[bisq-network/bisq] Optimize DaoState snapshot behaviour (#5609)
notifications at github.com
Fri Jul 9 18:09:22 CEST 2021
1.7 OOME on Linux
- I have had a couple of OOM errors with my live Bisq instance since upgrading to tag/v1.7.0.
- I have not had any OOM errors while testing this PR on mainnet, but my testing (mainnet) wallet is always empty, and there is no TX or trade history.
Linux OS Level Profiling (Very Simple)
I have been doing some linux sys-level profiling to see diffs in total physical RAM usage (native + heap). Native memory use is not shown by java heap profilers. The Linux OS memory profiler I'm using (`htop`) shows total physical mem usage (`RES`). It includes both on-heap and off-heap (native) memory usage in 1 value.
Addition of periodic `System.gc()` did not reduce overall RAM usage when linux's jvm `MAXRam=4g`. (Maybe because this 'low' MAXRam value is forcing more frequent GC, making the coded GC redundant and ignored.)
RES RAM Mb / Tag v1.7 pre PR / MAXRam=4g / no GC opts / 3 trials
RES RAM Mb / v1.7 PR 5609 commit 8859503 / MAXRam=4g / no GC opts / 3 trials
Note: no effect from added System.gc()
If you want to merge, don't wait on me. I think Windows is the biggest worry(?), and I can't help there.
But I will still do more testing and let you know what I find.
Today: I am going to see how it is for MAXRam=4g today, with chimp's jvm GC options.
And run tests using jvm opt MAXRam=8g, with and without chimp's jvm GC options.
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the bisq-github