1 Answers
Hi Learner,
You are right and I use this strategy ALL the time with non-critical workloads. Because I’m not using volatile instances types like the big ones or the GPU ones, I usually realize about a 70-80% savings over On-Demand. The risk is that I have seen some cases where Spot prices will move higher than On-Demand prices…it might happen in one AZ or all of them. It’s rare but it does happen. I usually set me Spot quite a bit higher than on-demand just as a safety margin. AWS lets you set a spot price up to 10x the on-demand price.
The second thing that might happen is capacity constraints in an AZ or Region. There is a set portion of on-demand ear-marked resources set aside to ensure that on-demand people get the resources they need when then demand them. Spot resources are a lower tier and can loose out on starting or restarting if there is a constraint. This has happened to me in the past with crowded regions on relatively common instances like the M, C and T instance types.
Overall, I like the strategy for disposable instances but not for production stuff.
–Scott
Thanks Scott! The explanation helps!