Showing posts with label Grid computing. Show all posts
Showing posts with label Grid computing. Show all posts

Monday, July 13, 2009

A way to stress-test GUI

What to do if you need to test the performance of your client-server application, which is not on the Web? For example, some kind of Lotus Notes one. The problem is - you can't even execute several Lotus instances on a single machine, so it's quite a tricky task to simulate multiple simultaneous users.

Virtualization is what you can use in this case. There's an example of working stress testing system consisting of Citrix XenApp terminal server running 10 Lotus Notes instances (eating 50 Mb of RAM each), and a load generator running IBM Rational Performance Tester (which is just a very advanced point-and-click thing), which can safely simulate 20 - 40 concurrent users.

Major bottlenecks of this setup are:
  1. RAM on Terminal Server (it's better to use 64 bit solution)
  2. Network bandwidth (at least 1 Gbit Ethernet)
  3. It takes much longer (up to 4 times) to implement and debug such test cases, compared to usual web testing scenarios

Wednesday, February 18, 2009

Grid Computing made easy with GridGain

I've just watched a 20 minute demo of GridGain usage for parallel computations. Looks quite impressive, at least easy due to Map-Reduce-like approach employed in a very simple form. Also it has a nice feature of transparent propagation of code changes between nodes, still it's not actually clear for me how they gonna manage changes in classpath, libraries, etc.

It seems that GridGain is not very popular (yet), at least I was unable to find a wiki article about it :) though I really sympathize its simplicity and wish them good luck.