June 20, 2008
This week at ISC in Dresden, Germany, storage vendor LSI is demonstrating the Engenio 7900, a storage product that they feel puts HPC centers right where they need to be: offering users a highly reliable storage solution that doesn't sacrifice performance or configurability.
LSI sees HPC centers as facing two types of workloads. On the one hand there is the purpose-oriented HPC shop that serves a very small number of users running one or two applications. For the purpose-oriented center, building a highly efficient storage architecture is a straightforward exercise that begins with characterizing the I/O patterns of a limited universe of applications and constructing the solution best suited to those patterns. On the other hand is the general purpose center, serving hundreds or thousands of users with a very diverse workload. This center faces a more significant challenge.
I/O workloads, and the storage systems that serve them, typically fall somewhere between two extremes: small blocks of random access on the one hand (file serving or metadata systems for example), and large blocks of sequential access (output files from large computational simulations) on the other. It is possible to build a system that can serve either of these workloads well, but usually not both at the same time.
With its new Engenio 7900 HPC storage system, LSI feels it has created a solution that prevents HPC centers from having to compromise one type of I/O workload in favor of another.
The 7900 is LSI's flagship storage product. The system features sixteen 4 Gbps FC or eight 20 Gbps IB host connectivity, with 6.4 GB/s performance on sequential disk reads and 5.4 GB/s on sequential writes. The unit will ship with the capability of supporting 256 SATA drives, with growth to 448 drives in future updates.
The company has placed a lot of emphasis on reliability and future-proofing with this product. On the reliability side, the 7900 features battery-backed controller cache mirroring, redundant data paths with automated failover, proactive drive health monitoring, and a hardware assisted RAID-6 among other features. To ensure that a customer storage solution remains relevant in the face of changing technology, the 7900's host interface cards are field-replacable daughter cards on the front end of the controller, allowing users to purchase a 4 Gbps FC solution today and change over to 40 Gbps IB in the future without buying a totally new system.
The 7900 is able to serve so much diversity in the target job mix by offering a wide range of configuration options, from disabling mirrored write cached to changing the stripe width -- all accessible from system management tools and even a command line interface if needed. In some cases these parameters are settable by the user for specific jobs, allowing centers to work with users to tune the type of I/O service they get to best serve their particular needs.
Another advantage that LSI cites is the array's ability to simultaneously support multiple RAID configurations among the trays in a system. A user might have some trays configured for RAID 10 supporting an Oracle installation, some for RAID 6, and some for RAID 5 offering protection but without sacrificing as much performance.
And while this is a new announcement, the product has already been used in the field by Oak Ridge National Labs since fall of last year, with another 51 units in acquisition by customers at major HPC facilities.
Why the emphasis on reliability with this product, especially in the face of several recent announcements by others in the storage space that were explicitly trading off reliability in favor of larger, less expensive solutions? I talked with Steve Hochberg, senior director for federal sales in the US, and Dave Ellis, director of HPC architecture, about the product.
As Hochberg explained, "We used to talk to users who would say, 'Oh, it's just scratch disk. We don't really care if it goes down, we'll just replace the disk and restart the job.'" But with dramatically increasing system sizes, and an increasing reliance on supercomputing in enterprise mission-critical applications, customers have become much more sensitive to the losses in productivity inherent in keeping a multi-thousand core HPC system idle because the filesystem is down.
One of the interesting things I found while talking to Hochberg and Ellis was the structure of the company itself. Although LSI sells its gear through OEMs, it cultivates a close relationship with HPC users to ensure that it has a direct connection to storage requirements. Hochberg described LSI's "sell to" organization, which manages sales to HPC OEMs who then brand LSI's gear as their own before plunking it on your machine room floor. This is complemented by LSI's "sell with" organization that goes in with OEMs for large customers to make sure that their user requirements make it into current and future solutions offered by the company.
Today the company is facing an uphill battle with its new product. IBM, Cray, SGI, SiCortex, Dell, and others all OEM storage gear from DDN for their large HPC compute systems. Several of these vendors also OEM LSI gear, notably SGI, but tend to target it for other applications. The 7900 is LSI's shot at a product that can make them the de facto standard as the storage engine for big compute.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?