> On Tue, 2003-07-22 at 12:50, David Boyes wrote: > > Doesn't really solve the same problem, Adam. You're still > limited to the > > maximum bandwith of a single adapter; equal-cost routing > and the other layer > > 3 solutions just decide which one to use. > > I don't think so: equal-cost-multipath lets you split traffic to the > same destination across multiple links, and therefore (with help from > the router at the other end) aggregate your bandwidth across > them. It's > explicitly been added to the VM stack to do load balancing. > You can do > it with either OSPF or static routing on the VM side, and can > use up to > four equal-cost links.
However, all you're doing is statistically selecting between a number of fixed-speed pipes. The actual transmission of the packet occurs at the speed of the selected pipe. Your speed increase is only proportional to a percentage of the number of links in the bundle due to additional transmission opportunities. Packet arrival (and thus path selection) is self-similar, not Poisson distributed. Different problem. I wonder if the multilink PPPoE code would work over the s/390 Ethernet implementation. That might do the trick.
