Monday 27 February 2012

How many upgrade tests should I run

This is a common conundrum for clients and technical consultants that are running the upgrade process.  How many times should they perform the upgrade before they can be confident that it’s going to work properly on the “go-live” weekend.  The true answer is, it depends…   Complexity and data volume are huge factors when trying to determine this number.

After the initial “upgrade” that is going to bring across code and data, you generally fine tune the upgrade process to upgrade the least amount of items properly.  My preference for an upgrade is a strictly “data only” approach.  This means that there are less moving parts and less to go wrong on the go-live weekend.  I generally prefer to execute this conversion at least 3 times to be confident that things are going to work well over the go-live weekend.

At the end of the 3 times, I’ll understand the following:

  • timing (so I know if something is going wrong)
  • results and how to ensure that things have worked (table counts, row counts, index counts)
  • when I can get some rest and when I need to pay attention to the process (timing again)
  • Documentation and general familiarity
  • Backup and restore points
  • Disk and CPU usage patterns and what is normal

This being said, I’ve been on projects that have a super tight window for the execution of the upgrade routines and this means that the the process has been run about 8 times.  I suggest between 3 & 8 executions.

No comments: