The different approaches to build colonies of Linux workstations into high-performance compute clusters is discussed. The first approach is to export the local file system and mount it on every node in the cluster. The second approach is to put a fiber channel adapter in every node and have each write to a common data pool while a shared file system manages the locking. A different approach is to use a parallel file system where clients stripe data over multiple servers that each have a direct-attached disk. The high-performance compute clusters that use less local storage will reduce the number of moving parts in the average compute node and will lead to a more resilient computer in the long run.
|Original language||English (US)|
|Number of pages||1|
|Journal||Scientific Computing and Instrumentation|
|State||Published - Jun 1 2004|