NAS, SANs, and SSPs

The following Q&A-with storage consultant and book author Jon William Toigo-was excerpted from an online chat session hosted by www.searchstorage.com, a Web-based search engine site for storage-related articles and information. Toigo is the author of and both published by Prentice Hall PTR.

What do you think will be the hottest storage issues of 2001?

Virtualization and DAFS (Direct Access File System). Virtualization is required to deliver the value proposition of SANs (storage area networks), and DAFS is the ultimate SAN spoiler.

How is the progress of DAFS going?

DAFS is rolling out smoothly. Keep an eye on www.dafscollaborative.org for more details.

Is EMC's attempt to get a Virtual Interface (VI) spec for NFS/CIFS adopted by SNIA (Storage Networking Industry Association) related to DAFS, or an attempt to split the whole DAFS initiative?

EMC appears to be endeavoring to start a working group at SNIA to develop an NFS over VI "standard." The technology seems similar in concept to DAFS, in that it would provide a mechanism to bypass protocol stack overhead and buffering issues associated with network file systems employed over IP networks. In so doing, it would presumably facilitate high-speed, high-performance storage access. DAFS, unlike the EMC idea, is not a port of NFS to VI but is a completely new protocol that will enable high-speed block-level storage access for NAS devices (and NAS/SAN hybrids) using far fewer instructions than are used in an NFS data access. If I were suspicious of EMC, I might suspect that it was trying to derail the DAFS effort or at least gain some bragging rights regarding its work on a "comparable" solution. In fact, its effort confirms the value of DAFS. Interestingly, CrosStor Software, which was recently acquired by EMC, was a long-time player in the DAFS development effort.

What do you think will be the biggest issues facing organizations implementing-or thinking about implementing-a SAN?

The key issue will be to virtualize heterogeneous storage devices while maintaining the ability to conduct restore operations and other large-scale data movements efficiently.

Will 2001 see the emergence of truly heterogeneous SANs?

You will see increased work toward realizing heterogeneous SANs, which I define as a storage infrastructure with device agnosticism-both servers and storage devices. However, I do not believe that all of the issues of a SAN operating system or of SAN transports (Fibre Channel or iSCSI) will be completely resolved until next year or later. There is still a great deal of work to be done on transports to guarantee interoperability, manageability, and scalability. However, those issues pale in comparison to the Herculean problems confronting SAN OS developers.

When will there be one set of standards established, relative to managing the various devices? Everyone today is doing his own thing, and I don't see how this will aid in establishing heterogeneous SANs.

A baseline framework needs to be established for networked storage infrastructure so that products from different vendors can compete on a level playing field and on a price/function/performance basis. Getting the plumbing standards together is a good thing. iSCSI and 2Gbps Fibre Channel will both hit the streets this year. Then comes the myriad of issues involved in standardizing the SAN operating system-next to which the challenges for building SAN interconnects are child's play. Once some standards are in place, however, end users will begin to reap the benefits of interoperable solutions.

As storage over IP becomes more practical, will it replace Fibre Channel, or will each have a place in SANs?

IP-based (or iSCSI) SANs hold great promise for realizing the objective of open, highly manageable, secure SANs. I believe that IP SANs will get the lead (over Fibre Channel) with powerhouses like Cisco behind it. Fibre Channel may continue to find adherents as the protocol suite is improved over time.

Will IP-based network storage solutions overtake Fibre Channel SANs in popularity?

Yes, I believe they will. IP SANs won't require new plumbing. They'll leverage existing IT staff skills and knowledge, be manageable via standard IP network management tools, and won't confront the interoperability issues that have plagued Fibre Channels SANs. I concede that FC and even InfiniBand offer some superior capabilities over IP-based SANs as presently conceived. I also believed that Token Ring was a superior technology to Ethernet and that SSA was a superior interconnect over parallel SCSI. But the best technology doesn't necessarily win. With backers like Cisco and IBM, iSCSI is a sure winner. FC will be supported as a piggyback protocol over IP. Cisco's work with Brocade ensures this, and it is required to link isolated islands of FC SANs.

SAN implementations are currently limited by the plethora of interoperability issues with respect to host bus adapters (HBAs), storage devices, and switches. There has been a lot of focus on switch interoperability. Do you see standards in HBA implementations, particularly with regard to multiprotocol and VI implementations?

Vendors have no stake in making their HBAs cross-product-compliant until there is a market-based reason for doing so. When a dominant topology and set of standards for SANs arise (perhaps from Cisco with its iSCSI solutions), the result will likely be fast, widespread, and industry-wide adoption by HBA makers. Then the customers will police the vendors and purchase only those products with demonstrated interoperability in the field.

When will storage vendors start manufacturing products that are truly interoperable? If it does happen, which vendor will be the first?

I think you are already seeing earnest efforts among SAN switch vendors in the FC world and among SAN router/gateway vendors in the iSCSI world toward producing interoperable products. Companies like Cisco and NetConvergence and possibly Vicom may already be ahead of the curve in delivering products designed to create a truly open IP-based SAN. I believe that the first true SANs-non-proprietary, open, capable of offering dynamically scalable virtual volumes to servers, highly manageable and secure-will come from smaller vendors rather than larger ones. Big storage and server vendors simply have too much to lose from an open solution. You may well see a spoiler technology in the form of the DAFS that delivers the benefits expected of SANs-but from a NAS/SAN hybrid solution-before year-end.

Do you see SCSI over IP as being a serious product, rather than hype, in 2001?

I am concerned about the hype, but I think that iSCSI is well on its way to becoming a reality. iSCSI is growing out of the IETF (Internet Engineering Task Force), and from that crucible will come something tangible. This year? Maybe.

Will true standards emerge this year? If so, what will they be?

The Fibre Channel Industry Association (FCIA) and ANSI T-11 committee that are working on Fibre Channel improvements are a collective juggernaut. I expect they will rectify many of the problems with FC interoperability and will articulate a faster FC-based SAN interconnect by mid-year. The iSCSI development effort is tougher to read. I think that the IETF IPS working group will have an RFC ready for implementation over Gigabit Ethernet by the summer and that vendors will be showing iSCSI SANs at Networld+Interop and other shows in the fall. A number of companies are already rolling out products based on the original proposed spec that will need to be field-upgraded once iSCSI arrives. Once it's a stable standard, I expect the IP world to beat out FC by virtue of the sheer market presence of IP networks. That's why FC HBA vendors have been hedging their bets over the past few months.

Please comment on Fibre Channel versus Gigabit Ethernet technologies for SANs. Will they co-exist?

I am sure they will. They already do. The alliance between Brocade and Cisco testifies to the need for a GbE interconnect to unite "SAN islands" created by early adopters of FC-based SANs. GbE will have an edge over FC because the SANs can be configured using existing infrastructure and skills.

If SCSI over IP is going to rule, how will they address the performance limitations of IP as a storage transport, or will they ignore it and just make the whole pipe faster with 10GbE?

This is an interesting question because it reflects some fiction and some truths that are appearing in equal doses in many trade press articles and at numerous conferences. TCP, not IP, is the potential limiter of iSCSI efficiency. There is certainly work to be done to make SCSI work effectively with TCP, which is being done by the IETF IPS Working Group. Yes, bigger pipes will help offset some overhead issues. Ultimately, however, IP SANs will compete in every respect with FC SANs-and will exceed the capabilities of FC with respect to networking, management, and security.

How do you see the HBA market evolving?

I am betting that the current leaders in Fibre Channel HBAs will get on board with iSCSI-based products as the technology matures. Adaptec may be first out of the blocks with an HBA based on the original iSCSI spec that also supports Adaptec's proprietary technology.

Is there such a thing as a "wide area SAN?"

I've heard that term as well as "SWAN" used. I think of it as re-packaging an old concept that dates back to channel extension. Basically, a wide area SAN consists of an interconnection between local and remote SAN components via a WAN. The question is: Will the SAN/ WAN interconnect tunnel the SAN protocol through the WAN, or will it run SCSI (or Fibre Channel) as an application, natively, across the WAN? If you don't understand the difference, look at the war that is breaking out in the IETF IPS Working Group over a suggestion that FCIP and iFCP be merged into a single standard. One involves tunneling; the other, operation of the protocol as an application.

Will this be the year that storage service providers take off or hit rock bottom?

If corporate America hits a recession, that's usually accompanied by a spike in outsourcing activity. Management says outsourcing-of which SSPs are one part-enables the company to focus on core competencies and to get out of the expensive IT business, but this is usually a rationalization for cost-cutting and downsizing. SSPs will potentially profit from lean economic times. On the other hand, high-speed networks have not rolled out at the speed anticipated only a year ago. The result is that SSP services are not what vendors originally touted. You won't see a lot of remote storage leasing arrangements because the bandwidth doesn't exist in many areas. You will see more collocation and insourcing/management arrangements, especially as the high cost to manage storage comes back to haunt everyone who has been deploying storage on an ad hoc basis for so many years. The real issue here is storage management cost. Ultimately, however, I don't believe that you solve the problem by outsourcing it.

Do you think SSPs will play a major role (like ISPs) in the future?

SSPs are definitely a wave of the future. They are awaiting high-bandwidth pipes to deliver their original promise to provide a remote storage utility service. In the short run, SSPs will benefit from the willingness of companies to outsource their storage problems. They will need to develop a more universal and dependable set of services, however, if they're going to be able to weather changes in IT outsourcing behavior.

Where do you see these SSP companies in the next year with the growth of, and need for, storage and storage management?

Data management costs will drive SSP adoption in the near future. However, it remains to be seen whether SSPs will be able to reduce overall storage costs. If they can, they will have tremendous staying power.

Click here to enlarge image

Jon William Toigo

This article was originally published on March 01, 2001