Nodes

From NEClusterWiki
Revision as of 03:10, 10 April 2013 by Ondrejch (talk | contribs) (node34)
Jump to navigation Jump to search

Information on Installed Nodes

There are 4 "generations" of nodes installed on the cluster. The oldest nodes are AMD dual core, the medium-old nodes are Intel Core 2 quad cores, and the latest nodes are Intel Xeon quad cores and Intel Core i7 quad cores with HyperThreading. Our last acquisition is the AMD Opteron Supernode. The specifications of each node are given in detail below.

All nodes use Fedora 18 as their operating system.

AMD Opteron supernode

Opteron supernode has 4 CPUs, 12 real cores each, that is 48 physical cores in total. This machine is to be used for highly parallel computations with huge MPI traffic. Please ask the admins if you would like to use it. The supernode has a small disk array mounted in /data. Since the user of the supernode would most likely use only that node, the user should use that filesystem for I/O.

Node Name Node CPU RAM Swap
node1 4x AMD Opteron 6234 @ 2.4 GHz 128 GB 100 GB

Intel Core i7 Nodes

Core i7s nodes are quad cores with HyperThreading. This means that they have 4 cores, but present 8 cores to the operating system. Therefore you will see these nodes having 8 CPUs when looking at them in Ganglia. Since they actually only have 4 cores, you will not get a linear increase in speed once the CPU is running more than 4 threads. However, since there still is a perceptible increase in speed, it is beneficial to use all 8 threads rather than just limiting yourself to 4.

We currently have 15 of these nodes in the cluster.

Node Name Node CPU RAM Swap
node3 Core i7 860 @ 2.80 GHz 12 GB 90 GB
node4 Core i7 860 @ 2.80 GHz 12 GB 230 GB
node5 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node6 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node7 Core i7 860 @ 2.80 GHz 8 GB 90 GB
node8 Core i7 860 @ 2.80 GHz 12 GB 90 GB
node9 Core i7 2600 @ 3.40 GHz 16 GB 184 GB
node10 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node11 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node12 Core i7 860 @ 2.80 GHz 12 GB 230 GB
node13 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node14 Core i7 860 @ 2.80 GHz 8 GB 230 GB
node15 Core i7 860 @ 2.80 GHz 8 GB 80 GB
node21 Core i7 2600 @ 3.40 GHz 16 GB 184 GB
node31 Core i7 2600 @ 3.40 GHz 16 GB 184 GB

Intel Xeon Virtual Nodes

There are five virtual nodes running inside Windows workstations. Each of them has a quad core Intel Xeon of SandyBridge architecture, identical to the Core i7s, but without HyperThreading. This machines are connected to the rest of the cluster nodes by a virtual private network (VPN), which is ~1.5 order of magnitude slower than the Gigabit interconnect used by non-virtual nodes. Therefore these machines should not be used for MPI jobs. Virtual nodes are best suited for sequential programs, OpenMP jobs, and code development.

Node Name Node CPU RAM Swap
vnode1 Xeon E31220 @ 3.10 GHz 8 GB 50 GB
vnode2 Xeon E31220 @ 3.10 GHz 8 GB 50 GB
vnode3 Xeon E31220 @ 3.10 GHz 8 GB 50 GB
vnode4 Xeon E31220 @ 3.10 GHz 8 GB 50 GB
vnode5 Xeon E31220 @ 3.10 GHz 8 GB 50 GB

Intel Core 2 Quad Nodes

These are slightly older machines. They have 4 physical cores and no HyperThreading, so when you look at them in Ganglia you will see that they have 4 CPUs. They are still reasonably fast.

We currently have 8 of these nodes in the cluster. In addition, node18 has a similar architecture, but has 2 CPUs (8 cores total).

Node Name Node CPU RAM Swap
node2 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 90 GB
node18 2x 4-core Xeon L5420 @ 2.50 GHz 8 GB 80 GB
node25 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 649 GB
node26 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 649 GB
node27 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 649 GB
node28 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 10 GB
node29 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 90 GB
node30 Core 2 Quad Q8200 @ 2.33 GHz 8 GB 649 GB
node34 Core 2 Quad Q6600 @ 2.40 GHz 8 GB 50 GB

AMD Dual Core Nodes

These are the "ancient" machines. They have two physical cores, and also no HyperThreading. If speed isn't that important you should use these nodes. They're also good at adding some speed to your MPI MCNP runs when using the balance flag. Compiling and other disk I/O tasks are also suited well to these nodes, since all nodes have the same delay when writing to the file servers.

We currently have 7 of these nodes in the cluster.

Node Name Node CPU RAM Swap
node16 Athlon 64 X2 5200+ 8 GB 184 GB
node17 Athlon 64 X2 5200+ 8 GB 184 GB
node19 Athlon 64 X2 5200+ 8 GB 10 GB
node20 Athlon 64 X2 5200+ 8 GB 184 GB
node22 Athlon 64 X2 5200+ 8 GB 10 GB
node23 Athlon 64 X2 5200+ 8 GB 10 GB
node24 Athlon 64 X2 5200+ 8 GB 10 GB