Intel eats crow on software RAID

SAS on motherboard shocker

IDF When Intel releases its Sandy Bridge-based two-socket "Romley" platform in the middle of next year, its "Patsburg" platform controller hub (PCH) will include support for serial attached SCSI (SAS).

By putting SAS support on the motherboard, Intel is embracing what it formerly shunned: software RAID.

"I'll plead guilty. We stood up here 10 years ago and told you software RAID sucked, you didn't want it, it wasn't a viable solution," Susan Bobholz of Intel's storage product marketing group told attendees at a Wednesday SAS and RAID session at the Intel Developer Forum in San Francisco.

"But that's one of those things that's starting to change in the industry," she added. "Software RAID is no longer the evil stepchild of the enterprise anymore."

Citing testing done by Tom's Hardware, Bobholz claimed that "Software RAID equals or outperforms hardware RAID these days. And this is because the host processors have gotten so much faster and you've got all the different cores going on."

Referring to software-RAID I/O performance of years past, she said: "Historically, it was slower. It just was. And it's not anymore. And in fact in a lot of cases it's a higher-performing solution."

Bobholz also discussed CPU usage. "Another thing that we tried to convince everyone — which was true way back when — is that in order to free up the CPU on your host processor, you needed to move your processing of your RAID onto another CPU in the system. But as CPUs have gotten better and faster, and you've all these different cores involved, we are anticipating that the CPU load will be very low — in some cases we're looking at less than five per cent of the CPU to get a full-performing RAID subsystem."

When announcing the support for SAS in the Patsburg PCH, Clive D'Souza of Intel's storage technical marketing group said: "Intel has gone and done what naturally complements our solution from the silicon perspective, and we have integrated SAS onto the chipset. That's a major performance statement, so I'm going to give it a few seconds to sink in."

Patsburg will also support SATA and USB. A slide in D'Souza's presentation included the USB 2.0 logo, but a Patsburg engineer with whom The Reg spoke would neither confirm nor deny whether USB 3.0 would join the party.

Intel's implementation of Patsburg's onboard SAS controller will be SAS 2.1 compliant, support up to 10m data cables, RAID 0/1/10/5, and expanders and enclosures. D'Souza also claimed that "We will be exceeding what the requirements of the T10 SAS-specs are."

Although software RAID may no longer be the evil stepchild of the enterprise, "evil" might be a word that SAS host bus adapter (HBA) makers such as Adaptec, HighPoint, LSI and other might want to apply to Intel, as the addition of SAS support to the motherboard obviates the need for their cards.

When listing the advantages of onboard SAS, D'Souza extolled the elimination of PCI HBAs: "The fact that we have integrated our SAS solution onto the chipset, we actually free up a PCI slot — and in the server world that's a big deal."

That's a boon to IT folks, said D'Souza, because "It opens up the possibility for all the data centers to go and add more functionalities that need a PCI slot."

But before HBA makers use their sharp-edged cards to either slash their wrists or attack Intel engineers, they should take comfort in the fact that their software stacks are welcome in Intel's onboard-SAS world.

As D'Souza explained: "By definition of our implementation, we are not restricting or constricting any RAID technology. If your software stack supports it, we support it." Bobholz added: "It's always up to the software," including, for example, the ability to support RAID 6, hot sparing, or variable allocation of host memory for RAID buffer duty.

Bobholz declined to give speeds and feeds for the Patsburg SAS implementation, specifically dodging a question about RAID 5, which has been traditionally problematic in terms of software-RAID performance.

"We've had our processor silicon back for 30 days," she explained. "We've got teams in the lab, locked in the lab, we're feeding them pizzas and candies under the door as they're working on really looking at the actual performance and functionality of our drivers. We will be providing that information down the road after we've done some actual measurements."

As brightly as Bobholz and D'Souza painted the onboard SAS future to be, Bobholz did note that software RAID remains dependent on the robustness of the operating system upon which it is running.

"I know some operating system vendors don't like us to say this," she admitted, "but should the operating system crash, yeah, your RAID stack will go down with it." ®