Update
Happily I was able to record both my presentation and The Great Debate with Stu!
Although I’m a long-time Fibre Channel user, and certainly not the sort of person who denies the capabilities of that hallowed storage protocol, is it really necessary anymore? On Thursday, October 6, I’ll be taking up the challenge and debating Stu Miniman on this very topic as part of the Interop Conference and Expo in New York. Knowing my enthusiasm for iSCSI, organizers Mike Fratto and Howard Marks made me the rebel upstart, leaving Stu to defend the tired old lady!
Let me make it perfectly clear ahead of time: I’m actually something of a defender of Fibre Channel myself! I cringe whenever the “superiority” of Ethernet is raised, but this has more to do with the high standards of typical Fibre Channel SAN implementations than any real technical quality. In other words, it’s fairly straightforward to deploy a high-performance, reliable Fibre Channel SAN and fairly difficult to do the same with Ethernet-based protocols.
That said, I see the era of Fibre Channel coming to an end. 8 Gb Fibre Channel is fairly compelling as an upgrade from 4 Gb, but the jury is out when it comes to 16 Gb or even 32 Gb off in the future. By that time, Ethernet SAN will be widely available and supported, and will likely boast 40 Gb or 100 Gb performance to boot. Who is going to adopt a 32 Gb Fibre Channel SAN when Ethernet is three times faster?
But all of this is the future. Today, the battle lines are drawn between 8 Gb Fibre Channel and 1 Gb or 10 Gb iSCSI and NFS. This is the baseline for my Interop debate. I am not arguing about the future of SAN, or even iSCSI versus NFS. Rather, I am arguing that most businesses would be best served by implementing an iSCSI SAN rather than purchasing Fibre Channel today.
I intend to record and stream our Interop debate, perhaps even real-time. Watch this blog for a recording, and follow me on twitter to see if a live stream is available. See you on Thursday!
We have some data about adoption of iSCSI and FC storage courtesy of IDC in our Demartek iSCSI Deployment Guide 2011 http://www.demartek.com/Demartek_iSCSI_Deployment_Guide.html on our website. Free download.
FCoE is actually great. Mainly because it skips the infamous TCP overhead. The reason many people shy away from it is the lack of the “ecosystem”. Not many deploy it, too new, they are very few software vendors, no target for Windows, poor documentation, hard to find expertise. It’s like deploying Solaris for a web server. Sure it gets the job done, but its a pain the ass to find competent people to manage it.
The sheer widespreadness and momentum of iSCSI will probably eat up all potential FCoE customers in the mid-tier.
let’s not forget that iSCSI is just warming up, until 10GBE it sucked donkey balls in performance. It was absolutely impossible to run any serious traffic over puny 125MBs. Any “hot data” couldn’t be run over it, because of terrible latencies. You were basically limited to toyish scenarios that were not too demanding. Now you can run almost any application without even blinking. it is downright stellar. Just wait until RDMA becomes widespread!
I almost forgot this point. iSCSI is SCSI over TCP/IP. Notice I didn’t say anything about layer 2. It is layer 2 agnostic. You can, in theory, run it over FC or Infiniband, even over the Internet, which has a multitude of mediums. iSCSI’s fate is not tied to Ethernet’s fate.