As described in #4490, CMS needs a wide-area glidein pool with 200,000 cores. We have a test setup that simulates a large pool to see how we can scale to that level. We use only a fraction of the normal hardware for execute nodes by running many startds per core and running sleeps jobs that don't consume CPU or memory.

This page documents both the HTCondor configuration (i.e. changes to the default RPM) as well as any Linux kernel parameter tuning required, divided into three sections: configs/tunings on the central manager, configs/tunings required on the submit nodes, and configs/tunings on the execute machines.



Central Manager

The central managers are running on HA mode, and They have 128GB of memory and 32cores. they have collectors trees with 200 child collectors.

Central Manager Configuration:

Submit Machines

These are the submit machines:

Execute Machines