Skip to content

How to debug webapp performance reactively?

Strategy

There are of course many strategies for finding and fixing performance issues in any application. The most common is probably

Classic performance debugging
  1. Find the bottle-neck
  2. Fix the bottle-neck
  3. Repeat
Do some initial cleanup to eliminate stupid issues

However, locating the precise bottle-neck might be difficult and/or time consuming. A more brute force approach is therefore often useful:

  1. Roughly where is the problem?
  2. Look for obvious mistakes and fix them
  3. Implement cheap, non-intrusive measures which fix common performance issues

If this solves the problem, NICE! Communicate the errors/mistakes so they are not repeated and you are done. If not, try classic performance debugging as described above.

Concrete tips on how to find the bottle-neck

Use case: Three tier Java webapp (silo) with RDBMS as backend. The webapp utilize one external webservice. The database, the webapp and the webservice are deployed on three different hosts.

Possible problems:

  • Slow web layer
  • Slow business layer
  • Slow persistence layer
  • Slow database
  • Slow webservice
  • Network

Strategy: Divide & conquer

  • Start with the web page and locate the function call that is slow. (Measurements are required yes.)
  • Partition the problem: measure time spent in webapp, by database and by webservice.
  • Investigate each in turn to check for possible (cheap) improvements.

If no simple improvements/fixes can be found, evaluate a new implementation, design or architecture.

If you suspect the network to be the problem, try measuring the time it takes to fetch a static resource. On *nix based platforms {} and {} can give an indication: 

time wget -O /dev/null --server-response http://wiki.community.objectware.no/includes/css/master.css