2 ms·
Having just finished benchmarking a new project I'm writing on nearly all EC2 instance types I have to say nothing beats benchmarking. Create a spreadsheet, wr
by rdoherty 13y ago
Having just finished benchmarking a new project I'm writing on nearly all EC2 instance types I have to say nothing beats benchmarking.
Create a spreadsheet, write some benchmarking tools, make sure you aren't actually benchmarking another service the instance depends on (db, API, etc) and go to town. Vary # of request/sec, # of parallel requests, # of app workers/threads/etc.
Write down all your results and then compute amount of 'work' (reqs/sec, computations, db inserts/updates, etc) vs hourly or RI pricing. That should get you some good numbers on what instance type is most cost effective for your workload.
- thejosh 13y agoOr just the machine you landed on. EC2 hosts vary wildy in terms of performance.
- 1SaltwaterC 13y agoPer CloudWatch stats, it varies for machine and region. We have a 6 machines array in three regions receiving round robin traffic from ELB's. It is network bound. However, the CPU graphs say otherwise. Here's the output of three machines in three regions: http://i.imgur.com/nadGmDH.png http://i.imgur.com/nadGmDH.png while their siblings follow the same trend. The working set resides in RAM. The disk and network follow the same path. These are instance store machines sharing as little as possible. Failing over a multi-AZ RDS for maintenance reasons did not change the graphs.
- qpleple 13y agoCan you tell us what tool you used? Or maybe share your results/conclusions?