Investigate performance consequences of enabling/disabling parallel GC. about semantic HOT 3 OPEN

Here's some anecdata for you: at Awake we saw, roughly, a 6x slowdown because of parallel GC on one of our GRPC server components (another component exhibited similar poor performance due to parallel GC and improved dramatically once disabled).

However, not everything has benefited from disabling parallel GC and I'm sure you could tune it to behave better.

from semantic.

Hi, Can I pick this up?

from semantic.

@avirlrma: Thanks so much for your initiative! We’d definitely welcome any measurements you take, but I think the high-order bit for us will be measurements taken operating this at scale at GitHub. If you’re still interested, I’d just request that you preserve as much information about your requirements, operating environment, and load as possible so we can contrast our findings with yours effectively.

from semantic.