jffs zlib tuning

[NoiseEHC]

Then it is the last nail to the coffin of asm-izing zlib. As I see the 
LZO code is cache friendly so there is no way to speed it up.
I am sorry wasting you time. :)


imm wrote:
>> On 7 Jan 2008, at 19:28, Benjamin M. Schwartz wrote:
>>
>>     
>>> Nonetheless, I decided to benchmark the compressors on
>>> http://xs-dev.laptop.org/~cscott/olpc/streams/joyride/build1513/ 
>>> devel_jffs2/xo-1-olpc-stream-joyride-build-1513-20080105_1602- 
>>> devel_jffs2-tree.tar.bz2
>>>
>>> Results:
>>> Uncompressed             732M
>>> LZO -1 (fast)            345M
>>> LZO -9 (slow)            298M
>>> Gzip (defaults)          269M
>>> Bzip2 (as downloaded)    236M
>>>
>>> Note that fast vs. slow refers to compression speed.   
>>> Decompression is
>>> the same speed in either case.
>>>
>>> These tests were run on the whole tar-file.  As noted by Bernie, the
>>> differences would likely be smaller on 4KB blocks.
>>>       
>> I had some notes, which I can't now find, for a real-time sample  
>> logger we built...
>> Anyway, we did some tests compressing small blocks, and I am sure  
>> we saw much less of a difference than Benjamin saw on his big block.
>> Also, I'm pretty sure bzip2 fared very badly in our test, the small  
>> blocks didn't suit it at all (and it was slower, of course.)
>> Of course, the statistics of our sample data was probably not  
>> representative, so YMMV...
>>
>>
>>
>>     
>
> _______________________________________________
> Devel mailing list
> Devel at lists.laptop.org
> http://lists.laptop.org/listinfo/devel
>
>
>