Scale Performance Metrics

This page is for collecting Scale performance metrics. These metrics do not reflect Scale's total throughput, but rather the throughput through a small (2) number of nodes.

How to perform a performance test in the Scale team's Kiln environment:

1. Start with an empty cluster with a new database
2. Make you sure you have 2 empty nodes (no DC/OS tasks or Scale jobs) with 4 CPUs each (pause all other nodes)
3. Pause the scheduler
4. Queue up 5,000 Scale Hello jobs
5. Unpause the scheduler (useful to note the time as you do so) and allow jobs to run to completion
6. Record how many of the Scale Hello jobs completed in the first 15 minutes

Previous Tests:

05/16/2017 - Issue 841 (Test Jobs) - 1,618 jobs in 15 min - 1.8 jobs/second

05/16/2017 - Issue 841 (Test Jobs) - 1,567 jobs in 15 min - 1.7 jobs/second

05/17/2017 - Issue 841 (Test Jobs) - 1,932 jobs in 15 min - 2.1 jobs/second

05/17/2017 - Issue 845 (Queue Improvement) - 1,790 jobs in 15 min - 2.0 jobs/second

06/12/2017 - Issue 852 (Refactor Scheduler) - 2,123 jobs in 15 min - 2.4 jobs/second

09/06/2017 - Issue 928 (job_exe refactor) - 2,227 jobs in 15 min - 2.5 jobs/second

12/20/2017 - Issue 1013 (Job Completion Message) - 3,055 jobs in 15 min - 3.4 jobs/second