iSCSI - trying to break the language barrier
A plain language guide to the pitfalls of iSCSI
The iSCSI promise is that cheap and well-known Ethernet could be used to substitute for Fibre Channel, arguably simpler and less costly. Gartner DataQuest's VP and Research Director, Nick Allen, thinks so; "iSCSI promises to let users operate SAN, NAS, LAN and wide-area networks as a single integrated network." Nevertheless, iSCSI, in development since 2000, has not happened yet. Cisco has introduced iSCSI routers, the 5420 and 5428 for example, and other vendors have kit available or ready. But there are no signs of any mass adoption despite the technology being promising. Why is this? iSCSI's Interrupted Adoption
The basic technical problems are network speed, processing overhead, and customer inertia. To send SCSI commands across an IP network requires that a TCP/IP packet be wrapped around them at the sending end of the link and stripped off at the other end. An Ethernet network interface card (NIC) doesn't do this. If the server CPU does it then its performance on other applications is degraded. So-called TCP/IP offload engine (TOEs) cards are appearing, equivalent in function to FC HBAs. Adaptec and Emulex both have such products. Emulex' Alan Wallman, VP European Operations, says that Emulex is waiting on endorsement by a substantial OEM before making its latest product available. What are the OEMs waiting for? Part of the delay is due to network performance. If storage traffic is sent over the everyday LAN then its bandwidth could be eaten up. It requires gigabit Ethernet, even 10 gig Ethernet in heavy duty situations. There is also the issue of FC being known and iSCSI being new. Positiv's Marketing Director, Graeme Rowe, explains: "The thing to bear in mind is that none of this is here yet. People need to stick with the things that are mature and we won't be pushing anyone out to the bleeding edge just yet." There is also a cost aspect to iSCSI's adoption. A TOE card is more expensive than a basic Ethernet NIC, for example. Transtec's Wolfgang Bauer, head of its storage business unit, says: "At the moment first products are quite expensive, so the savings from less expensive cabling might be eaten up by the high costs for components." SAN Extension Niche
With iSCSI being unready for full-blown SAN use the vendors positioned it as a way of extending the reach of Fibre Channel SANs. Cisco provide FC ports on its iSCSI kit so that its iSCSI router can connect to existing FC SANs. Ironically this hardly simplifies an organisation's overall network infrastructure; it complicates it instead. Virtually all iSCSI vendors and storage integrators agree that iSCSI will complement FC SANs and not replace them. Brocade, the leading FC switch vendor, is one such. Paul Trowbridge, its European Marketing Director, says: "It is generally agreed, across the industry, that IP-based storage networks will be an extension to the existing Fibre Channel-based solutions." What all this means is that network performance worries, iSCSI component costs and inertia are holding back full-blown iSCSI adoption. Yet the technology is very appealing, not least to suppliers of Ethernet and IP products who foresee a whole new application area for their kit in the storage market. In forthcoming features we'll look at what suppliers are doing to overcome these worries, and examine the TOE cards, iSCSI routers and, finally, performance compared to Fibre Channel.