Petroleum Geo-Services Corp. (PGS) was looking for Fibre Channel parallel file system performance on Ethernet NAS money. It found this happy intersection with a proprietary Linux-based file system running on clustered hardware by Panasas Inc.
Like most geological survey companies, PGS must process huge amounts of data very quickly: 1,500 terabytes (TB) flashes through its disk systems in Houston, London and Perth, Australia, in a typical month. According to PGS' director of global computer resources Andrew Wrench, that number can sometimes get up to 2,000.
"We're like any other factory," was how Wrench put it. "How we move data, and how fast we move data, is how we make money."
Another application, Wrench said, will use 50% of PGS' compute capacity this year. A handful of Panasas units (100 TB) are able to reduce the total elapsed time taken up by that application by about 15%.
"So as a very minimum, the benefit to us is a reduction of 7.5% of our total compute," Wrench said, "Which would cost 5-10 times the amount to purchase, as did the handful of Panasas shelves."
Traditionally, PGS has managed its data with a SAN that used parallel file systems -- both Advanced Digital Information Corp.'s StorNext File System and IBM's General Parallel File System -- on Engenio Technologies Inc. hardware, as well as a handful of Network Appliance Inc. FAS840s filers in each of its locations running NFS to store metadata on the massive systems.
"We were looking for a more cost-effective way of running parallel file systems on Linux," Wrench said. "And we were looking to get away from the performance bottlenecks and capacity limits we encountered with NFS."
Panasas' clusters run a proprietary parallel file system called Active Scale File System, PanFS for short. The file system is an advantage over the SAN-based file systems, Wrench said, because Panasas doesn't charge a per-node licensing fee.
"The cost of parallel file systems on every node or CPU was becoming oppressive," Wrench said. "It was around $1,000 for each license, and in our environment, that limits growth."
The older equipment will not be wiped out entirely, Wrench said -- rather, it will move down the storage food chain from primary to secondary storage, and will be gradually phased out when the equipment depreciates.
"They're mature and they work well," Wrench said, "but both are far too expensive to continue to install, license and manage. Panasas has become our de facto standard for disk expansion."
In the course of evaluating new file systems, PGS also bandied about products from Cluster File System Inc., BlueArc Corp. and TeraScale LLC. Their technologies may be comparable, Wrench said, but Panasas got there first.
"I'm not a university -- I can't run eight products side by side and study their theoretical differences," he said. "Panasas got to us first, and was a stable enough product to meet our needs. That's all there was to it."