Exposing Fibre Channel

Exposing Fibre Channel

By Kon Leong

GigaLabs Inc.

In the parable about the emperor`s fine clothes, it took a child to point out that the emperor was wearing no clothes. Today, the storage industry is so enamored with the fine-spun yarn of Fibre Channel vendors that it seems it`s now time to ask: Are they wearing any clothes?

I think not. Let me show you what I mean through a point-by-point comparison of Fibre Channel and SCSI.


Fibre Channel proponents claim Fibre Channel is a lot faster than SCSI.

Fact: SCSI is now running at 80MBps; Fibre Channel at 100MBps. And SCSI will ramp up to 160MBps next year.

Fact: Fibre Channel`s performance does not match the hype. Network Computing magazine tested three Fibre Channel RAID arrays in November 1997. Their performances were so lackluster that they compared them to 40MBps SCSI arrays.


Fibre Channel proponents claim Fibre Channel can be used for longer distances than SCSI.

Fact: With copper, both Fibre Channel and differential SCSI can be used for up to 25 to 30 meters. With inexpensive expanders, twice the distance is possible with SCSI.

Fact: With fiber, 2km is possible with Fibre Channel. However, SCSI can go much farther using existing ANSI standards. For example, with HIPPI, distances of 20km on single-mode fiber are possible.

Fact: For distances beyond 20km, SCSI over a switched ATM carrier network is also available.

Number of devices:

Proponents claim FC-AL can support up to 126 devices, versus 8 devices for single-ended SCSI and 16 for differential SCSI.

Fact: FC-AL performance begins declining when more than a handful of devices are running concurrently.

Fact: SCSI`s smaller connectors allow you to fit four connectors into a single PCI slot, for a maximum of 64 devices per PCI slot.

Fact: SCSI LUN Bridge Technology can extend the number of devices to 900 per host adapter.

Fact: With a SCSI switch, the number of SCSI devices that can be physically linked is practically unlimited.

Switched topology:

Fibre Channel vendors claim Fibre Channel provides the flexibility and power of a switched topology, while SCSI does not.

Fact: SCSI can be switched and networked. GigaLabs, for example, already ships SCSI switches.

In direct contrast to the recent hype, Fibre Channel vendors have chosen to downplay the following areas of concern:

- Uncertain adoption, uncertain investment: Unlike SCSI, Fibre Channel technology is not proven, nor is it certain that user and vendor communities will irrevocably commit to it. If widespread adoption of Fibre Channel fails to materialize, early adopters may run the risk of being locked into a niche market without vendor support. Early investments in Fibre Channel on the network side suffered such a fate. SCSI, on the other hand, represents a far safer investment with broad support from vendors and users.

- Uncertain interoperability within the same standard: It`s unlikely that users who invested in "quarter-speed" Fibre Channel a few years ago will be able to use these products with "full-speed" Fibre Channel devices, even if the products are made by the same vendor. With Fibre Channel, forward and backward compatibility is simply not assured. In contrast, SCSI has always promised and delivered backward and forward compatibility.

- Uncertain interoperability: Proprietary versions of Fibre Channel exist. Sun`s Fibre Channel devices, for example, will not work with Hewlett-Packard devices.

- Uncertain connectivity to peripherals: Currently, Fibre Channel users have limited peripheral options. Only a few vendors are shipping Fibre Channel RAID arrays, and Seagate is the only manufacturer shipping Fibre Channel drives. Users may have to wait a long time before other Fibre Channel peripherals are available. In contrast, every peripheral vendor offers a SCSI version of its product.

The Fibre Channel hype grossly underestimates the power of the SCSI legacy. The ability to network the SCSI bus enables users to protect their legacy investments.

SCSI has been the industry standard for over a decade, so users and vendors have already invested significantly in SCSI technology. Moreover, the implementation of networked SCSI is as simple as "plug-n-play"; users do not have to change existing configurations. The closest analogy would be a shift from shared Ethernet to switched Ethernet, where a switch is simply dropped into the existing configuration.

Now that SCSI can be switched, networked, and extended to 20km, I have not heard a single cogent reason why users should move to Fibre Channel. For users to contemplate a migration to the new technology, the price-performance advantage would have to be a quantum leap forward, and not just an improvement.

If you want to peer into the future of Fibre Channel and SCSI, look at the ATM vs. Ethernet battle. The similarities between the two debates are uncanny. Vendors started the ATM hype, the press amplified it, and soon ATM was all the talk at trade shows and technology conferences. As for Ethernet, those that remained faithful to this technology were branded reactionaries. No one listened to users who worried about the costs of forklift upgrades and about protecting legacy investments. But Ethernet fought back. Ethernet went switched, and it went faster. And the death knell sounded for ATM in LANs. Similarly, SCSI has gone switched and it has gone faster. The next step is up to users.

One sorry consequence of the ATM vs. Ethernet debacle is that everyone lost. Vendors had invested heavily in R&D users invested prematurely in ATM-based LAN infrastructures; and in hindsight many member of the press and analyst communities ended up looking a bit silly.

Maybe this time we should look before we leap. Until vendors show signs of responsibility and start listening to users before they jump on the latest technology bandwagon, we are doomed to repeat past errors--until eventually, it`s the users who point out that the vendors aren`t wearing any clothes.

Click here to enlarge image

Kon Leong is president of GigaLabs Inc. in Sunnyvale, CA.

This article was originally published on March 01, 1998