Comparing Storage Connection Costs: A Worksheet

We've all seen the PowerPoint slide that depicts the relative pricing of 10GbE, InfiniBand, and Fibre Channel. Usually, the upshot is a plug for the presenter's wares.

Central to the argument is the calculation of price per gigabit/second -- the sum of per-port costs of host adapters and switches, divided by the Gbit/s delivered by the technology. Marketing math? Maybe, but the results can prove interesting.

It's an intriguing exercise overall, and after viewing a few of these slides, we at Byte and Switch decided to undertake our own comparative worksheet of per-port pricing. The result is in the table below:

Table 1: Storage Connections Compared

By way of comparison, here are two price-per-gigabit worksheets from vendors with axes to grind (names have been withheld):Table 2: Supplier Worksheet Results

There are a couple of surprises here: First, if one uses the price-per-gigabit metric, it does look (as some sources insist) as though InfiniBand is a good buy after all -- coming in lower even than gigabit Ethernet. And 4-Gbit/s Fibre Channel looks like a better deal than 2-Gbit/s by the same measure.

Without the price-per-gigabit calculation, InfiniBand runs behind 4-Gbit/s Fibre Channel in terms of costs per switch port, even though it's cheaper in adapter costs.

As for 10-Gbit/s Ethernet, it's no surprise to see it still up in the ether price-wise. After all, that's why components vendors are peddling new wares aimed at lower-priced finished goods starting in 2009.

The villain of the piece looks like 2-Gbit/s Fibre Channel, which seems to cost more than 4-Gbit/s FC overall. But there are lots of variables: A 4-Gbit/s switch with 2-Gbit/s may not have the same negative price profile.Which leads to a key point: This kind of worksheet is only as good as the pricing you feed into it. In our case, we obtained per-port pricing for adapters and switches by dividing the base prices for products from vendors and online sources like CDW, PC Connection, and SANDirect by the number of ports they provide. We averaged multiple prices obtained this way to get our basic figures. We used server adapter prices where applicable, and we compared managed switches, not unmanaged ones.

We know there are lots of pricing sources not accessible to us, including discounts from vendors for existing customers, so we can't claim to be scientific about our approach. That said, we are open to suggestion: Do you think these are valid comparisons overall? Do you have different prices to share? Do you favor a particular interconnect for reasons other than its price tag -- cost savings that don't show up on this worksheet? Hit that message board, or contact us. We're all ears.