PPPL Cluster Hardware
The PPPL cluster consists of several sub-clusters.
kestrel - the main cluster.
This group of 240+ systems, linked via a low latency gigabit Ethernet,
provides the
majority of PPPL's scientific computing hardware. Each system
contains AMD Opteron CPUs (dual and quadcore), and at least 2GB of
memory per core.
kite - a larger
memory cluster with a very high speed (10Gb) Infiniband low latency
interconnect. This group of 24 systems, each with two AMD Opteron
CPUs and 6 GB of memory.
sque - a pool of systems
reserved for single CPU jobs. These systems each contain two CPUs and
between 2 and 4 GB of RAM.
ganesh - a pool of systems
reserved for large memory jobs. These systems each contain four CPUs
and
between 16 and 32 GB of RAM.
kruskal - a low latency
cluster of 80 systems, each with 8 CPU cores and 16 GB of memory,
linked via an Infiniband interconnect. The cluster is segmented into a
limited access 'large' partition of 512 cores, and a smaller open
access partition of 128 cores.
The PPPL cluster also has two large memory systems:
mhd - this is an SGI Symetrical Multiprocessing (SMP) system
with Itanium 2 processors and 192 GB of memory. It uses the NUMAlink
architecture to allow any process to address any or all memory in the
system. Its use is deprecated in favor of the stix system.
stix - this is a large
memory Symetrical MultiProcessing (SMP) system with 80 CPUs and
440 GB of memory. The CPUs are 3.0GHz Xeon quadcore. As an SMP system,
a user can allocate very large amounts of memory from a single process,
allowing codes to run with part or all the memory on the system.
