High Performance Computing (HPC)

Punching through the HPC Performance Barrier

Fusion's ioMemory platform has been built, from the ground-up, to minimize latency to make use of idle CPU time. This translates into faster processing with less scale-out to meet performance needs.

“This new technology will allow us to meet the performance requirements critical to fulfilling our national  security missions well into the future, while dramatically reducing power consumption and satisfying new energy conservation initiatives.”
-Mark Seager, Head of Advanced Computing Technology, Lawrence Livermore National Laboratory.

Turn Grid Servers Into Processing Powerhouses

Grid environments typically approach high-performing computing performance by scaling out commodity servers. Their distributed model works well on a small scale, but costs grow exponentially as systems grow in size. Fusion's ioMemory dramatically improves the processing efficiency of servers so that each server will handle much larger workloads, use less memory, and complete jobs faster to be more readily available.

Read how Lawrence Livermore National Labs (LLNL) used ioMemory to process a 68 Billion Node Graph with a single four-socket Intel server.

Condense Computing Clusters

Latency is key in clustered environments. ioMemory's 15-microsecond write latencies ensure that processors in clustered environments are kept busy actively processing requests instead of managing thousands of contexts and threads. In addition, unlike SSDs and non-native PCIe solid-state devices, Fusion's VSL doesn't make today's multi-core processors wait on RAID controllers for batched I/O. Instead, it offers each core direct access to terabytes NAND flash in parallel, and up to the full PCI Express bandwidth. This makes each server in the cluster capable of far more processing without ever needing to go across the network, and makes servers much more resilient to I/O bursts. It is ideal for tackling the thorny random access patterns of file system metadata (directories, allocation maps, file attributes) and small files, as well as the random access of large files, which, though rare, can lead to disproportionately large fragmentation and read access slowdown.

Read how Lawrence Livermore National Labs (LLNL) achieved 52 million IOPS and 400 GB/s aggregated bandwidth in just two racks of servers.

Slash Data Center Footprint Costs

By improving the workload capability of each server, HPC environments can reduce the costs of hardware, middleware software licenses, IT maintenance, floor space, and power and cooling. Many Fusion-io customers quickly realize a complete return on investment, giving them budget to further improve their computing capabilities. (See the Answers.com case study for an example.)

Read how Lawrence Livermore National Labs (LLNL) achieved performance levels in just two racks of appliances that would take a comparable hard disk-based solution over 54 racks.

 

 

Lawrence Livermore National Labs achieved the equivalent of 54 racks of servers in just two racks!