I have an application that must run on Windows. This application creates data models on the order of ~2 Billion calculations per model. Today a 933Mhz machine can process roughly 50 of these per day. I have been tasked to build a cluster of 20 high end (3.0Ghz) PC's to distribute and share the processing power. NOTE: This application and its throughput is solely dependent on clock cycles, memory and server architectures provide no speed increase.
What I need to do is have some sort of scheduling or queueing with a master-slave relationship. I want the master to watch and know when a model is finished on one of the slave boxes, so it can start another one.
I have looked at the Globus Toolkit and Condor, and I've even looked at basic Windows clustering and W2k Enterprise Server. These all look like they could do it with a good bit of customization. Does anyone know the best way for me to proceed?
Thanks.
What I need to do is have some sort of scheduling or queueing with a master-slave relationship. I want the master to watch and know when a model is finished on one of the slave boxes, so it can start another one.
I have looked at the Globus Toolkit and Condor, and I've even looked at basic Windows clustering and W2k Enterprise Server. These all look like they could do it with a good bit of customization. Does anyone know the best way for me to proceed?
Thanks.
