We received the following comment from a VAR based in Florida on our piece covering the newly proposed Fibre Channel over Ethernet (FCoE) standard:
The concept of FC over Ethernet has very limited value. According to this article, the FCoE consortium is targeting this at low-to-mid range servers, over 10 GbE, and as a convergence technology. While 10 GbE makes sense from the storage array to the switch, it makes little to no sense from the server to the switch for two reasons: first, low to mid range servers — where this is targeted, don’t have the I/O requirements to saturate 1 GbE much less 10 GbE, and their PCI busses would not be able to handle anywhere near 10GbE throughput (do the math), and second, the reason that FCP exists is not because of throughput, but deterministic response time, which is guaranteed by the FC protocol wheras the Ethernet protocol becomes more non-deterministic with high load. This lack of deterministic response time will not be fixed with FCoE.
For these reasons, FCoE to the server does not make sense on the low/mid (don’t need the throughput and couldn’t handle it anyway) or the high end (Ethernet lacks the predictable response time of FCP). So back to the question of FCoE over 10 GbE from the storage array to the switch — if the storage arrays were 10GbE capable, why not just use iSCSI, which is already supported and in wide use in the enterprise despite what some manufacturers’ marketing and media reports say? My personal opinion is that this is an effort on the part of manufacturers who are behind in iSCSI to change the game in an effort to compete, and provides little to no value to consumers.
We have the feeling this could develop into an interesting discussion in the industry over the next year or so as FC and iSCSI, originally at odds in the market, have increasingly been combined in tiered storage environments and multiprotocol systems. Still, combining the two protocols–especially in the same data stream–could become a thorny issue.
What do you think?