To: Fellow Storage Weenies
The following is a six month old research report from the Morgan Keegan brokerage firm, recently brought to my attention. It is very bullish on Fibre Channel, vis-a-vis IP/GE.
I debated whether to post this information on the NTAP thread or the G&K thread; it seems both appropriate and at least slightly out of place for either. Though no explicit Gorilla Game language was used, the inferences are many. Since I am not on top of recent discussions on storage on the G&K thread, I decided to post it on the NTAP thread.
Possible references to NTAP?
"VIRTUAL INTERFACE is much more than a clustering enabler. VI is a standard that allows applications to circumvent the TCP/IP protocol stack for direct application communications. Though generally considered a clustering enabler, it can support any type of traffic between two servers. For example, we believe that at least one leading NAS vendor is developing a mapping of the NFS and CIFS network file system protocols over VI in order to bypass the inefficient TCP/IP stack in most client systems.
Fibre Channel is the only open interconnect standard that can support SCSI and VI concurrently today. SCSI over Fibre Channel is already broadly commercialized, and we believe that at least two host bus adapter vendors will offer hardware implementation of the FC-VI standard by mid-year. This suggests that SANs supporting VI and SCSI concurrently will be rolled out in 2000. We therefore expect to see at least one NAS vendor begin to offer Fibre Channel as a high performance transport for file system traffic. This would be a remarkable event in the history of NAS, given the historical tendency of this industry to preach IP/Ethernet ubiquity."
If others here believe this report warrants further discussion, we can decide whether the discussion should move elsewhere.
Greg
===========================================================
System Area Networks-Who's Disrupting Whom?
Over the last few months, representatives from Adaptec, Cisco, 3Com, and EMC have turned up the volume on the concept of using IP/Ethernet as an alternative to Fibre Channel for storage networking. This idea has been floating around for years. Now that George Gilder is keen to the concept, however, the investment community is beginning to take notice. We believe the noise will only get louder going forward given the market opportunity at stake.
Ironically, some proponents of the IP Storage initiative are characterizing the concept as disruptive to Fibre Channel. We disagree. In our view, the battle that is brewing reaches well beyond the storage interconnect to include server-to-server communications in the corporate and the .com computing environments. The incumbent technology in this battlefield is TCP/IP/Ethernet; the disruptive technology is Fibre Channel. These are just the weapons, however, for the essence of this conflict is the rivalry between the computing industry (ie. Intel, Compaq, Sun, etc.) and the networking industry (ie. Cisco, 3Com, etc.) for mindshare control in the emerging market for what we will term System Area Networks. This version of the SAN concept incorporates both server-to-storage and server-to-server data traffic in the data center on a single communications platform, allowing Network Attached Storage (NAS), Storage Area Networking (SAN), and clustering convergence to form a new networking ecosystem.
Due in part to the threat of Cisco becoming the thought leader in this emerging version of the SAN, computing industry leaders Intel, Compaq, Dell, Hewlett-Packard, IBM, Microsoft and Sun Microsystems are now working to develop Infiniband. This new server I/O interconnect standard attempts to merge the strengths of Fibre Channel with the latest version of the IP addressing scheme, IP version 6, to create an entirely new internetworking protocol stack that would reside entirely in hardware. What is significant about this aspect of Infiniband is that it is NOT backward compatible with version 4 of the IP standard, meaning that, if it gains traction, end users may consider an upgrade to Infiniband from IP for backbone traffic. As a result, the Infiniband effort is directly targeting the IP installed base with this technology. This puts LAN vendors in a weaker position if they attempt to support Infiniband, mainly because their promise of backwards compatibility across all IP standards will now be challenged, and the market positioning of Ethernet will be diluted.
On the software side, the compute intensive TCP/IP protocol stack would give way to the extremely efficient Virtual Interface (VI) architecture. This standard eliminates the high CPU cycle overhead associated with processing the TCP/IP stack by providing applications with a direct connection to hardware-implemented networking platforms.
In other words, the computing industry is proposing a new networking paradigm that could make TCP/IP over Ethernet obsolete. Will this happen? It's too early to call, but in the meantime we believe the computing system vendors need a horse to ride as they attempt to fight off the encroachment of the IP ecosystem within their sphere of influence. Herein lies a compelling new market opportunity for the Fibre Channel industry.
Fibre Channel is a hardware-implemented transport with industry standard mappings for both SCSI and VI. As a side note, IP traffic can also be transmitted over Fibre Channel, and a significant effort has transpired in the industry over the last several years to standardize this technology. However, we believe that the aforementioned performance issues related to IP may IP-Fibre Channel traffic to SAN management applications. On the other hand, most of the commercial applications to date have involved the use of serial SCSI over Fibre Channel to support storage I/O traffic between servers and storage. However, we believe that commercial applications of Virtual Interface over Fibre Channel, or FC-VI, will begin to roll out in 2000. Most of these applications will involve the clustering of servers for high availability. Practically speaking, though, they could involve any type of server-to server exchange that traditionally is transported over TCP/IP/Ethernet.
So what's the point? WE BELIEVE FIBRE CHANNEL WILL EMERGE THIS YEAR AS A LEGITIMATE THREAT TO LAN VENDORS IN THE DATA CENTER.
Here's why:
VIRTUAL INTERFACE is much more than a clustering enabler. VI is a standard that allows applications to circumvent the TCP/IP protocol stack for direct application communications. Though generally considered a clustering enabler, it can support any type of traffic between two servers. For example, we believe that at least one leading NAS vendor is developing a mapping of the NFS and CIFS network file system protocols over VI in order to bypass the inefficient TCP/IP stack in most client systems.
Fibre Channel is the only open interconnect standard that can support SCSI and VI concurrently today. SCSI over Fibre Channel is already broadly commercialized, and we believe that at least two host bus adapter vendors will offer hardware implementation of the FC-VI standard by mid-year. This suggests that SANs supporting VI and SCSI concurrently will be rolled out in 2000. We therefore expect to see at least one NAS vendor begin to offer Fibre Channel as a high performance transport for file system traffic. This would be a remarkable event in the history of NAS, given the historical tendency of this industry to preach IP/Ethernet ubiquity.
Existing Fibre Channel deployments can therefore be leveraged to deliver a common platform for block and file level storage traffic as well as clustering. With support for concurrent operation of VI and SCSI, we believe Fibre Channel can emerge as a high speed System Area Network solution that drives the convergence of NAS, SAN, and cluster networking in the data center. Furthermore, we believe this convergence could over time reach a level of functional integration that provides a common framework for access and control to System Area Network resources.
Fibre Channel and Gigabit Ethernet share the same physical layer signaling technology. As the physical layer signaling standards for gigabaud networking evolve, we believe these two technologies will progress in lockstep fashion to higher bandwidth levels. Furthermore, the Ethernet standards have always trailed Fibre Channel in terms of bandwidth utilization, suggesting that Fibre Channel will remain the performance leader through the 10 gigabaud bandwidth level and beyond.
We believe the noise level from the networking camp and a select group of independent storage vendors will rise over the course of 2000 as they struggle to assemble an alternative open-standards based System Area Network platform that is IP-centric. Two core issues in using IP for these implementations are (1) its high-latency characteristics are simply impractical for bandwidth-intensive storage traffic and (2) the small packet size for IP-based messages is infeasible for block-level file transfers that are commonplace in the storage world. As a result, the standards that govern the TCP/IP stack must be revised whereby (1) direct memory access is enabled in the standard and (2) larger block sizes are facilitated to accommodate SCSI traffic. For the vendors that are moving this direction, SCSI over IP is the critical first hurdle. However, we believe that this will prove to be challenging for those vendors, mainly because this initiative is hardware-intensive and requires broad industry support. Therefore, a hardware-based protocol stack must be implemented in new silicon solutions. As a result of these complications, the current roadmap suggests an official standard for SCSI over IP will not be finalized until late 2002. All the while, Fibre Channel connections will be gaining traction in the data center, both for SCSI and VI implementations.
Already Cisco has published a draft standard proposal for SCSI over TCP/IP, and an Internet Engineering Task Force committee is in the early stages of development. However, we believe the urgency of the situation will lead certain proponents of the IP way to circumvent the standards process in order to get to market early with a solution. "Early" is a rather optimistic term, though, given the challenges that must be overcome to deliver a viable alternative to Fibre Channel. They include at least the following:
* Strong commitment from the storage industry, which has invested heavily in Fibre Channel
* Cooperation from the computing industry, which by doing so would be ceding control of the data center to LAN vendors
* A standard disk drive interface for TCP/IP, which requires the cooperation of the disk drive vendors.
* A standard for hardware implementations of TCP/IP, since software versions require too many CPU cycles
The primary conclusion of this dialogue is this: Fibre Channel is clearly the foundation for SANs. However, we are expanding our definition of the target market for Fibre Channel to SYSTEM AREA NETWORKS, to include not only SANs, but the convergence of NAS and SAN as well as server-to-server clustering configurations. The bottom line is that the target market for Fibre Channel appears to be growing significantly, leveraging its success in storage implementations. |
