User: Password:
Subscribe / Log in / New account

Fast zero page deduplication?

Fast zero page deduplication?

Posted Apr 7, 2013 2:05 UTC (Sun) by dlang (subscriber, #313)
In reply to: Fast zero page deduplication? by djm1021
Parent article: In-kernel memory compression

caches are getting larger, but last I looked, the speed ratio between the CPU and memory was continuing to grow (although not as fast as it was).

you can spend quite a few CPU cycles between the time that you tell the cpu to fetch the next chunk of ram into the cache and the time it's available. I don't know how many, and it will vary from chip to chip (with the increasing impact of ARM/MIPS and other non-X86 chips, I expect that this is getting to be a more interesting concern than it ever has been). The ability to vary the speed of the CPU without varying the speed of the memory makes this even more interesting :)

I don't know how many CPU cycles you spend in the compression. It would be interesting to find out how this compares to the 'free' cycles that are available on current CPUs when walking though memory.

I have never looked at any LZO code, so I'm not in the position to make modifications there.

> Else I think the algorithm in zram and the zcache patch recently posted by Wanpeng Li should work just fine.

I expect they work just fine as well, I'm just suggesting a possible improvement (and then explaining why I'm doing so when it sounds like I'm not getting the point across)

(Log in to post comments)

Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds