Parallel Platters: File Systems for HPC Clusters (Part One)
High Performance Computing (HPC) clusters are easier, and cheaper, than ever to put together. If you have an interest in pulling together your own cluster, or maybe you just want to understand more about cluster technology, it’s necessary to grok the differences between clusters and standard systems.
High Performance Computing (HPC) clusters are easier, and cheaper, than ever to put together. If you have an interest in pulling together your own cluster, or maybe you just want to understand more about cluster technology, it’s necessary to grok the differences between clusters and standard systems. One of the differences is storage requirements. While a standard server or PC can get by with a single disk and a standard file system, clusters need something a little more advanced.
In an effort to scratch the proverbial I/O itch, this multi-part article is designed as a high level survey that just touches upon file systems (and file system issues) used by clusters. My goal is to at least give you some ideas of available file system options and some links that allow you to investigate. Before I start, however, I want to discuss some of the enabling technologies for high performance parallel file systems.
Enabling Technologies for Cluster File Systems
As I discussed in a previous article Infiniband (IB) has very good performance and the price has been dropping steadily. DDR (Double Data Rate) Infiniband is now pretty much the standard for all IB systems. DDR IB has a theoretical bandwidth of 20 Gbps (Giga-bits per second), a latency less than 3 microseconds (depending upon how you measure it), an N/2 of about 110 bytes and a very high message rate. The price has dropped to an average cost of about $1,200-$1,400 per port. In addition, Myrinet 10GigE is…
Please log in to view this content.
Not Yet a Member?
Register with LinuxMagazine.com and get free access to the entire archive, including:
