Michigan Ross iMpact

Leading in Thought & Action

Grid Description

Grid Environment

Our Grid environment consists of 6 public servers; RC1-RC5 & RCX (which serves as the headnode), 3 private Faculty servers (RCP1-3) and a file server (RC-FILE). Most are fairly recent Dell servers. The characteristics of the servers is as follows

RCX: 2-CPU 2-core w/ 16GB RAM ~143GB scratch disk space Redhat ES4.2

RC1: 2-CPU 6-core w/ 48GB RAM ~388GB scratch disk space Redhat ES4.2

RC2, RC3: 2-CPU 4-core w/ 32GB RAM ~245GB scratch disk space Redhat ES4.2

RC4: 1-CPU 2-core w/ 8GB RAM ~202GB scratch disk space Redhat ES4.2

RC5: 2-CPU 2-core w/ 8GB RAM ~120GB scratch disk space Redhat ES6.4

RCP1, RCP2: 2-CPU 2-core w/ 24GB RAM ~256GB scratch disk space Redhat ES4.2

RCP3: 2-CPU 6-core w/ 48GB RAM ~900GB scratch disk space Redhat ES4.2

RC-FILE: 16-CPU 4-core w/ 32GB RAM ~11.4TB user disk space

The disks on the servers other than RCFILE are for the OS plus scratch space. All user file space is on RCFILE and is attached to the various servers via NFS. The user file system uses RAID for disk failure robustness and is incorporated into CS's enterprise backup system.

The private servers represent systems purchased by individual faculty members for their own use (plus the occasional RA). These can be added to the Grid on an "as needed" basis and are financed by the individual faculty member.

The Grid is designed to load-share jobs across all servers. The Platform LSF software is used to accomplish the load sharing. Clients log into the Headnode, RCX, and initiate their jobs from there. Where those jobs ultimately execute are either on (a) the Public node with the smallest load or (b) the private node associated with the user. LSF has extensive queuing and routing capabilities built in.

Third-Party Applications

Applications installed include SAS, Stata, Matlab, Gauss, Mathematica, R and Splus.