Hi, for my purposes I am using AWS Batch with EC2 spot to run multiple jobs created every 15 minutes with each lasting ~2-10 seconds for computation.
From my understanding AWS Batch creates an instance for each job, where Spot prices a minimum of one minute's charge for each, although I am uncertain of the exact instance type chosen by the "optimal" selection's logic with best-fit.
As a result of my low duration workload, I was planning to create a load-balancing system that combines jobs (Each is calculated independently and are not easily multi-processed) before submitting them to AWS Batch, choosing a specific instance type instead of the "optimal" selection to reach the 1 minute time for a given instance consistently to avoid wasted time (However I still want the work to be computed in parallel over multiple instances whenever otherwise possible).
I was wondering if there were any faults in my understand of AWS Batch and Spot above, perhaps where AWS Batch packages the jobs similarly to how I was planning?
AWS 官方已更新 3 年前