Intel squeezes one million IOPS from desktop
Seven SSDs = 5,000 hard drives
IDF Intel has managed to squeeze one million IOPs out of a two-socket desktop tower.
Intel Fellow Rick Coulson showed off the eye-opening setup during a keynote Tuesday at the Intel Developer Forum in San Francisco. The secret ingredients are SSDs tightly coupled to a host in a highly tuned setup that Coulson called "co-optimization."
Coulson's storage technologies group works to improve SSD performance. But that's not all. "Perhaps more importantly we look at the platform and see how the SSD and the platform play together," he said. "We look at the interface between the SSD and the system, software, overhead, those sort of things."
Since SSDs are, according to Coulson, about 100 times faster in terms of latency than hard drives, there are performance bottlenecks to be overcome in the physical interface, the protocols on that interface, the software driver, and the chipset interface.
"So as we look at optimizing some of those things," he said, "like interrupts, driver speculation, improving the physical interface between SSDs, and the system, we expect great gains in power, performance, and cost."
Gains in performance, for example, that add up to 1M IOPS (input/output operations per second) in Coulson's lab, where he hooked up a dual Xeon 5500 desktop tower to seven Intel SSD prototypes - four in the tower and three in a PCIe expansion box - and ran a 4K read/write benchmark on it.
"This many I/Os per second is about four gigabytes per second of storage bandwidth," he claimed. "As a storage guy, that's a huge number, a very huge number."
He also pointed out that during the test, the CPU utilization of the tower was about 50 per cent. "That's really nice," he smiled.
To the non-storage guys in the audience, he said "I want to put this in perspective: this is about 5,000 disk drives worth of random performance. And 5,000 disk drives, when you think about it, is a lot of floor space. Maybe more importantly, that's maybe 50 kilowatts of power - and this whole setup is about 500 [watts]."
To be sure, Coulson's setup is still in Intel's labs, and is tweaked to perfection. But no matter how you cut it, one million IOPS - 1,076,600, to be exact - is one hell of a lot of desktop I/O. ®
STEC SSD on SPC-1
There was no "SAS HBA", it was a 5908 RAID controller. Moreover, 45K/17K IOPS @ 4K is about 100MBytes/sec. Average SPC-1 I/O size is 8.3K. Therefore at 5,600IOPS the STEC SSDs are doing only about 45MBytes/sec. each.
Now, you are suggesting that the STEC SSDs could have done several times more IOPS, but they were bottlenecked in the SPC-1 benchmark by the IBM 5908 RAID controller.
You are suggesting that IBM put more than twice as many SSDs on that RAID controller than what it could handle?!?!
Given that the key metric here is Dollars/IOP, and that the $100K worth of SSDs were 95% of the system price, it would be extraordinarily silly to put more SSDs on a RAID controller than were needed.
By the way, Ideas International is the analyst firm formally appointed by the Storage Performance Council to cover technical SPC benchmark analyses, and Gary Burgess is the head analyst. See here:
There you have it. The STEC SSD is actually costs MORE per IOP than HDD. If these observations are all so wrong, maybe IBM should take it up with Gary Burgess?
And while you are at it, when do we get to see the SPC mandated "Full DIsclosure Report" that remains missing from IBM's SPC-1 submission? IBM was supposed to have submitted it before the end of July...
Re: Not just Intel... look at STEC BS
Disclaimer : I work for IBM, not STEC.
Before casting aspersions, maybe the "anon coward" here should do some homework. The limit in the SPC-1 pSeries SPC is not the STEC drive, but the SAS HBA and ESM units used to connect to the SSD. If you benchmark an STEC drive in the right enclosure / fabric you can SUSTAIN around 45,000 IOPs (read) and 17,000 IOPs (write) from a device when doing 4KB blocks.
SPC-1 uses between 8KB and 16KB blocks, and as we all know, SSD IOPs half as you double the block size, therefore at say 12KB this would bring the number down to around 6,000 IOPs for SPD-1
Go do the maths before you start mud-slinging.
Just because it's a technical site doesn't mean you don't have to write properly.