Smart SAN Switches: Not This Year

NEW YORK -- A roundtable panel of executives from storage processor startups at the RBC Capital Markets conference here today agreed that intelligent SAN switches won't arrive in any significant way until next year at the earliest.

The panelists included representatives of Aarohi Inc., Aristos Logic, Astute Networks Inc., iVivity Inc., and Silverback Systems Inc. All of these startups are in various stages of developing silicon -- for either Fibre Channel, Ethernet, or both -- to speed up certain SAN operations.

Interest in intelligent SAN switches -- which are supposed to offer multiple high-level data management features in the network fabric -- heated up at the end of last year, as Brocade Communications Systems Inc. (Nasdaq: BRCD) announced it would buy Rhapsody Networks, and Sun Microsystems Inc. (Nasdaq: SUNW) bought Pirus Networks.

Others developing such smart storage switches include Cisco Systems Inc. (Nasdaq: CSCO), Candera Inc. (formerly Confluence Networks), Maranti Networks, and MaXXan Systems Inc.

But while there will be some "token" deployments of these switches in the coming months, the category won't take off any sooner than 2004, say the storage processor startups, several of which are suppliers to those smart SAN switch developers. One of the stumbling blocks is that the new systems need to be able to integrate with the existing SAN infrastructures -- without causing any disruptions.It's also a more complex problem than some first realized, noted Vipul Mehta, co-founder and VP of business development at Aristos Logic.

"When we started talking to intelligent switch developers, the dream was to have them talk directly to JBODs [just a bunch of disks]," he said. "Over time, realities have tempered that."

In particular, such features as failover and RAID 5 "are very intense and complex functions," Mehta said. "To get that intelligence into the switch will take some time."

Sanjay Sehgal, COO of iVivity, said all elements of the infrastructure need to be able to take advantage of the SAN's intelligent features for there to be any real benefit to end users: "Intelligence needs to reside not just in the fabric but at the endpoints."

The panelists also concurred that the introduction of IP SANs based on iSCSI, the protocol for sending block-level data over IP networks, has been pushed out toward the end of this year.Roger Moyers, founder and COO of Astute Networks, said there's a lot of work happening on the initiator side but "there hasn't been a lot done on the target side."

SilverBack CEO Danny Biran, whose company is focused exclusively on providing chips to accelerate iSCSI traffic, agreed the IP storage technology has been delayed. But he said Microsoft Corp. (Nasdaq: MSFT) has released a beta version of its iSCSI stack for Windows .NET Server 2003. Microsoft had been expecting to include native iSCSI support in the initial version of .NET, slated to ship in April 2003, but now plans to release it as a follow-on update (see Microsoft Won't Ship iSCSI in .NET).

"It helped that Microsoft articulated iSCSI in .NET very clearly," said Biran.

As for 10-Gbit/s networking technologies, the executives said, the cost of the optical interfaces will have to fall before either 10-Gigabit Ethernet or 10-Gig Fibre Channel become feasible as a host or storage connection (see our report, 10-Gig SANs).

"At this point we see the need for aggregation at 10 Gig because that's where you need high performance," said Ameesh Divatia, CEO of Aarohi. As an interconnect, "10-Gig Fibre Channel is interesting, but until the optic prices come down, it's not going to happen."Biran added that servers and storage systems also need a higher-performance bus to be able to do anything useful with 10-Gbit/s networks: "There is no end node that can generate or consume 10 Gig today."

Todd Spangler, US Editor, Byte and Switch
http://www.byteandswitch.com