High-Performance Computing in the Accelerator Systems Division at APS

APS has a cluster of over 200 Sun workstations, with a common server and operating system.
Use Distributed Queueing System (DQS) to manage 24 of ASD's fastest workstations.
When combined with ASD-developed software tools, this provides a powerful concurrent computing environment.

Concurrent Computing at APS
--Problems and Solutions--

How to use resources that are distributed around a building or lab?

DQS provides load-based queueing of jobs to heterogeneous networked workstations.
It incorporates features that reduce the impact on interactive users.
It supports both concurrent and parallel computing

Concurrent Computing at APS
--Problems and Solutions--

How to prepare 100's or 1000's of input files?

Use scripts (simple programs) to prepare specific simulation input files from templates.
Each input file might have, e.g., a different value for a parameter or a different seed.
Typically this is a simple matter of text substitution.
Decide on a method for making locally unique root filenames for simulations. (You can't call them all FOR006.DAT.)

Concurrent Computing at APS
--Problems and Solutions--

How to process 100's or 1000's of output files?

Trival but vital point: Use the same root filename for all input and output for a run.
Use codes that are compatible with an automated postprocessing system.
We use APS-developed SDDS (Self-Describing Data Sets), a group of ~70 generic programs using a common data protocol.
Use scripts to combine the SDDS modules into data processing algorithms.
The number of simulations in a set is arbitrary when this approach is used.

Computing Activities Depending on the Concurrent Approach

Top-up safety simulations.
"Proves" the safety of a proposed new APS operating mode by simulating ~3000 fault scenarios and configurations.
Simulation of collective effects (impedance, IBS) in storage rings.
Understanding SURF sawtooth instability.
Testing 4th generation concepts.
Finding APS impedance model.
Light source design validation.
Able to simulate large numbers of randomly perturbed machines for dynamic apertures, etc.

Other Intensive Computing Activities and Interests at APS

Automated optimization of a new rf gun design using a PIC code. (Presently done using PARMELA.)
SASE FEL simulations.
Lattice calibration of the APS ring.