About admin


optimizing spot instance allocation (part deux)

See part 1 for context. This post explains the control/planner strategy I settled on to minimize the cost and make things as stable as possible to mitigate disruptions caused by VMs being re-allocated. Continue reading

optimizing spot instance allocation

There is surprisingly little information on how to optimize costs using the AWS spot instance market. There are several services that provide management on top of the spot market if you have an architecture that supports an interruptible workload but very little in the way of how to go about doing it yourself other than surface level advice on setting up autoscaling groups. To remedy the situation here’s an outline of how I’ve solved part of the problem for CI (continuous integration) type of workload Continue reading

learn some math

This is in response to the math myth.

The author couldn’t be more wrong even if he tried. My current project is using GLPK to do some basic mixed integer programming to optimize AWS spot instance allocation. If I had not taken linear algebra, calculus, and a few courses in linear programming the idea would not even have crossed my mind that I could use mixed integer programming to solve the spot allocation problem and that’s just the first half.

The second half can be considered a problem in control theory because it requires taking the new allocations and gracefully transitioning from the old set of allocations. You can go even further and say that the whole thing would be even better if I understood more about stochastic processes and could potentially model the spot market and make predictions ahead of time to simplify the control problem and get ahead of the price fluctuations.

Saying all you need is Excel and 8th grade is in the words of one famous physicist “not even wrong”. If you’re in an engineering discipline then the more math you know the better.