deletions | additions       diff --git a/Queueing.tex b/Queueing.tex  index 2d16cf1..cd1bbd2 100644  --- a/Queueing.tex  +++ b/Queueing.tex 
...
The problem is effectively 3 separate \href{http://en.wikipedia.org/wiki/M/G/k_queue}{M/G/c queues}. This makes the problem of calculating the waiting time (given the $\lambda(t)$ we calculated and the distribution $G$ from the data) analytically impossible. However, there are approximations which we could use, but we would have to figure out which ones work best, if at all, (especially considering papers like \href{http://link.springer.com/article/10.1007%2Fs11134-009-9133-x#page-1}{this one} which describe the issues with approximating M/G/c waiting times). Perhaps it's a good idea to check our approximations against simulations.  The Prezi system automatically  sends jobs to the VM which is closest to the end of a billing period, but that still has enough time to run the job. job, so we don't have to worry about that.