[rhelv6-list] Your opinion regarding NFS vs. iSCSI

Lamar Owen lowen at pari.edu
Mon Apr 30 17:47:24 UTC 2012


On Monday, April 30, 2012 10:59:11 AM Jussi Silvennoinen wrote:
> I really don't see any need or use for FCoE. I do like the idea of a 
> single communications channel (redundant) but FCoE is a poor excuse for a 
> solution towards that. iSCSI is much simpler protocol but suffers the same 
> single fabric shortcoming.

Ethernet is just as easy to deploy multifabric as FC is; you just design with multiple physical switches or sets of switches (just like FC).  I wouldn't deploy Ethernet for the SAN on the same physical switches and topology as my LAN anyway, assuming existing switch availability or new switch purchase budget for that task.  If the budget is tight, at least try to implement the backup or secondary SAN fabric on its own switches (I'd not put the primary on a dedicated fabric with the backup overlaying the LAN; the backup should always be more reliable than the primary).  If using full active/active fabrics, one on its own hardware and links and one overlaying the LAN might be serviceable.

Having said that, I am FC multifabric here.  But the same overall SAN architecture rules apply with Ethernet as with FC, really, for either iSCSI or FCoE.  You can even carry LAN traffic on FC (IPoFC).

But to do it because 'we already have 10GbE' seems to indicate that SAN overlay on the LAN is being contemplated, and this is where things can get hairy.  You really do want completely separate networks for SAN and LAN, if only to eliminate spanning-tree issues on the LAN creating issues for the SAN.

FC has other advantages over the Ethernet technologies, too, but it does cost quite a bit more.

IMO, YMMV, etc.




More information about the rhelv6-list mailing list