User`s guide

2 007-5474-003
1: SGI Altix XE Cluster Quick-reference
instance, one application may run on 16 processors in the cluster while another application runs
on a different set of 8 processors. Very large clusters may run dozens of separate, independent
applications at the same time.
Important: In a cluster using older and newer compute nodes (i.e. XE310, XE320 and XE340
nodes) parallel calculations will be executed at the rate of the slowest node. If a cluster consists of
multiple older and newer compute nodes, the job scheduler would try to select a set of identical
nodes on which to run a calculation.
Typically, each process of an MPI job runs exclusively on a processor. Multiple processes can
share a single processor, through standard Linux context switching, but this can have a significant
effect on application performance. A parallel program can only finish when all of its sub-processes
have finished. If one process is delayed because it is sharing a processor and memory with another
application, then the entire parallel program is delayed. This gets slightly more complicated when
systems have multiple processors (and/or multiple cores) that share memory, but the basic rule is
that a process is run on a dedicated processor core.
These are the primary hardware component types in the rackmounted cluster:
• Head node(s)
(SGI Altix XE250 or XE270 2U-high servers)
• Compute nodes
(SGI Altix XE320 or XE340 1U-high servers)
(SGI Altix XE250 or XE270 2U-high servers)
• Network interconnect components
(Gigabit Ethernet switches, InfiniBand switches, PCI cards, and cables)
The head node is connected to the interconnect network and also to the “outside world”, typically
via the local area network (LAN). The head node is the point of submittal for all MPI application
runs in the cluster. An MPI job is started from the head node and the sub-processes are distributed
to the cluster compute nodes from the head node. The main process on the head node will wait for
the sub-processes to finish. For large clusters or clusters that run many MPI jobs, multiple head
nodes may be used to distribute the load.
The compute nodes are identical or compatible computing systems that run the primary processes
of MPI applications. These compute nodes connect to each other via the interconnect network.
The network interconnect components are typically Gigabit Ethernet or InfiniBand. The MPI
messages are passed across this network between the processes. This compute node network does