CVMFS on clusters

LHC CVMFS examples contains information. The recommendation for clients 2.1 is _minimum_ 40 gigabytes of node-local cache.

Another source for information: https://www.gridpp.ac.uk/wiki/CVMFS_configuration_at_RAL

Each cluster should also have a HTTP proxy running, preferably Squid (+the AliMon cvmfs monitor). This machine needs ~20 gigabytes of memory, but 100 gigabytes of disk is enough. All files will be read from the cache anyway.

cvmfs-5.ndgf.org

cvmfs-5 has opened up the proxy to the jade cluster using DDR infiniband, allowing it to serve files at high speed.