Hi Rodrigo,
Rodrigo S de Castro wrote:
> Nitin Gupta wrote:
>> I was testing WKdm de/compression algo and surprisingly it's not
>> working!! The de-compressed and original text do not match. Rodrigo's
>> patch contains exactly this algorithm implementation (did he make some
>> changes which I can't see?) and it works (though, I never myself tried
>> that patch).
>>
>
> What exactly is not working?
>
> I remember I changed some things in WKdm to make it compile. If I
> recall correctly, WKdm uses some userspace libraries not available in
> kernel space (ceiling/floor math functions?) and I implemented these
> functions somewhere in the code and made them available for WKdm to
> work. If this is the problem you are facing, I may take a look at the
> code to recall with more details.
>
Ah...I just did not look into WKdm code carefully. It is currently
hard-coded to work only when number of input words is 1024 (i.e. 4K
pages) and I was testing it for 2, 256, 512 etc. no. of words -- it's
working perfectly well for 4K input. I have not yet started taking it to
kernel space and was just testing it in user-space. There are many
optimizations I can see in current WKdm implementation and I'll first
try to get some of these optimizations working in user-space and then
adapt it to kernel space.
>> Can somebody please look into WKdm implementation
>> (http://www.cs.utexas.edu/users/oops/compressed-caching/WKdm.tgz) and
>> tell what's wrong?
>>
>> (WK4x4 seems to work okay, while LZO is pain in neck to get it into
>> kernel...so I've dropped LZO idea for now).
>>
>
> LZO works in my patch, but I cannot remember what I have done for that.
>
Yes, I saw your patch has included LZO. But, I could not figure out how
much memory footprint it requires during compression of a page. Due to
design of compressed cache I am trying to implement, it will be
difficult to use any compression algorithms as 'black box'. Since WKdm
and WK4x4 are well documented and easy to understand, I am giving much
more stress to them.
Do you have any documents describing LZO algorithm in detail -- I could
not find any of them myself ? Or, maybe your own study of LZO ? This
will be of great help :)
>> WKdm is fastest with nice compression ratios so it'll be nice to have
>> it working instead of WK4x4 (we can always include both -or more- of
>> them later but for now.....).
>>
>
> Remember that WKdm is very nice with swap cache data, but not with
> general page cache data.
>
What about WK4x4 -- does it shows good (I mean, somewhere between LZO
and WKdm) compression for page cache pages as well?
Best Regards,
Nitin
|
