SANs, storage area networks, are an integral part of the global information infrastructure today. Distance limitations (latency) of the existing protocols such as Fibre Channel, FICON, ESCON have been resolved so that information/data centers can be located at nearly any distance from each other. The host of IP tunneling protocol embodiments in switches and routers have nearly but not quite eliminated the barriers. Switching and routing technology advances (and new standards such as MPLS) enable the “Any time, Any level and Any place” on demand provision of data/information to the requestor. Security methodology has also advanced but is not yet perfect to eliminate the possibility of destructive intent.

Optical connectivity technologies at multiple levels – WDM, CWDM, and DWDM – plus photonic transport evolutions enabled native and multiple protocol transport particularly in the Metro and Campus environments. Advances in the copper layers also have enabled higher speed transport at the 10g level and GigE to the desktop is a reality (but the question remains who need this at the desk top given the computer bus speed levels and limits).  We continue to feed the need for speed at all levels. Now we can transport video at realistic speeds when desired if the high speed connectivity is available. SANs have provided a mechanism for sharing and storing of high volumes of data. Yet, as improvements are embodied in the infrastructure, new limitations and bottlenecks are located.

As the transport and delivery speeds increase, the limitations of the computer/server bus is becoming a limiting factor. Unfortunately, the PCI bus and newest evolutions are inadequate beyond a year or two horizon for higher speed realization. They are fine for most applications but the volumes of data and processing requirements necessitate changes. Blade servers offer the opportunity for high capacity cluster computing with high availability and low TCO costs. Short path lengths help but protocol and bus improvements are vital. Incorporation of switching and routing capabilities on a blade (or in the server) plus the increased density and smaller server footprints offer significant improvement in Reliability, Availability and Serviceability. Improvement here reduces TCO significantly. The emerging technologies offer significant opportunities and rewards for those with a design eye to customer needs.

Any simplification industry can enable will be richly rewarded. Embodiments of the emerging clustering stand – Infiniband – on the surface offers an interesting opportunity for it captures the best of many worlds. However, there is reticence on the part of the manufacturers (despite lip service to the contrary) to fully commit to this technology. Advantages can be enumerated but need to be placed in the proper perspective. There is an installed base of SANs and connectivity (switches/routers/server/storage devices) that will not be replaced. Change will be evolutionary BUT will be accelerated by the Blade Server impacts for they are approaching critical mass and will reach nearly a $3 Billion market by the end of 2005.

The following addresses the existing environment and possible evolutionary path to simplify the current complexities in the global information infrastructure. First we will briefly describe a typical SAN and connectivity infrastructure. Second, we will address a possible evolution of this infrastructure. Third, we will present a possible topology configuration using blade servers and Infiniband.

Figure 1 == Typical SAN Installation

Figure 2 == Possible SAN Topology Evolution

Figure 3  == potential Blade and SAN Environment

Each will be discussed separately.

Fig. 1 Typical Current SAN Connectivity Requirements

Nearly all the storage area network infrastructure today is Fibre channel. This was originally developed in the late 1980's as a higher performance standard to FDDI. It was designed by computer centric personnel versus network centric thus the high performance and guarantee delivery elements in FC.

This complex technology (including SANs) did not reach critical mass until the late 1900's. It is now the dominant force in SANs but is faced with severe transport distance limitations. Considerable effort has been expended in devising solutions to eliminate the distance limitation of Fibre Channel. However, solutions are limited and the field is littered with failed start-up companies who thought “they” had the solution (much to the chagrin of many VCs).

Tunneling solutions do abound and do provide a measure of relief for the distance question.  Key protocols remain FC (Fibre channel), SCSI and iSCSI, plus Ethernet in a variety of LAN speeds from 10, 100, 1000 and even 10G over the wide area.

SANs and the connectivity solution must evolve if they are to keep pace with the ever increasing information (voice, video, and data) demands of the global environment (business and personal).

A possible evolutionary scenario is discussed in the next section.

Fig. 2. EVOLVING SAN CONNECTIVITY COMPLEXITY

In the next short term evolutionary phase, we see the emergence of blade servers at the edge (plus in back and front end locales) and continued utilization of RAID, Tape storage arrays, and stand alone storage device form the competitors in this space. Storage routing becomes a more dominant aspect as part of the full mesh, non-blocking environments. Switching and routing intelligence will be moves to the devices and some will be dedicated to specific tasks.

Virtualization (server and storage) increases in importance here. In some cases, Ethernet will be the primary transport protocol but Fibre channel will retain a position in the evolution. Cost considerations become significant with added utilization of RAID, ATA devices and lower cost servers. They will not obviate the need for sophisticated systems as installed today but may represent a significant percentage of the expansion of the infrastructure.  Serial attached SCSI (ANSI T.10 spec) may grow in importance (including iSCSI). Manufacturers may incorporate ATA technology for lower cost blade systems. 

We foresee a key role for Ethernet, particularly 10GbE in future embodiments. Eventually, we would expect to see the transport speed difference between FC (current 2Gbps installed) and Ethernet (current availability of 10GbE resolved for this will be necessary to simplify operations.

One of the problems that rapidly become evident is the severe limitation of the PCI bus in increasing speeds though out the systems. Evolutions of the PCI bus will only satisfy the demands for one to two years. Clearly new standards will be needed to resolve the bottlenecks to packet and cell transport speed from the server to the user. We do feel it will continue to be an amalgam of protocols and standers in any topology. Infiniband does offer some intriguing possibilities for consideration when combined with the acceptance of the blade server.

Fig. 3 - A Topology with BLADE SERVERS and Switches

Blade Servers or Rack Mounted

Utilizing blade servers minimizes the cabling needs and simplifies the equipment topology demands. It is simple to provide redundancy at low cost and to implement virtualization as needed.  Infiniband was developed as a single unifying industry wide server I/O interconnect (IBTA spec. and SRP - SCSI protocol over IB). It is designed and functions as a simultaneous Network, Storage, and IPC interconnect. Many feel it has the potential to replace FC as box to box server storage interconnect but we do not have that grandiose opinion for complete acceptance faces a number of significant challenges including developing and maintaining traction given the large company defections lately, overcoming economic challenges in these trying times (as will most newer replacement technologies) for it requires new hardware and software.

We feel the Infiniband technology in a Blade Server design can aid resolving TCO issues, adds simplicity in topology and operation at multiple levels, and can be readily incorporated at the server level. It does need several major “design and installation” wins from the majors. To date, this has not occurred and many early Infiniband companies have fallen by the wayside. JNI though (an established company) has moved forward with real products through their alliance with Mellanox as have others. This is not an endorsement of the JNI position but merely indicative of the current state of product availability versus vaporware.

Summary and Opinion

Despite the obvious challenges and at times slippage in traction, we think it will play a role in the next several years for it does offer appreciable advantages in reliability, serviceability and resolving availability challenges. It will, however, need to be “pulled through” will emerging blade server and cluster switching technology.

We suggest our clients closely review the advantages of the new standards and consider incorporation in future blade server designs after conducting their own market demand research before committing large development funding.