The QB3 Shared Cluster is a co-operative resource available to all UCSF QB3 faculty members. It was founded by QB3 PIs, from their own funds, primarily for the benefit of QB3 researchers. The cluster provides a very large number of CPU cores (currently over 4300) and temporary space on a high performance storage device (currently approximately 12TB), all managed by a full time systems administrator. Please note that the cluster is not a core facility. It receives no guaranteed institutional support, but relies on user contributions for all costs. Since the cluster's beginnings in late 2004, these contributions total over $1.8M in addition to a large seed hardware component. The cluster's annually recurring costs include salary and benefits for the systems administrator and support contracts for the storage and network hardware, totaling approximately $170,000. In addition, new nodes are added to the cluster on roughly an annual basis, depending, of course, on the level of support received from the userbase. In exchange for these contributions, users are granted shares of the cluster that translate into priority access to the computing power.
To serve the greater UCSF community, guest accounts on the shared cluster are available. These accounts allow for the use of up to 100 CPU cores when these cores would otherwise be idle. Guests are, of course, welcome and encouraged to contribute to the cluster and gain full access.
For details on the hardware available to users, please see Cluster Topology. While this wiki provides basic instructions on operating in the cluster's Linux environment, please note that no application support is available to cluster users. All users are subscribed to a mailing list and are encouraged to discuss any application issues they encounter there.
The cluster is managed by the steering committee, consisting of Andrej Sali (Chair), Tom Ferrin, Sharmila Majumdar, and Brian Shoichet.
Job Scheduling and Contributions
The shared cluster uses Grid Engine to schedule and run jobs. Priority is given to users in labs which have contributed to the cluster. For full details on how this works, please see Cluster Usage. In brief, labs which have contributed to the cluster are guaranteed immediate access to a number of CPU cores proportional to their contribution. The remaining cluster capacity is available to all participating QB3 labs (both contributing and non-contributing). A history of cluster usage can be seen here.
Estimating the Cluster Share Resulting from a Contribution
Full details on how cluster share is calculated and the complete list of labs which have contributed to the cluster and their resulting shares are available on this page. If you are contemplating contributing cash to the shared cluster, you can estimate your resulting share in the cluster and guaranteed CPU cores by doing the following:
- First, calculate the number of standard Processor Units (PUs) by determing how many "Standard Nodes" the contribution would buy:
- PUs = ($(Amount of contribution) / (4673 * 1.095)) * 24
- Next, calculate what fraction of the total number of PUs this represents:
- Share = PUs above / (11001 + PUs above)
- Finally, multiply this share by the number of CPU cores in the cluster:
- Guaranteed CPU cores = Share * 4326
Please note that there is no minimum contribution -- every little bit helps! Hardware donations (of relatively recent vintage, please) are also gladly accepted. Such donations will be benchmarked against the current cluster hardware in order to assign shares.