Design considerations for 10-Gbit iSCSI

The biggest challenge to Fibre Channel may come from efforts by a range of players to bring storage-area networks (SANs) to the low-cost, high-volume economics of Ethernet. But this is no engineering slam-dunk.

Running the ubiquitous SCSI command mechanisms over Ethernet at anything near wire speeds for 1G and 10G connections requires specialized hardware and a lot of design savvy, notes David Black in his article. Black is a senior technologist at array maker EMC Corp. and a co-chair of the Internet Engineering Task Force's Internet Protocol storage group, which defined the iSCSI protocol for Ethernet.

Indeed, two online contributors burrow even deeper into the issues that Black raises. Shridhar Mukund, director of engineering at Adaptec Inc., examines the implications for programmers of the move to multicore ASICs for the TCP offload engines (TOEs) used to accelerate SANs based on iSCSI. And Uri Elzur, the strategic marketing manager at Broadcom Corp., describes the RDMA Consortium's effort to solve the problem of TCP framing for TOEs.

Steven J. Sears, a member of the technical staff at Network Appliance Inc., contends that Infiniband provides a viable alternative route to 10G storage, thanks to its low latency and high bandwidth-despite the fact that some Infiniband companies have backed away from product plans or folded altogether.

While these 10G networks drive forward, two classes of lower-end SAN products are emerging. Online, Maxtor Corp. interface engineers Duncan Penman and Martin Parry look at the rise of 1.5-Gbit/second serial ATA systems and examine some of the signal-integrity and power-distribution problems they face. And LSI Logic Corp.'s Microsoft alliance manager, Calvin Chen, writes about how serial attached SCSI and multipath I/O can be combined to deliver storage systems that could be less complex and more affordable than those based on Fibre Channel.

EET