Techworld discussed SAN routing and the spread of iSCSI storage with Tom Clark. Clark is director of SAN technology for McDATA, having joined with the Nishan acquisition, where he was director of technology. Nishan was the iSCSI/IP storage pioneer. What does Tom Clark think of the relative strengths of Fibre Channel (FC) and iSCSI?

FC at the top of the pyramid
Fibre Channel (FC) is well-established in the data centre, the top of the business pyramid, but is prevented from spreading by three factors, Clark says; “GigE followed on from FC. It needed switching to catch up, and then a protocol to move block data over it – iSCSI. Now it has spread quite rapidly; from first product to real customer deployment in three years.”

IP SANs have been shown to function across 1,000 mile plus distances, ones too long for the cost of FC to be justifiable.

“IP is a complement to FC. FC can’t solve long distance problem; it’s only suitable for local or metro distances. That’s a driver to IP. FC can’t address the mid-market because of HBA cost; around $1500. Putting them on a sub-$5000 server is out of joint. This is also a driver to IP SAN technology.”

“FC is a layer 2 architecture and designed specifically so, because of a minimal protocol overhead. As you build out you get stability problems. Fabric reconfigurations propagate across the SAN. A state change in one port causes a storm of broadcast chatter. SANs can have 900 ports or more. You can’t afford to have this instability.”

This is the third driver to iSCSI technology, because it is a layer three protocol and isolates state changes and other faults to SAN sections. ISCSI kit is also much more affordable than FC and opens up a new IP SAN market, one for the mid-tier and one for the long-distance linking of FC SANs. Clarke says, “Technology becomes more affordable to the mid-tier and to enterprise branches as well, for the Dell space and CLARiiON and HDS 900 mid-tier storage arrays.”

There’s also the replication and disaster recovery (DR) angle. Mid-tier companies are just as concerned about this, Clark thinks; “We take everything for granted but it’s really so precarious. We turn on the switch and the light comes on. We take it for granted, until there’s a power outage. SME guys have he same problem as larger enterprises. They want availability and safeguards, just like enterprise shops. (With iSCSI) they can now reasonably afford to have a DR set-up. The distance capability is there. Now if there is a power outage they don’t lose customers.”

Clark says it’s not an issue of losing data so much as losing timely access to data. If you can’t serve your customers somebody else will. The affordability of IP SANs alters enterprise behaviour also. “Enterprises can SAN-up branches, running block I/O across an ordinary NIC, or a TCP/IP offload NIC or pay a little more for full gigabit speed on Alachritech cards. There is a wide variety of cards.”

This choice is not generally available with FC, even though less expensive FC gear is available. But, Clark says, “The market doesn’t want cheap FC if it comes with enterprise complexity. IP storage is simpler to deploy and less scary for the mid-tier DAS buyers. (They know) buying more DAS compounds the storage admin burden. ISCSI host connect is just SCSI and Ethernet; it’s understandable.”

IP provides more storage access options than FC. “We can accommodate wireless links. We can use a wireless bridge to link sites through. Once you move blocks to IP then the world opens up in terms of customer options for communications. The SAN routing capability lets us align storage to DR needs . (When disaster happens) you can redirect apps from the broken SAN to a DR SAN thousands of miles distant. The DR SAN is kept up with replication.”

Brickbats For Brocade and CIsco
Clark sees the development of FC directors getting iSCSI blades. McDATA is in a prime position to capitalise on this compared to its competitors. In his personal opinion, “Brocade has essentially failed to go upstairs with directors and has failed to bring out mid-tier stuff with the Rhapsody acquisition. Cisco is new to this space and has a long road to establish itself. The storage guys in data centres are completely different from the ones managing the Cisco Catalyst routers and switches. It’s not easy. Storage data is far too critical to play with…. What’s key are your references and experience. Brocade does okay on the FC side. They don’t have good solutions for DR. They are late to market. They stalled so long on the Rhapsody acquisition. They can’t do what we can do today; neither can Cisco.”

Concerning Cisco, Clark personally thinks, “The 5420 and 5428 were pathetic performers.“ It is as if Cisco has essentially stopped marketing them. “They’re banking on newer iSCSI blades for the MDS line. But that’s an expensive blade. It’s not really competitive. Today we have a competitive advantage.” McDATA iSCSI products have 256MB buffers in the long haul ports, compared to competitors’ 64KB ones. “It’s obscene overkill. We can use the fattest (network) pipes available, coast to coast. Brocade ports are not designed for the long haul. Cisco has not shown they can drive as many iSCSI sessions as we can.”

“Cisco doesn’t have SAN routing. It’s all layer 2. This undermines the high availability of data. Notification broadcasts go all the way across. I know there are instances where this causes problems.” But these instances are not publicly citable. He sees iSCSI as being, “analogous to the LAN market, where bridging was the first solution. Now routing is preferred. It isolates faults.”

IP Storage To Be The Rule
And the future? Clark is clear; “ISCSI SANs are a given. In five years’ time you’ll have DAS for boot and local files on a server. For the rest it’s shared storage. There’s so much to gain. SANs will rule just like shared printers today. Why will you want to have server-dedicated storage? Why would you want everyone to have their own printer today?”