Sanity checking Globus Benchmarks

I have tried to check that what I am doing yields comparable results to tests Globus people have done themselves.

In the current Tycho, R-GMA, Globus tests one of the setups varies the number of clients/consumers querying a registry with a fixed number of records. The Globus guys submitted a paper to HPDC this year which tested the same variables (although obviously with different hardware). We use 1000 records in our tests and they vary it between 1-500, they use up to 800 clients. I changed our test scripts to match their setup for 500 clients and got some results. If my test results track theirs then it adds more credibility to our testing (and proves I’m not being a retard).

I grabbed a screen shot of the Globus graph. Then I plotted my results in OpenOffice and made the graph box about the same size as the Globus graph. Finally I used Gimp to paste my graph over the top of theirs.

The graphs finish at 800 clients (which just doesn’t quite fit on this Blog page). You can click them to open them on their own.

Graph1 (Graph from Globus paper):

Graph 2 (My results on a graph about the same dimensions):

Graph 3 (Combined):