Supercomputing is a general term that encompasses any high-speed computational process, and whose definition changes as new computing methods are developed. As per today, there are three areas of interest in supercomputing at the IAC: distributed, parallel and grid computing; although interrelated, there are some slight distinctions between the terms.
The IAC also has several high-performance Linux machines, open to anyone, where CPU- or memory-intensive jobs can be run.
For the latest hardware and software upgrades, service interruptions, etc. see the Supercomputing News
Parallel computing is the simultaneous execution of the same task (split up and specially adapted) on multiple processors in order to obtain faster results (see also: Parallel computing by Wikipedia). It takes place entirely within one computer and adding additional processors allows for tasks to be performed simultaneously thus increasing computing power and speed. At present, there is only one Beowulf-type clusters at the IAC:
Diodo, the 100-cpu beowulf cluster installed here in the IAC headquarters, was retired from service at the end of 2012 after many hardware failures.
Distributed computing is the process of running a single computational task on more than one distinct computer (see also: Distributed computing by Wikipedia ). It is usually considered that distributed computing takes place on a single network, while grid computing (see below) can utilize the processing power of machines outside of a defined network. At the IAC the Condor system is installed (working, but still in testing phase), which allows for distributed opportunistic computing. Condor is an scavenging system, which allows you to make use of idle machines across the IAC to run your programs. For more information see the Condor system page at the IAC.
Grid computing offers a model for solving massive computational problems using large numbers of computers arranged as clusters embedded in a distributed telecommunications infrastructure. Grid computing's focus on the ability to support computation across administrative domains sets it apart from traditional distributed computing. Grid computing has the design goal of solving problems too big for any single supercomputer, whilst retaining the flexibility to work on multiple smaller problems. Thus grid computing provides a multi-user environment (see also: Grid computing by Wikipedia). If you are interested in Grid Computing, please join the Grid Mailing list at the IAC, in order to discuss future developments, suggestions, etc.
Users who need to run CPU- or memory-intensive jobs, which are unsuitable for either parallel processing or for the Condor system, can access any of several high performance Linux PCs. These machines are open to any user and do not require advanced reservation.
| Name | Location | CPU | RAM | HD space | OS |
|---|---|---|---|---|---|
| denso | Remote access | 2 x AMD Dual core 2.0 GHz | 16 GB | ~130 GB | Fedora 17 64bit |
| dardo | Remote access | 2 x AMD Quad Core 2.4 GHz | 64 GB | ~270 GB | Fedora 17 64bit |
To check if any other high-performance machines are available, you can use
the following Condor commands:
condor_status -constraint 'TotalCpus >= 4' #
PCs with 4 or more CPUs
condor_status -constraint 'Memory > 4000' #
PCs with 4GB or more RAM per CPU
condor_status -constraint 'TotalCpus > 4 && Memory > 4000' -sort
KFlops # sort by CPU speed.