Starting with the Urika-GX model, the Markley Group plans to offer the whole Cray supercomputer line on a cloud rental basis.
Seattle-based Cray Inc. has cut a deal with the Markley Group to provide its supercomputers on a pay-as-you-go basis among Markley's cloud computing offerings, adding new meaning to the term "supercomputing-as-a-service."
Starting with the Urika-GX model, Markley plans eventually to offer the whole Cray supercomputer line on a cloud rental basis.
Figure 1: *The Cray Urika-GX with pre-integrated Cray Graph Engine (CGE) for pattern-matching capabilities will be the initial supercomputer-as-a-service offered with Cray's entire line to follow. (Source: Cray) *
Today, to run programmes in minutes on a supercomputer, compared to hours or days on a conventional server, you need to either buy your own or beg for time on one owned (usually) by a U.S. National Lab. Academics routinely use national lab supercomputers, but even if given permission, have to queue up their job with those from dozens of others. Only the U.S. national lab itself routinely has access to all the resources of a supercomputer for a single job.
"There are no other supercomputers in the cloud," Jeff Flanagan, executive vice president at the Markley Group told EE Times. "Our unique offering uses an independent reservation model by the week or even month for exclusive access to a Cray supercomputer—not sharing it with others."
Figure 2: The Markley Group cloud data centre, located in Boston, houses massive arrays of servers and storage units. (Source: Markley Group)
The reservation holder does not have to waste time uploading their petabytes of data or debug their programme during their supercomputer reservation either. Markley Group provides to tools to prepare all aspects of your supercomputer programming so all you need to rent is enough time to run the runs.
Figure 3: The Markley Group cloud data centre houses massive storage arrays for Big Data access. (Source: Markley Group)
"You can upload your data ahead of time, then use our virtual machine to emulate the Cray supercomputer. That way you are ready to put your programming and data on the Cray from the get go," said Flanagan.
According to Cray, many users have been asking for supercomputers-as-a-service as a result of the massive amounts of data that businesses are having to process, sometimes taking weeks of months to get results back that management wants in hours or days.
Figure 4: The Markley Group cloud data centre uses state-of-the-art chilled water in row cooling (foreground). (Source: Markley Group)
"What is different today is the need for supercomputing beyond the national lab," Fred Kohout, Cray senior vice president, told EE Times.
The Markley Group also claims to have been queried by their users as to when they would start offering supercomputer services.
"We have pharmaceutical and genomic clients using arrays of blades today, but a Cray supercomputer will get their answers more quickly," Patrick Gilmore, the Markley Group's chief technology officer, told EE Times. "Now they can get in a few minutes when they used to have to wait hours or days. We estimate up to a five times faster execution time for these kinds of results."
Figure 5: The Markley Group cloud data centre covers 727 million square feet with a million dedicated to the new supercomputer-as-a-service. (Source: Markley Group)
But why would Cray offer a supercomputer-as-a-service? Won't is cannibalise sales of their supercomputers to Big Pharm, life sciences, aerospace, government, banking and more? Not so, according to Kohout.
"We see this as an 'and' proposition, not an 'or,'" Kohout told EE Times. "The more people that use a Cray supercomputer, the more they will want one, and Markley will even host it for them so they don't have to build-out their own data centre."
Markley Group also guarantees 100% uptime, backed by a Service Level Agreement the company claims is aimed at generating the maximum profit for the customer—what Kohout calls "deep client engagements."
The first Cray computer to be hosted by the Markley Group is the Cray Urika-GX integrated with Cray's own Graphic Engine for quick pattern-matching capabilities using highly-scalable parallelisation. The Urika-GX houses 48 Intel Xeon E5-2600 v4 nodes for 1,728 cores total, plus 35TB of PCIe interfaced solid-state-drive on-node memory and the Aries optical interconnect.
First published by EE Times U.S.