It might be useful to see the output of "ovs-dpctl dump-flows" while your test is running, to see how many active flows there are and how much traffic is running through each of them. Thanks, Ben. On Tue, Mar 06, 2012 at 08:53:03AM +0100, Kristoffer Egefelt wrote: > *ovs-appctl bond/show bond0 on both sending and receiving server:* > * > * > *root@node0301:*/home/Kristoffer# ovs-appctl bond/show bond0 > bond_mode: balance-tcp > bond-hash-algorithm: balance-tcp > bond-hash-basis: 0 > updelay: 0 ms > downdelay: 0 ms > next rebalance: 9692 ms > lacp_negotiated: true > > slave eth1: enabled > may_enable: true > hash 126: 1 kB load > hash 200: 1 kB load > hash 245: 3 kB load > > slave eth0: enabled > active slave > may_enable: true > hash 143: 100182 kB load > > *root@node0302*:/home/Kristoffer# ovs-appctl bond/show bond0 > bond_mode: balance-tcp > bond-hash-algorithm: balance-tcp > bond-hash-basis: 0 > updelay: 0 ms > downdelay: 0 ms > next rebalance: 4974 ms > lacp_negotiated: true > > slave eth1: enabled > may_enable: true > hash 28: 0 kB load > hash 197: 1 kB load > hash 216: 0 kB load > hash 218: 6 kB load > hash 227: 0 kB load > > slave eth0: enabled > active slave > may_enable: true > hash 58: 0 kB load > hash 90: 0 kB load > hash 161: 0 kB load > hash 166: 0 kB load > hash 178: 0 kB load > > On Mon, Mar 5, 2012 at 6:06 PM, Ben Pfaff <b...@nicira.com> wrote: > > > Yes, I'd expect that to work (assuming that "ab" means "application"). > > > > What does "ovs-appctl bond/show <bondname>" print? > > > > On Mon, Mar 05, 2012 at 09:39:59AM +0100, Kristoffer Egefelt wrote: > > > I see. > > > Then shouldn't a simple ab with 2 or more sessions balance the traffic > > > across 2 uplinks, using source port hashing? > > > It looks like this is not the case when I'm testing on v1.3.0... > > > > > > Thanks > > > > > > On Fri, Mar 2, 2012 at 10:36 PM, Ben Pfaff <b...@nicira.com> wrote: > > > > > > > Looking at the documentation, it says, "such as destination MAC > > > > address, IP address, and TCP port." The phrase "such as" introduces a > > > > list of examples and is not meant to be exclusive. > > > > > > > > On Fri, Mar 02, 2012 at 07:52:08AM -0800, Ben Pfaff wrote: > > > > > Does it say destination IP/port only? At any rate, balance-tcp > > balances > > > > > on L2 through L4 packet headers, including source and destination > > ports. > > > > > > > > > > On Fri, Mar 02, 2012 at 11:10:18AM +0100, Kristoffer Egefelt wrote: > > > > > > Ahh, yes - balance-tcp. > > > > > > I'ts balancing using destination ip/port - is it possible to do > > source > > > > > > ip/port balancing? > > > > > > > > > > > > Thanks > > > > > > > > > > > > On Thu, Mar 1, 2012 at 4:34 PM, Ben Pfaff <b...@nicira.com> wrote: > > > > > > > > > > > > > On Thu, Mar 01, 2012 at 03:55:41PM +0100, Kristoffer Egefelt > > wrote: > > > > > > > > Outbound traffic seems to depend on SLB bonding (where a source > > > > switch > > > > > > > port > > > > > > > > is needed for every session to utilize a link). > > > > > > > > Is it somehow possible to use policies like the linux kernel > > > > bonding > > > > > > > driver > > > > > > > > is offering? (layer2+3 / layer3+4) ? > > > > > > > > > > > > > > Yes. See ovs-vswitchd.conf.db(5) for configuration details. > > > > > > > > > > > > > _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
