The Clash Of The Networks - Industry Trend or Event

Computer Technology Review, Oct, 2000 by Bob Rumer

Customers routinely ask important questions such as: "When will InfiniBand deploy?" "Will NAS replace or complement SAN?" "Will Fibre Channel migrate to 4 Gb/s or 10 Gb/s?" As a semiconductor vendor servicing multiple markets, Vitesse is in the enviable business of selling bullets to all sides.

LAN - The 700 Pound Gorilla

No network protocol has quenched the world's thirst for bandwidth and scalability like Ethernet. Nor has any protocol provided total universal interoperability on the international scale as Ethernet. Was it just four years ago that magazines such as this one debated how ATM would impact the LAN? Ethernet, in all of its incarnations from 10 Mb/s to 10 Gb/s, increases its momentum year by year. Customers fearlessly flock to Ethernet as the industry repeatedly specifies and successfully implements each next generation with a 10x performance boost and a 3x cost increase. A unique combination of solid technology, strong leadership, clear focus, and a broad vendor base ensures interoperability and quick adoption in a competitive business environment.

The move to switches in the last five years has leveraged the benefits of Ethernet to provide optimal end-user networks. Today, there is no doubt that Ethernet is the dominant datacom protocol. However, what does Ethernet want to be when it grows up? Will 10 Gb/s implementations move it into the Metropolitan Area Network (MAN) domain? Most likely--there is a lot of excess dark fiber to fill. The Wide Area Network (WAN)? Most likely not--too much infrastructure exists in a market which moves relatively slowly. Into the SAN? There are some opportunities here.

Today, Ethernet targets storage with highly optimized Network Attached Storage (NAS) servers which efficiently convert file-oriented data from Windows (CIFS) and Unix (NFS) into block-oriented storage traffic. This application is gaining momentum due to ease-of-use, cost and interoperability. Initially NAS was viewed as disruptive technology to SAN. However, its real strength is as a complementary network to SAN.

Longer term, Ethernet is developing storage-oriented traffic encapsulation protocols which would move storage-over-IP. When the politics surrounding storage-over-IP settle down, Ethernet will transport SCSI block-oriented traffic and thereby target applications in SAN. At this early stage, the crystal ball is a little hazy but it should clear up by the end of the year.

SAN-Moving, Managing And Maintaining Your Data

SAN is the "Killer App" for Fibre Channel. The connectivity, redundancy and reliability of high performance demanded of storage networks are all addressed by Fibre Channel. In its early stages, Fibre Channel promised unification of storage and networking data into a single protocol, just as InfiniBand does today, but the early applications in storage carried momentum into SAN. While this has been a match made in heaven, SAN suffers from delayed widespread adoption due to cost, interoperability woes, a lack of centralized management and complexity. All of these are attributable to the central weakness of Fibre Channel--a lack of direction at inception. Fortunately SAN found Fibre Channel which has leveraged the benefits of Fibre Channel over SCSI and is quickly improving its weaknesses.

The benefits of SAN are compelling for high-end "enterprise" users who realize that their data and its availability is of paramount importance to the growth of their company. Regardless of the protocol, SAN will continue to grow--it is unstoppable.

InfiniBand: Something For Everyone

As the new kid on the block, InfiniBand has gathered widespread support with industry leaders and a huge following of contributors. Initial adoption will use InfiniBand as a PCI replacement so that servers can be remotely located from I/O subsystems allowing power users to upgrade each system independently. The next application will be to link groups of servers into high performance, scalable clusters. Since InfiniBand will become standardized in servers, users will be looking for ways to connect to this already-paid-for feature rather than add Fibre Channel or Gigabit Ethernet adapters to their systems. This will provide significant momentum to the adoption of InfiniBand.

Unfortunately, InfiniBand was initially promoted as the ultimate network for all datacom traffic regardless of the application. This seems silly. However, InfiniBand looks like it will certainly succeed at its core applications: replacing PCI and standardizing clustering.

Similarly to Fibre Channel, the limiting factors to InfiniBand's ultimate growth may have been determined at its inception: the lack of a unified vision. A network conceived in 1999 cannot be all things to all people. However, the community seems to be regaining its focus and appears to be implementing a very strong set of features.

At the 10,000 foot level, these three fabrics each solve serious problems well: LAN--end user connectivity/scalability, SAN--data protection/management/availability, InfiniBand--performance/scalability. Because of this, each fabric will grow healthily in these separate arenas. Regardless of protocol, the three network model will remain optimal for years into the future. The question is: Will one network be able to make significant inroads into the other's application space?