SAS smacks down FC
Posted in Storage Interconnects & RAID, Advisor - Tom Treadway by Tom TreadwayI’m not sure how this will go down, but I’m about to disagree slightly with my homeboy blogger JoeD regarding SAS’s effect on other storage options. It’s not that he’s necessarily wrong, but I’m going to present a different answer that comes from a different point-of-view.
First, Joe said that SAS is mostly a replacement for parallel SCSI. Of course he is 100% correct.
But it’s important to realize that SAS also replaces ATA. Before you jump up and say, “But SATA drives are so much cheaper!”, let me remind you that SAS isn’t just a drive interface. It’s also an adapter interface. It’s the interface on your motherboard or on your RAID controller. So if you were going to move from ATA to some new interface, you might not select SATA. That would lock you into SATA drives. But selecting SAS will allow you to use either SAS or SATA drives, as I’ve mentioned before. So while most people won’t be able to justify more expensive SAS drives, they will at least have the option to use SAS drives.
Next, let’s look at FC. It has two purposes – as a fabric and as a drive interface. I’m not going to touch the fabric purpose. Unless you squint and hold your mouth just right, SAS has no real play there. But as a drive interface, let’s look at what FC drives are being attached to.
If you own a FC-to-FC MFA (Mid Fabric Appliance) or RAID controller and you like to hang lots and lots of FC JBODs behind it, or put those JBODs really far away, then you obviously have to keep buying FC drives. But if you buy a FC-to-FC MFA or RAID controller, then you should be asking your vendor to make a FC-to-SAS version. SAS has all the reliability and performance of FC, and you also have the option of using cheaper and larger SATA drives.
You may be using FC drives simply because you have an external JBOD with drives outside your server, and you might not like nasty parallel SCSI cables. This is a typical “direct attach” model. Or you may have multiple hosts connected to that storage for failover purposes, and of course parallel SCSI cables is a nightmare in this configuration. This is obviously the ideal usage model for SAS. Anybody using FC in such small systems would be crazy to not switch to SAS. And like I’ve mentioned before, you will have the option of using cheaper SATA drives.
Lastly, believe it or not, SAS will encroach on some number of iSCSI sales. Like the FC SAN example, I’m going to ignore all the larger, more complicated fabrics where iSCSI makes sense. But there are companies that sell iSCSI as a direct attach solution. And originally this really didn’t seem that stupid of an idea - it has better cabling than parallel SCSI, it’s not too hard to manage, etc. But now with SAS, that idea actually is kind of stupid. SAS has much higher performance, is much cheaper, and, again, it supports SATA.
Let’s summarize:
-
SAS replaces parallel SCSI.
SAS host adapters replace ATA.
SAS supports SATA drives.
SAS replaces direct-attach FC.
SAS replaces FC in external RAID systems.
SAS replaces direct-attach iSCSI.
FC is used in high-end SANs.
iSCSI is used in low-end SANs.
Hmm, I wonder what happens when 10Gb iSCSI is released? Where does FC go? Well, I don’t pity the FC vendors. They dug their own grave by selling two port FC adapters for $1000 and FC switches for $500 per port. SAS and iSCSI may have never happened if they hadn’t been so greedy.
But I got off topic. SAS will take major market share from FC, some from ATA/SATA, and even share from iSCSI.
TT
October 29th, 2005 at 12:29 pm
TT,
As much as I’d love to disagree with you… I don’t! Oh well - we’ll fight another day. If you check out one of my see previous blogs you’ll see I actually mention the SAS interconnect as well. I was speaking strictly to SAS drives relative to other drives.
You’re right on the SAS interconnect front. In fact, I’ll go as far as to say that the interconnect technology and SAS switches, when they turn up, will be more disruptive to the storage industry then the drives themselves.
No worries though - I’m sure we’ll disagree soon!
Blog ya later!
Joe
November 2nd, 2005 at 12:12 pm
Sorry about that, Joe. My template for new posts starts out with “Joe, you couldn’t be more wrong”. I guess I forgot to delete it this time.
TT
December 2nd, 2005 at 8:14 pm
Tom,
About SAS as a fabric and an interconnect. 2 or more points I’d like to add.
1) FC as a drive interconnect has shared bandwidth between all the devices so any amount of sequential disk I/O (with Maxtor Atas 15K II SAS at the WORLD RECORD of 98MB/s) is going to saturate the FC shared bus. SAS on the otherhand is point to point at 3ooMB/s (full duplex unlike our FC and SATA hald -duplex buddies)
2) SAS supports Wide Ports, which means that bandwidth is scalable and almost unbounded. For more bandwidth just add more connections (up to 128 of them)! Doubling the number of connections from the server to the storage cabinet or expander doubles the bandwidth between them and so the connection back to the host, or from expander to expander will never become a bottleneck. Yes, FC has port aggregation but its done back at the driver level. SAS implements wide lanes at the port layer and is transparent to the driver level.
3) SAS supports multi-initiator and so many servers can be clustered to provide performance and/or server fault tolerance. If you look at the majority of clusters today, they are FC based - WHY? …because there’s NO alternative - until SAS. For an “in-the-rack” cluster with 2-5 server or server blades, 1 or 2 storage cabinets with 1-2 fanout expanders the cost effectiveness of SAS over FC is compelling.
Joe
December 3rd, 2005 at 6:07 am
Joe,
Yes, yes, and yes!
Regarding FC, it’s of course moving to 4Gb/s, but it just seems too late. As you point out, it would take only 4 drives to saturate a 4Gb FC bus.
Regarding wide ports on SAS, yeah, I just love the idea of 12Gb/s going to 24Gb/s in the next few years. And I suppose SAS could go wider than 4X very easily. Smokin’!
Regarding SAS clustering, what are you hearing about Microsoft clustering, and it no longer being allowed with PCI RAID? (See previous post here.) Sounds like we need SAS-to-SAS RAID controllers. Do you agree?
Damn good to hear from you, buddy. Keep in touch.
TT