ac100 team mailing list archive

[Marc Dietrich <marvin24@xxxxxx>]

On Wednesday 11 April 2012 22:43:18 Jani Monoses wrote:
> > One thing missing (not really missing, but kinda wrong) is CONFIG_ZCACHE=y
> > instead of "=m" (compressor for clean buffer cache pages). Zcache cannot
> > really be used as a module and upstream has a patch to disable module
> > support. You may also activate CONFIG_FRONTSWAP (compresses dirty buffer
> > cache pages). It works fine here, the effect isn't large though.
> 
> Ok. IIRC frontswap or zcache was causing regressions at one point last
> year right? I think that is why
> I did not look at it anymore. I may be confusing that with other
> caching support bits though.

I don't remember such problems (maybe because of other trouble), but I merged 
quite a few bug fixes over the time. Plus, I don't see any issues here, but 
well... .

Reading my previous mail again, I need to clarify that ZCACHE is the backend 
for cleancache (the one for clean buffer cache) and frontcache (for dirty buffer 
cache). So cleancache only works if zcache=y, otherwise 
/sys/kernel/debug/cleancache shows only zeros. Same holds for frontswap.

So either we try ZCACHE=y, CLEANCACHE=y, and (optionally) FRONTSWAP=y
or set all to "N" to be on the safe side.

I prefer to enable all, because we have a low mem machine were every bit is 
precious and to have a bit larger testing base until precise is released - I 
guess in April 2012 - oh no, time passes so fast.

So, I'll leave it up to you to enable the caches or not. Remember that you 
also have to specify "zcache" on the kernel cmdline to use them (which is off 
by default I think).

Marc