Hi Team,
Any tips for optimizing these values. I'm thinking the defaults of 100 and 200 are probably to low for my server but imagine there might be a sweet spot maximum value before scheduler performance starts degrading.

To put this into context I'm trying to test with tens of thousands of clients with WU that take 70 seconds to process and the scheduler is struggling often returning zero work units to clients at high loads.

I've already tune other aspects of my server around number of validators/assimilators/deleters, fastcgi, winding down logging, hosting files on a different server (S3 with CloudFront on AWS), plenty of iops on my DB and large DB host.

Any performance tips appreciated.

Cheers
Seth

Hi Seth,
can you elaborate what values you want to tweak? My guess is you want to extend the shared memory array that keeps the work in RAM because your feeder is not fast enough for the scheduler. That is usually not a problem and only limited by the amount of RAM your server has. I've seen setups with 500 or even 800 slots. That should help in your case where there is a high throughput in the system. Just increase it and find the right spot for your project.

Hi Christian,
Just finished a great performance run where 2,000 seemed to be the sweet spot for me. It was choking at 1,000 mainly as new clients were spinning up as they tend to download more work at their startup.

I had around 23K vCPU running at those settings and begun to be throttled by my DB (need a bigger one!).

It's good!

Cheers
Seth