> On Apr 5, 2014, at 4:26 PM, Skylar Thompson <[email protected]> wrote > > Agreed. FC also has the benefit of preventing over-subscription of a > channel by design - you might saturate a link but you won't experience > dropped frames. FC channel aggregation blows the socks off LACP as well; > the usage on each individual channel is far more uniform than with LACP > since an incoming are sent to the first channel with available credits > rather than using some arbitrary hashing algorithm. >
Modern Ethernet switching fabric (leaf and spine) designs provide the same design features. You can get zero over subscription, they don't require lacp (or the hashing limitations), don't require spanning tree (meaning all links are active, so you can scale with an N+1 design), and scale to massive throughputs. With 10gig nics to the storage devices and 40gig links between leaf and spine switches you can get massive east/west bandwidth, which is becoming critical with modern workloads (FCoE, Hadoop, VMware VSAN, etc) I just got back from Interop and I saw multiple presentations on this style of design and a case study of a clustered IP storage grid achieving greater then 1Tbps of IO across the cluster. (Aside: I was really surprised to not see any other lopsa members there! (Except one). Seems like there should be more overlap between these crowds, but that's a topic for another thread) -David > Skylar > _______________________________________________ > Discuss mailing list > [email protected] > https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss > This list provided by the League of Professional System Administrators > http://lopsa.org/ _______________________________________________ Discuss mailing list [email protected] https://lists.lopsa.org/cgi-bin/mailman/listinfo/discuss This list provided by the League of Professional System Administrators http://lopsa.org/
