Writing up results for paper

I have completed enough of the R-GMA and Globus results to prepare a paper for Cluster 2006.

I could run a maximum 100,000 registry entries test for Globus because the MDS Index runs out of memory after 4,000 records (with 1.5GB of heap). I couldn’t run the full many clients test with R-GMA because we get a stack error after 150 clients.

We understand the Globus out of memory because it marks everything up in XML and stores it in memory - this takes lots of space.

The R-GMA error is likely a bug (and is more interesting). In R-GMA you declare the query you will make when you create your consumer. This makes sense for long lived queries (’keep telling the load every 5 minutes’) but is not good if you are going to make a series of different queries because you must destroy and recreate your consumer to run another query. In our tests we make a bunch of randomly generated queries in a loop which means we are creating and destroying 1 consumer for every query. When running 150 of these looping consumers the R-GMA registry servlet crashes with an out of stack error. Increasing the stack size does not help. I am liaising with the R-GMA people to find out what is happening, they have just sent me some test code which I will try to modify to make the bug reproducible.

The short term goal is writing all of this up for the Cluster 2006 deadline on Friday.