[GSOC] HAMMER2 compression feature week12 report

[Daniel Flores]

Hello everyone,
here is my report for week 12.

During this week I was mostly bug-hunting with the help from my mentor. The
most important result of that is the fact that now we pass, in the most
cases, the blogbech test. There are still some problems in certain
circumstances, but they are rare and we'll continue bug-hunting during the
last week of GSOC.

Regarding the blogbench results, sadly, there is not much to say about
them, because the score it gives doesn't make much sense in our case and it
varies without any observable correlation to the compression mode used (and
even if there's no compression). This, probably, happens because blogbench
seems to write only zeros instead of some actual data... But it served us
really well as a general stress-test when it comes to file creation,
concurrent reading/writing, file deletion, etc.

Another test I performed this week was the test of zero-checking
performance. This test was suggested earlier by Samuel and I just used the
same command that he suggested: “dd if=/dev/zero bs=64k count=5000”. This
command was executed 10 times for each mode.
You can see the results in this table [1].

Generally, as you can see, zero-checking greatly improves the performance
and the results are the same for zero-checking and all compression modes,
as it was expected, because the compression modes execute the same actions
as the basic zero-checking mode.

In previous report I also promised to try to perform some tests on files
that can't fit either in the category of fully compressible or fully
incompressible files. Sadly, I don't have much to present here either,
because most of files I tried, such as some .odt and .pdf's were mostly
almost completely incompressible for both compression modes that we have
(LZ4 and ZLIB). So ZLIB doesn't really change much. There was one .pdf file
where LZ4 managed to compress 2 blocks out of 21 and ZLIB managed to
compress 7 of them, so in some cases there is a difference, but probably
there aren't many files like this. I'll continue looking into this matter
and I'll try more different files.

The next week is the last week of coding. What I'll be doing during it is
more bug-hunting and more refinement of the code, so, I hope, it will be
completely clean by the end of the week.

I'll appreciate your comments, suggestions and criticism. You can check out
the code in my repository, branch “hammer2_compression” [2].


Daniel

[1]
http://leaf.dragonflybsd.org/~iostream/hammer2_zero_checking_performance.html
[2] git://leaf.dragonflybsd.org/~iostream/dragonfly.git
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dragonflybsd.org/pipermail/kernel/attachments/20130909/41970565/attachment.html>