How to Fix Performance Problems

• Performance of a system can easily be improved, with a little planning and experience/skills

  • But there is a asymptotic aspect of improving performance, you eventually get less bang (small % increase in performance metric) to the time spend doing so

  • Thus avoid major spending too much time optimising, unless it is required (SLA)

• Under time-to-market pressure, it is both wise and effective to choose a solution that gets the job done simply and quickly, but less efficiently than some other solution

  • performance is a part of usability, and often it must eventually be considered more carefully.

• Be aware when optimising, locally it might be slower on your machines compared to production

  • The benchmarks used or how they are collated might not be accurate

  • The compiler might do it's own optimisations after several repetitions

  • Just measuring, will influence the performance too

How

• Need to analyse the system

  • Need metrics

  • Need evidence

  • Need to be able to replay this area of optimisation, to test your fix works

• Need to find the bottlenecks

  • places where most of the resources are consumed

  • No need to focus on optimising anything if it counts for a small % of computation time.

  • Focus on the big rocks first, the easy wins

• To do this, you will need tools, llike a profiler or other apps

  • logs will also help too

• Focus on things that will improve the system at least 2x for that metric

  • Or some big factor

• Any change that is made, will require testing, which needs to be factored in

  • Is the change to hard to test (or measure)?

• Well designed systems, especially if modular will contain the performance issue in the module (segratated section) and can just focus on that

  • May allow you to redesign that module instead, as better allocatio of time

• Might be the case, that all the easy wins do not help attian your performance goal

  • This can lead to doing the hard and time consuming stuff

  • It might be better to redesign system in this case

    • Less time to improve system and/or much better performance improvements

Examples

• Using a database connection pool, instead of connecting to database everytime per call

• Using indexes in databases for common column looks up

• Using caching for non changing data (or hardly changing data)

• unnecessary I/O in inner loops

• leaving in debugging statements that are no longer needed

• Too many logs

• unnecessary memory allocation

• inexpert use of libraries