Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
Recently an industry ruckus was stirred by an announcement made by Spirent, a network performance analysis and test manufacturer, which reported it had run a new test on Fibre Channel fabric switches that none could pass. Spirent’s test simulated the maximum industry standard of 239 switches in a single Fibre Channel fabric.
None of the switch vendors, it turned out, could actually support the standard, even though some had specified such support in their product data sheets. Incensed that Spirent could perform such an unfair test, the Fibre Channel Industry Association returned fire, stating that Spirent’s test had nothing to do with real world conditions and that Spirent was simply trying to sell more test software.
Hypocrites! declared Spirent, pointing out that the Fibre Channel emperors were in fact stark naked when it came to inflated claims.
Opportunists! responded the FCIA, for attempting to turn a non-issue into a means to sell test suites.
As it turns out, though, this heated exchange is really much ado about nothing, especially in light of new SAN routing technology on the market.
The magic number of 239 switches in a single fabric is the product of the address space for Fibre Channel fabric domains — 8 bits, or 256 possible addresses, minus 17 reserved addresses. That leaves 239 possible unique addresses that could be assigned to different fabric switches in a single, flat network.
This theoretical maximum has little to do with day-to-day storage area networking, though, and in practice most vendors cannot reasonably support more than 20 or so switches in a single fabric. So while Spirent’s test suite has little practical value, as defensively stated by the FCIA, neither do vendor data sheet claims of supporting 239 switches. Egg on both your faces, an impartial judge would declare.
Of Honesty and Hypocrisy
In the early days of Fibre Channel technology (7 years or so ago), I received a call from a storage vendor about arbitrated loop technology. The standard for arbitrated loop specifies up to 126 end devices per loop, with up to a 10 km segment per device.
The inquirer asked if we had actually tested this theoretical maximum, to which I replied, of course not, and to which the inquirer responded with shock and indignation. How could we possibly sell a product that had not been tested to the maximum standard? It was the sheerest hypocrisy on our part.
Obviously, in the real world, no one would run an arbitrated loop that was a total of 2,530 km (126 x 20 km in both directions per device) in circumference. If it worked at all, it would have had pathetic response time, and a test to demonstrate this maximum would have been pointless. To the inquirer’s credit, however, support for 126 devices at the maximum of 10 km per device, even if valid by standard, should not have appeared on vendor data sheets.
The problem, though, is the first vendor that's truly honest is easy prey to the conscience-challenged competition. For Fibre Channel fabric switches, for example, the data sheet should really just specify the number of switches actually tested and supported in a single fabric (e.g., 24).
But since there's no law preventing other vendors from listing the theoretical maximum, and because a ‘24‛ simply doesn't look all that impressive compared to a ‘239,‛ the gross exaggerations of one vendor forces all others into opportunist submission: we all support 239 switches in a fabric. Large asterisk. Read the best practices manual.