Its a really hard question to answer, especially when you’re creating something new.

After making some changes to become more “compliant” with Facebooks TOS I decided that I’d re-launch our “Who isnt my friend?” (now called “Friend-Watch“) application by simply adding it to my wall, and then doing nothing. Anyone who has followed this blog in the past knows that this app took off all by itself in the past, so it shouldnt be any surprise that I’d like to see if that happens again (if i do nothing to help it).

That was about 4 days ago now, so today I decided to measure my performance. In order to do this I created the following graphs:

• Engagement stats
  

• System performance

The Engatement stats are interesting because without me doing anything, they’re all pointed in the right direction…so i guess we can watch this one for a while longer and see what happens!

What interests me the most is the “System performance” graph. I was trying to answer the question “how do i know if i need more hardware?”. Previously when i wrote this app I had a massive issue with scale. I think that issue is pretty much resolved, but now i need to ask myself “how do i know when i need to scale it?” Thats where this graph came from.

I’ve drawn 3 key details on the “System performance” graph:

• Total audits – This shows how many audits the system has done as to date
• Total audits on day – This shows how many audits the system did on this day
• Total audits per user * 100 – This shows how many times each user was audited on average, multiplied by 100 to blow up any issues so they’re more easily seen

At first i just implemented “total audits” and “total audits on day”. I made the assumption that when i hit the servers limit i would see the “total audits on day” flat line. While i dont think this is an invalid assumption, it would take a few days for me to notice this…not cool. Then i had a better idea, i should measure how many times each user has been audited.

In general the way this application works is pretty simple. We check each users account once every 7 hours (ideally…) based on a queuing system. There are two “worker” processes, so I expect that ideally we should see around 10-12 audits per user per day (which is something like whats happening in reality), but whats most important from a “system health” type situation is that whatever the current level of “audits per user * 100″ is doesnt change too much from the previous day.

I’m sure there are a lot more details we can abstract out, but I just thought this was interesting and worthy of writing. Enjoy