Linux Cluster Hardware

Hardware description of euler.phys.cmu.edu, the small linux cluster operated by the Widom research group in the CMU Physics Department.

Networking

The head node is named euler.phys.cmu.edu, with a webserver alias alloy.phys.cmu.edu hosting the alloy database. The file server is eulers.phys.cmu.edu. Authentication is via internal user names and passwords and via the CMU Andrew system. All nodes are connected internally by a 10 gigabit network switch although each node may be limited by its own network interface speed if lower. Nodes euler and eulers are connected to the Wean Hall public network.

Nodes

NameChipNCPU x Ncore CPU (GHz)FSB (MHz) L2 cache (MB)Mem (GB)Disk (GB)
euler.physXeon Gold 51222 x 43.6 26661796240 SSD
eulers.phys (old)Xeon E5-26302 x 62.3 13331532500+23000
eulers.phys (new)Xeon Silver 43141 x 162.4 320016128960 SSD+51000+50000
n01.lc.physXeon X56752 x 63.06 13331248500
n02.lc.physXeon X56752 x 63.06 13331248500
n03.lc.physXeon X56752 x 63.06 13331248500
n04.lc.physXeon X56752 x 63.06 13331248500
n05.lc.physXeon X56752 x 63.06 13331248500
n06.lc.physXeon X56752 x 63.06 13331248500
n07.lc.physXeon E5-26702 x 82.60 16002064500
n08.lc.physXeon E5-26702 x 82.60 16002064500
n09.lc.physXeon E5-26702 x 82.60 16002064500
n10.lc.physXeon E5-26702 x 82.60 16002064500
n11.lc.physXeon E5-2680v32 x 122.50 213330641000
n12.lc.physXeon E5-2650v42 x 122.20 240030128240 SSD
n13.lc.physXeon E5-2650v42 x 122.20 240030128240 SSD
n14.lc.physXeon Gold 5218R2 x 202.10 266727.596500 SSD
n15.lc.physXeon Gold 5218R2 x 202.10 266727.596500 SSD
n16.lc.physXeon Gold 5218R2 x 202.10 266727.596500 SSD
g01.lc.physXeon E5-2660v32 x 102.60 21332564500
g02.lc.physXeon E5-2650v22 x 82.60 18662064250
Total 21 380 137810220+24000

GPU

g01 has one Tesla K80 for running GPU-accelerated VASP. g02 has two Tesla K40 GPU's.

Storage

The file server eulers has 51TB on a RAID array serving the /home directories on the cluster. The file server runs a backup program tob, backing up to the head node euler. Usage is described in /home/alloy.db/HOWTO.restore. Each compute node has temporary free disk space in /tmp.