OITJ—What is Liquid Computing’s (LC) approach to the oil and gas vertical?
Weston—I moved to LC from Sun Microsystems (which has exited the HPC business) because it represented something new—both a startup and new computing paradigm with oil and gas as one of its initial targets. Our first target is upstream HPC. We’ll then move into ERP with our ‘fabric-based’ architecture.
OITJ—What’s ‘fabric-based?’
Weston—The HPC market has settled on clustered PCs linked with Gigabit Ethernet, Myrianet or Infiniband. In reality these are somewhat ‘Rube Goldberg’ solutions. Our ‘LiquidIQ’ fabric eliminates these ad-hoc connectivity solutions, replacing them with a dedicated network and blade computing infrastructure. LC’s founders came from Nortel and have leveraged the blade technology originally developed for the telecommunications industry. This means that you can have as many nodes as you need. Everything on the fabric is visible by everything else—network, storage, applications and so on.
OITJ—Sounds like ‘virtualization.’
Weston—The virtualization concept has been applied in software à la VMWare but our solution does not have the performance trade-off of traditional virtualization. Tests with the two major fluid flow simulator vendors show VIP performance up by 40% and linear scalability to 64 CPUs.
OITJ—We’ve heard that in some clusters, CPU usage may be only 10% of the theoretical rating.
Weston—Sure, because of the communications bottleneck. The LC fabric gets 90% efficiency from processors – even dual core devices. Moreover the power and cooling issues are fixed because the technology was developed for the strict environmental requirements telcos. Our liquid cooling leverages technology developed for gaming machines and lets CPUs run 10-15° cooler. This leads to greater efficiency and reliability.
OITJ—Have you addressed the seismic market yet?
Weston—Not yet, we are concentrating on the reservoir simulator market. But our solution should be good for seismic processing. It is easy to maintain, cheap on people and price competitive.
OITJ—Does the fabric address shared memory issue?
Weston—Yes. It is great for larger models. Everything is shared. In future TESLA GPU devices will become shared resources on the Fabric. In fact HPC will just be another compute resource on the system. Today we can deliver 2,000 node systems with considerable upside—the bus length can stretch to 300 m!
OITJ—What sort of compiler support is there?
Weston—We are ‘completely COTS!’ We support Red Hat Linux, AMD and whatever runs on these. VIP and Eclipse binaries and SCALI all ran. We provide tools that manage, monitor components continuously and reconfigure virtual machines to assure quality of service by adding resources or limiting application access.
OITJ—You mentioned ERP?
Weston—Yes we plan to offer technical and commercial computing on single architecture. The fabric will run on Oracle’s ‘Unbreakable Linux’ and as Landmark’s R5000 release will support the Oracle 10G Rack, we will be one of first certified platforms. Oracle Rack will actually work on the fabric!
More from Nick.Weston@liquidcomputing.com.
© Oil IT Journal - all rights reserved.