How to minimize system memory for a computational grid of linux servers

I have an application that is running on a total of 7 linux (fedora 3) servers with 2gig of ram each. The process will be run for seven days at a time and will be reading from the db at startup time (about 500 files with around 1 gig of data) and will then cache that in local data structures. The process alos allocates and releases a lot of memory and is written in python.

The process gets very slow after about 25 minutes of churning. When I check the free memory it is down to about 20meg and I thought the problem might be that. What is interesting is that the process is only using about 600 meg of memory, 1gig of it is is being used by the linux cache. I turned the swap off using the command swapoff -a and that releazed some memory. But how do I get the memory back from the linux cache which is using about 1 gig of memory. This memory is not let go even after I kill the process.

I've spend a good part of day reading articles about this. I came acorss some people who patched the kernal to set a max cache parameter, but I can't believe that there is nothing out of the box.

I don't want any of the files cached in memory and using my very precious memory since I'm already storing the data in my application.

Any suggestions on how I can get my memory back!!