Thanks for the detailed response. On Feb 27, 9:56 am, Mike Christie <micha...@cs.wisc.edu> wrote: > If you do not use ifaces, then IO will be routed based on the route > table. So I think probably, IO would go through the same NIC on the > server. Is this what you are seeing?
I actually don't have this environment setup yet. We're making a change in our architecture and I want to make sure we're going to be setting up the environment in a the best (highest availability, best perf) way. >If you wanted to use dm-multipath > to round robin over both NICs on the linux server then you would use a > ifaces to bind each session to each NIC. Ok, good to know. > > With your setup you could also accomplish this by just putting one > target and one server in one subnet and the other group in a different > one. The linux networking routing will take care of everything for you. I was initially thinking this was the only way to do this but then noticed the ifaces option and wanted to make sure I fully understood it. But if we can get away with not having to create another subnet and change loads of servers to deal with it, it would be much easier. > Are the two switches connected to each other? Not sure, will have to check on this. > If they were and you are > using one subnet, you would have better redundancy. Above you have 2 > paths to the target, but if the switches are connected you have 4 paths. > > If the switches are not connected then if one switched dies you cannot > access that target portal, and then if you got really unlucky and the > other target portal died then you cannot access the target and you are > out of luck. If OTOH the switches were connected then you could still > access the other target portal with both server nics. > > > > > If not using ifaces and one of the switches or nics in the server go > > down, will the multipathing know to switch to the other nic? even > > though they are on the same subnet? > > The network layer should figure out there is another NIC that can be > used and just use it. A problem might be while we are switching nics IO > could time out and both paths could be down if they both ended up using > the same nic due to the routing table. So you would want to setup > dm-multipath with a higher no_path_retry, because when you switch over > you might also have to relogin to the target through the new nic. > > If you used ifaces then the failover should be smoother. The other path > would already be logged in, so dm-multipath could just restart the IO > right away. Thanks again, I appreciate the insight, this is what I needed to know. Romeo -- You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-is...@googlegroups.com. To unsubscribe from this group, send email to open-iscsi+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/open-iscsi?hl=en.
