iSCSI and LAGG

[jadams]

My understanding of Link Aggregation is not that 4 NICs in a LAGG = 4 x the bandwidth. A single connection cannot use it all at once. However 4 connections can all achive max throughput of a single NIC.

How does this work with iSCSI? I'm assuming it doesnt take advantage of the LAGG.

 

[patrickdk]

Your understanding of 802.3ad is lacking.

When you bond the 4 interfaces into an this, you have the option of splitting traffic over them how ever you want. This includes by mac address, by ip address, and sometimes by port number, round robin, ...

A single packet will use the distribution method you pick. If you pick per ip, then all packets going from/to the same ip will use the same nic, and therefor limited to a single nic. But going to 4 different ip's could use 4 different nics, and then 4x the limit of a single nic.

Normally there is no roundrobin option for this, as that can cause out of order packets, and tcp connection resets, but if this isn't an issue for you, 1 connection could go over all your nics and not have the limit. But this is normally not what you want.

Iscsi would have the same issue. Personally I don't believe in runing iscsi over top of this. Iscsi has built in support that makes this solution just extra overhead with no gains, and that is multipathing. Iscsi will distribute it's workload over every path.

If you iscsi stack doesn't support multipath, then you will have issues, and have to do this, and you will have to make sure your distribution over the links and your iscsi ip mappings work out to make this happen correctly.

 

[jadams]

Cant be too lacking. Perhaps I just wasnt clear enough but what you just said reinforced what I thought I knew.

Thanks, I think.