The manager of a new supercomputing centre at Oxford
University said his latest storage choice, Pillar Data Systems'
Axiommidrange array,
reflects the fact that his high-performance computing (HPC) lab
is growing concerned with storage issues once the domain of
corporate storage managers.
When it opens its new facility in August, the newly refurbished
Oxford e-Research Centre (OeRC) will provide a central computing
and data repository for researchers at each of the 39 colleges at
the university. During the latest reinvestments in technology,
according to centre manager Dr. Jon Lockley, a new need for
datacentre storage has emerged in a computing environment once
solely concerned with CPU horsepower.
For example, "we have a particularly nasty chemistry
application, which can easily produce temporary files of up to 2
terabytes," Lockley said, a load that caused the previous systems,
which consisted of NFS file servers attached to a single 1 terabyte
(TB) unbranded JBOD, to crash. Aside from the JBODs, the existing
supercomputing center is also running EMC Corp.'s AX150 array.
Lockley said budgetary concerns were a factor in purchasing such
small systems, but added that until recently, they had been
sufficient, since supercomputing at Oxford remained focused on
providing high-octane "scratch space" for research
calculations.
Now, 2 TB chemistry files are just the tip of the iceberg as the
new datacentre storage plan gets going. Further complicating
matters is the fact that Lockley's team won't know exactly how many
researchers or how much data they'll bring with them until the
facility opens its doors.
"I've had people come up to me in the corridor and talk about
how great it is we're getting this new system, then ask me if I
have 20 terabytes," Lockleysaid. "We are probably going to have to
add more than 100 terabytes before the end of the year."
Lockley declined to specify exactly which vendors' products he
had evaluated prior to selecting Pillar's Axiom array, but said
they were storage area network (SAN), network attached storage
(NAS) and multiprotocol systems from "mostly legacy vendors."
The problem with the majority of products he looked at was that
"there were too many products where we would have had to make a
design decision from day one," in terms of storage capacity
allocation and provisioning, he said.
Pillar's array, meanwhile, scales its controllers separately
from disks, which Lockley said will give the university more
flexibility as the new datacentre gets up and running.
A software feature that sold Lockley on the Pillar array is a
part of the AxiomOne management console that allows him to create
"what-if" scenarios and ascertain from the array how many
controllers and disk bricks will be needed to maintain a predefined
level of service.
"We're headed into the great unknown with this project," Lockley
said. "We need all the help we can get when it comes to predicting
our capacity and storage purchases."
When it comes to performance, Lockley said he plans to rely on
Pillar's quality of service (QoS) feature to feed heavy files or
high-transaction database data into the compute cluster, which will
handle the heavy processing. He acknowledged that an enterprise
midrange array isn't generally associated with high-performance
computing, but said it's data management that will become the real
issue for the university once the new system is up and running.
"Everyone's having a bit of a rethink about storage right now,"
Lockley said. "Data management is going to be the issue for
us."
For example, Lockley said, backups for the system will now
probably also fall under his purview. "We used to be able to tell
people, don't count on us" when it came to protecting data, he
said. "But that was back when all we sold was compute power."