QLogic set to push 4Gbit/s Fibre Channel
Is there a future in the QLogic intermediate step?
By Bryan Betts, Techworld | Techworld | Published: 00:00, 02 September 2003
Fibre Channel fabrics are getting faster. They doubled in speed from the original 1 Gbit/s and next year they will double again to 4 Gbit/s.
If your immediate thought is: "Hang on a moment, wasn't the next step supposed to be 10 Gbit/s?" then you are at least half right. There has been a 4 Gbit/s specification for some time, but it was only planned for use inside subsystems such as disk arrays, where data bursts are capable of saturating a single fibre.
What's new is the idea of using it in the SAN fabric too; the question is whether it will find a useful niche, or merely confuse the buyer. The advantage, say 4 Gig's proponents, is that it is backward-compatible with slower devices, whereas 10Gig is not. But with 10Gig fabric products on the way, do they have time to establish 4Gig before it gets overtaken?
The main mover behind the 4Gig SAN is Fibre Channel specialist QLogic. Although not as well known for its switches as Brocade, say, QLogic has a much bigger role in SANs than many would suspect, because it also develops the Fibre Channel chips used in many disk drives and other storage components.
QLogic's VP of marketing Frank Berry says it was this latter business that spurred the idea of a faster fabric: "What triggered our interest was that we had already developed silicon for disk drives, which was not a controversial issue. So we had to develop silicon for RAID controllers and it turns out the chips for those are exactly the same as for fabric devices."
He spoke to customers and found many surprised to hear that 10Gig Fibre Channel would not provide the backward-compatibility they wanted. "We also found that the price of 10Gig was a big issue," he adds. "Intel introduced a 10Gbit Ethernet NIC at $7995 and that served as a great reference point for customers - even if it halves next year and the year after, it's still very expensive."
By comparison, he says that 4Gig products - switches, host bus adapters (HBAs) and chips for other devices - will launch early next year and at much the same price as QLogic's existing 2Gig Fibre Channel range. As with today's Fibre Channel switches, they will automatically sense the speed of devices attached to them, defaulting to the lowest common denominator if there is a mixture of speeds on a single fibre.
So far though, QLogic is the only major SAN fabric vendor to commit to 4Gig, and Berry acknowledges that even QLogic will release its 10Gig-capable switches first. These will have separate 10Gig-only ports for inter-switch links (ISL), which is where higher speeds are most valuable.
"We don't expect any hosts or storage to have 10G ports when we launch," he says. And he adds that even when 10Gig HBAs are available, only the fastest servers will be able to take advantage of them, as 10Gig's throughput is so high that only the second generation of PCI-Express buses, running at 8X speed, will be able to cope.
However, other vendors are sceptical. Most are planning to add 10Gig ISL ports to their switches and directors. They point out that even if it costs several times more than a 2 Gbit/s port, the fact that each 10Gig fibre replaces several slower ones means that 10 Gbit/s will still be an economical choice.
Berry insists that there is still room for 4Gig in the medium term, especially if it is in effect, free, as QLogic is promising. "The killer application for SANs has been backup. Most libraries, from the midrange up, have multiple fibre ports because they have multiple tape drives and those can saturate a single fibre. Essentially, 4Gig means you require half as many links."
He admits though that 10Gig is the long term route: "We expect in future to see 10G storage systems and switch ports, fanning out from the switch to 1Gig, 2Gig and 4Gig servers."
The conclusion has to be that yes, 4Gig could give a useful short term boost in some areas of the SAN. In the longer term the autosensing means it should not add much complexity. It is hard to see non-QLogic users picking it up though.