zswap and zram facebook multithreaded zlib and converting to nvcomp
wanfuse123 opened this issue · 4 comments
Hi!
I have gotten my system to run lz4hc multithreaded compression and decompression on my cpu using facebook's zlib implementation (which is multithreaded) , however I am thinking this might be even better if nvcomp was used to multithread the compression and decompresssion on the video card instead.
I read some science papers that indicated it might have a 2.5 to 3x boost over performing it on the cpu. (Especially since I only have a core2due cpu (and seems I always bury 60% of my core performance performing the compression/decompression.
I have an older video card (GeForce GTX 1050 Ti) .
Is this possible to do?
Is there directions on how to get it working. I could share my zswap and zram and facebook zlib implementation compile and installation for Ubuntu 22.04 if needed (or if anyone is interested) but I have no idea how to do this with nvcomp or even if my video card supports it.
Can you advise?
Thanks in advance.
This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.
Didn't get an answer in time. No longer needed.
This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.
Sorry for missing your question. The team has been swamped with 2.4 release work over the summer.
The examples and documentation we provide might be helpful with integration into zram/zswap, although we didn't use into this use case previously.
I'm going to close the issue based on #67 (comment). Feel free to reopen if you have more questions.