HP Aiming For Faster, Cheaper Linux Clusters

HP has upgraded its storage file system software to
transfer and store more data at a faster rate on Linux clusters.

The product, HP StorageWorks Scalable File Share (HP SFS), eases bandwidth
burdens by distributing files simultaneously across server and storage
clusters.

SFS can span dozens to thousands of clustered Linux servers,
making it
ideal for distributed applications in industries such as life sciences,
financial services and digital animation.

HP SFS 2.0 is even more powerful. The software shuttles data up to three times
the bandwidth (or 35 gigabytes per second), than version 1.0, said Kent
Koeninger, product manager for HP’s high performance computing and
technology group.

It also handles double the capacity at 512 terabytes and
now uses Voltaire’s InfiniBand interconnect technology, as well as gigabit
Ethernet, to transfer data across machines.

Koeninger said the Palo Alto, Calif., company has cut the starter costs for
SFS 2.0 in half, to $44,000, to help smaller enterprises afford large-scale
file sharing utilities. HP was able to do this by cutting the number of
ProLiant or Integrity and StorageWorks disk arrays the system needs to
perform.

“We were able to reduce what it costs to get into the system by half,”
Koeninger said. “We found [the original list price of $80,000 to $90,000] to
be too high a threshold for people to kick the tires.”

HP SFS 2.0 runs Red Hat Enterprise Linux versions 3 and 4 Fedora; SuSE Linux
Enterprise Server version 9, among others. The product will be available for
purchase in July.

The new performance perks are important at a time when compliance
regulations order corporations to save and recall files.

Most global file systems cannot support today’s computing demands because
many Linux clusters use older input/output data transfer techniques, such as
Network File System . SFS is built to leave traditional file
systems from IBM and Sun behind, according to Koeninger. Start-ups like
Panasas and Ibrix make competing object-based cluster file systems.

The first version of SFS was launched
around this time last year. The software is based on HP’s StorageWorks Grid
architecture, which is one of the reasons why it is able to zip data along: The
software scans three billion files in three seconds.

StorageWorks Grid divides storage, indexing, search and retrieval tasks
across a set of computing nodes or storage “smart cells.” Each smart cell
acts as a storage server that works in parallel with other smart cells on
the grid, delivering info
in real time.

HP SFS customer Shared Hierarchical Academic Research Computing Network, a
Canadian academic research network, recently purchased four new high
performance computing (HPC) clusters as part of a $20 million contract. The
group integrated 480 terabytes in HP SFS systems.