EMC president and incoming VMware chief executive Pat Gelsinger most likely shot down any hope that the company’s storage arrays would be built around the ARM architecture.
Gelsinger, who also helped orchestrate the VMworld show in San Francisco this week, presented an Aug. 29 keynote at the Hot Chips conference in Cupertino, Calif. Afterward, an audience member told Gelsinger that as many as 25 percent of all servers could be shipped around the low-power ARM architecture, then asked if Gelsinger agreed with that estimate.
The question was a bit pointed. Back in 2008, IBM PowerPC chips powered EMC’s storage arrays. Two years later, all four of EMC’s product lines—Symmetrix, Clariion, Celerra and Centera—shifted to the Intel Xeon processor.
Gelsinger is a former Intel enterprise executive who helped bring to market the Xeon processors that EMC adopted wholesale in 2010. EMC’s switch to the Xeon architecture away from PowerPC came about ten months after Gelsinger joined EMC as president of its Information Infrastructure Products group.
Gelsinger told the audience member that the situation is unlikely to change, even if ARM could deliver workloads at a fraction of the power of an X86 chip.
“If I take the power of the core to zero, the power requirements… to do a server workload, I/O memory whatever, is not changing a lot,” he said. “The individual X86 core—take a low power core, you’re only talking two or three watts to a core, at the lowest end is close to a watt. Take that to zero, it doesn’t matter—you’ve still got memory power, I/O power, et cetera associated with that. So fundamentally I don’t think the math makes sense when you go look at it.”
(In other words, even with a core requiring a bare minimum of wattage, associated components have their own power needs.)
At this point, EMC’s arrays are homogeneous, running on the same architecture. To alter that, Gelsinger continued, would add unnecessary complexity.
“That means I can’t schedule across the whole pool, I can’t pick up workloads from here and move them there, I can’t power them down, I can’t fail them over to other data centers—the cost of heterogeneity in a cloud computing model to me is very high,” he added. “So I don’t think the math works in terms of power/performance characteristics, and I don’t think the architecture works in terms of cloud-scale computing.”
Despite his skepticism, he suggested that “our software” would support any emerging model that actually met the math requirements.
In his speech, Gelsinger beat the drum for the cloud, a familiar refrain for those who attended VMworld in San Francisco earlier in the week. In some sense, Gelsinger’s argument against ARM makes sense: in a world of virtualized hardware, the physical CPUs supporting the VMs need to emphasize IOPS and computational power first, with low power following behind.
Gelsinger noted one surprising way in which Big Data would directly benefit EMC’s customers. A few years back, he said, an unnamed disk-drive supplier had gone out of compliance, and hard drives started failing over a period of about ten months. EMC had to offer field replacements, at the cost equivalent of roughly $200 million.
“But what we did then is that we started to look at the real data and we realized that we could have detected this approximately ten months sooner had we been looking at the full synthesis of real-time data that had been occurring with all of our drives,” he claimed. “But we hadn’t been gathering all of that data across all of our customers and all of the drives.”
By the middle of 2013, EMC will have installed data analytics on 70 percent to 100 percent of its drives, allowing the company to proactively detect and replace hard disks in danger of failing. “That’s the type of thing that Big Data is going to enable us and our supply chain to do, as we tie it to how we run our business,” Gelsinger said.
Image: Konstantin Sutyagin/Shutterstock.com