There were a few tweets about my last comments regarding the death of fibre channel that I would like to respond to. Scott Shimomura called me “clueless” and said I was anti-SAN. Well, to those individuals who agree with those comments: Let’s debate a few of the issues. Let’s start with facts:
1. Fibre channel sales have been dropping for a number of years and SAS sales are increasing.
2. Fibre channel’s biggest volume was when all enterprise disk drives had FC interfaces.
3. Scalable file systems today – whether they be NAS or parallel file systems – are connected via ethernet for the most part, with the latest being SAS backends.
I was a SAN guy in the 1990s and early 2000s. And in fact I tested some of the original Jaycor fibre channel cards (the company became JNI) on early MaxStrat Gen-5 (was purchased by Sun back in 1999) hardware with fibre channel interfaces back in 1996, if my memory serves me.
Back then we needed file systems that could take advantage of SAN performance and I worked with a number of file systems, helping them improve in performance to take advantage of this new technology.
Folks, nothing last forever: technologies come and go. I am not anti-SAN, anti-Fibre channel or anti-anything except maybe bad technology. The way I see it we have performance requirements that require scaling beyond what fibre channel can do. We have cost requirements that required this scaling to be cheaper than fibre channel. And we have fewer and fewer people that know about fibre channel technology.
So for you SAN people out there, tell me about an installation of a single namespace file system that has 10 PB in the namespace using a fibre channel SAN. I know of one and they struggle it, but I would like to hear about others. I know 100s using appliances.
To respond to me, please post in the Comments section below.
Labels: data storage, fibre channel
posted by: Henry Newman
As if fibre channel needed another arrow to deal with, this week LSI announced 12 Gb/sec SAS. The MSRP for SAS 9300-8e (single) is $450, SAS 9300-8i (single/kit) is $350/$430, SAS 9300-4i4e (single/kit) is $350/$430 and SAS 9300-4i (single/kit) is $245/$335. So for the best case, you get external 96 Gb/sec for $450, and in the worst case 48 Gb/sec for $430.
Previously, you could get 32 Gb/sec for $2,645 or $82.65 per Gb/sec. Now, for 96 Gb/sec you have a cost of $4.69 per Gb/sec, which is 17.63 times less than the old cost on a Gb/sec basis.
This is another nail in the coffin for fibre channel, as far as I am concerned. And it's a good reason to move your storage and compute close to each other to reduce the cost of buying your storage. I am sure that this will put increased pressure on fibre channel pricing.
Now, of course, we will have to wait a bit for storage vendors to have 12 Gb/sec SAS interfaces, but I do not think that will be too long. Maybe by the end of the year, I think.
So let’s think about this from a different point of view for storage interfaces. The SCSI interface was dominant for about ten years. Fibre channel has dominated for about twelve to fourteen years, and now we are entering a SAS-dominated world.
With SCSI to fibre channel, the change was not only for performance, but for distance. It surely was not for cost. The movement to SAS initially was purely a cost move. I think it was delayed because of the recession, but it was driven (unlike the fibre channel move) by the disk drive vendors. At least that is how I remember it. This might have some interesting long-term implications.
Labels: SAS, fibre channel, interface, Storage
posted by: Henry Newman