scylladb
feature(gemini): add PGO optimized build when releasing
how do you test it?
Really hard to test it. Only way is to do the release
what do we expect this to do ? improve throughput of gemini is capable of doing ?
if so it should be clearly measured
I think that if it's less then 1% improvement, it's not worth our time.
what do we expect this to do ? improve throughput of gemini is capable of doing ?
if so it should be clearly measured
I think that if it's less then 1% improvement, it's not worth our time.
There is improvement up to 10% in some arias of the code, like in the generator and partitions. It optimized some parts where GC is running, putting thing on the stack instead of the heap which helped GC pauses, overall i think it will be between 2-5% improvement on the throughput, really hard to tell, until this is released.
Since it's not a big change, only slight change in release part there is no reason to not keep it, as Go advances they will add more optimizations that will be part of the PGO and thus improve it.
what do we expect this to do ? improve throughput of gemini is capable of doing ? if so it should be clearly measured I think that if it's less then 1% improvement, it's not worth our time.
There is improvement up to 10% in some arias of the code, like in the generator and partitions. It optimized some parts where GC is running, putting thing on the stack instead of the heap which helped GC pauses, overall i think it will be between 2-5% improvement on the throughput, really hard to tell, until this is released.
Since it's not a big change, only slight change in release part there is no reason to not keep it, as Go advances they will add more optimizations that will be part of the PGO and thus improve it.
your estimation is 2-5%, if you think it's important enough, test in SCT before merging this change. sync with @soyacz when we should try it.
I would recommend if it's below 1% we don't merge it
we shouldn't be doing guesswork and figuring it out after releasing, we should figure things our before merging the into the code.
what do we expect this to do ? improve throughput of gemini is capable of doing ? if so it should be clearly measured I think that if it's less then 1% improvement, it's not worth our time.
There is improvement up to 10% in some arias of the code, like in the generator and partitions. It optimized some parts where GC is running, putting thing on the stack instead of the heap which helped GC pauses, overall i think it will be between 2-5% improvement on the throughput, really hard to tell, until this is released. Since it's not a big change, only slight change in release part there is no reason to not keep it, as Go advances they will add more optimizations that will be part of the PGO and thus improve it.
your estimation is 2-5%, if you think it's important enough, test in SCT before merging this change. sync with @soyacz when we should try it.
I would recommend if it's below 1% we don't merge it
we shouldn't be doing guesswork and figuring it out after releasing, we should figure things our before merging the into the code.
Will run a simple test with it