Oil ITJ—What is SGI’s raison d’être?
Bartling—SGI is solving technical computing’s hardest problems—in defense, manufacturing, media and sciences including space, pharmaceuticals genomics etc. The energy sector represents 10-15% of turnover. The common theme is very large data, massively parallel processing and visualization.
Oil IT J – What is your contribution to Energy IT?
Bartling—We have three product lines in the energy business. Storage management, computing and visualization. Storage management is SGI’s ‘best kept secret’ as data growth soars and distributed data management is the norm. Companies need to avoid replication and versioning. SGI’s CXFS* is ten times faster than NFS and scales to multi-Exabyte file systems and Terabyte file size. LightSand** allows for distributed file systems spanning thousands of kilometers. Distant parties share the same view of data supporting mirror failure/backup systems. This has become very important post 9/11 now that the Federal government mandates that backup systems need to be at least one time zone apart. These solutions leverage OC48 high bandwidth fiber—distributing businesses without burdening IT. One SGI customer has one PetaByte (PB) of data on spinning disks and is planning to go up to 5PB on disk with a forecast 50% capacity growth per year. Only SGI can handle this.
Oil ITJ—And on the computing front?
Bartling—With the new Altix 3000, SGI enters the high performance computing (HPC) cluster business. But our approach differs from other cluster vendors. To specify our Linux-based cluster offering we combined the good things from clusters and shared memory architectures (SMA). So SGI’s new clusters combine SMA’s single system image and global shared memory with the price and openness of the cluster. We have eliminated disadvantages such as the SMA’s proprietary OS and high price and cluster annoyances such as managing multiple OSs, node bottlenecks and poor connectivity. An SGI cluster ‘node’ is build with 64-128 processors, a single copy of the operating system and up to 8TB of globally shared memory. These machines run a combination of Red Hat Linux and SGI’s ProPack extended Linux. ProPack applies SGI’s supercomputer R&D and Cache Coherent Non Uniform Memory Access (CCNuma) architecture to cluster-based computing. CCNuma operates at 6.4 Gbytes/sec. as opposed to more conventional clusters running Myrianet at 1Gbit/sec. We recently sold a 256 processor system to Total in Pau for seismic processing. Paradigm Geo also likes our shared memory model and interconnect bandwidth. Another development is the use of graphics processing units—see below—for data processing. Look out for an announcement of a strategic move to Linux with our Onyx 4 soon.
Oil ITJ—Clusters are already widely used in seismic processing. But how applicable is the technology to reservoir simulation?
Bartling—CCNuma does offer improvement in seismic processing—systems can scale to 512 processors with just four instances of the operating system. But reservoir simulation is where the CCNuma architecture comes into its own. Schlumberger resells a CCNuma-based machine bundled with Eclipse. Total and Marathon use these systems and have reported that it outperforms everything except the IBM Power 4.
Oil ITJ—And the third line of business—Visualization?
Bartling—We continue to lead this segment and to innovate—especially with our high end Infinite Reality Onyx solutions. We have observed that the most influential demographic in visualization is the 12 year old male. Gamers have driven unbeatable price performance solutions from companies like ATI and 3DLabs—these are truly extraordinary machines. This is cool—so we’ll use the technology. Our next generation Onyx systems will use ATI cards. But the way we use them is somewhat different from the gamer. The Onyx 4 sports 24 cards on one computer—and soon 32. These are very high end Wildcat series graphics cards.
Oil ITJ—Who is using these systems in oil and gas?
Bartling—VoxelVision uses the Altix 3000 in an interesting manner—the computation is performed on the cluster with only the graphics file shipped to the workstation. Another remote visualization option is provided by our VizServer which has been used by BP to connect London to Baku and Cabinda. Elsewhere users have driven a 24 tile PowerWall*** with an incredible pixel density. This also allows them to load seismic attributes and reduce pixel depth.
Oil ITJ—What of SGI’s conventional MIPS/IRIX architecture?
Bartling—There are lots of good reasons to keep the MIPS chips. They run cool so a high density, low footprint is possible. This is important where speed of light latency issues come into play. We will continue to support and will up clock speeds. But oil and gas will probably move to Linux.
Oil ITJ—We have noted that although the conventional wisdom deprecates chip building today, NEC seems to have woken the HPC world up with its super-calculators****. What is SGI’s take on chip manufacturing today?
Bartling—That’s a good point! All I can say is that today we are increasing spend in Intel/Linux space.
Oil ITJ—How important is the oil and gas sector to SGI?
Bartling—Oil and gas customers have plenty of cash and lots of opportunities. The downside is the shrinking population. We have taken these issues and decomposed them into an IT problem. Our offering is to support decision making with less people by leveraging data transformation. Storage, processing, visualization and collaboration all come into play. Our balanced technologies combine to increase ‘insight velocity’.
*CXFS—SGI’s multi-platform file system.
**LightSand—SGI partner lightsand.com.
***PowerWall—from FakeSpace.com.
****NEC’s 40 teraflop Earth Simulator.
© Oil IT Journal - all rights reserved.