# What method should I use to model simplified performance

I have the data of some build times. These builds occur on a cloud and the runtime has been more varied than in the past. One thing that could be occurring is an improper sharing of resources between clients. To test this, I've downloading the dataset containing

Start Time
Agent ID
Length of Run


I want to test to see if when there are more machines running, the builds get slower. What method should I use to do this?

• I don't see how you can use these daya to discern how many machines are running. Sep 26 '14 at 16:11
• Imagine two agents, AgentA AgentB. AgentA starts at noon. AgentB starts at 1:00 pm. If I know they both ran for 2 hours each, then I can calculate what time periods they ran together. Sep 26 '14 at 17:51

Based on your answer to my question, I think you can construct an additional variable nComp in the dataset equal to the number of competitors associated with each observation. You probably should throw out the first few observations because you can expect there were competing agents active at the beginning that are not accounted for.
Then you can construct a scatterplot of LengthOfRun versus nComp. If it looks reasonably linear, then a simple linear regression should help you make an inference about the effect of number of competitors. If it's nonlinear, or the variability changes, you need a fancier regression model.
In addition, you could plot nComp versus other variables -- e.g. vs. StartTime to see if there is a trend.