[GSoC] HAMMER2 block compression feature

[Daniel Flores]

Hello.

My name is Daniel Flores and I’m a student from Spain. I study in Barcelona
School of Informatics, which is a faculty of Polytechnic University of
Catalonia, and it’s my last year before I graduate.

I sent a mail with this proposal to Matthew 4 days ago, but I didn't get a
reply yet, so I decided to send this to the mailing list, as it is
suggested on Dragonfly's GSoC page. Before submitting any official
application, I would like to ask for your opinion and criticism regarding
my proposal and whether I have the required skill and adequate planning to
successfully finish this project.

I’m interested in implementing the block compression feature in HAMMER2,
which is described on Dragonfly’s GSoC page [1]. The feature consists in
that any written block should be compressed, if possible, to 50% of its
size or less (in powers of 2 – 32 KB, 16 KB and so on, the initial size of
block being 64 KB). It should be possible to set that compression mode on
any directory, so that new contents written in it and its subdirectories
would be compressed.

I believe that I can succeed because I have programming experience in
C/C++, since we use those languages for most practices in our university. I
also completed courses centered on algorithms and their implementation in C
or C++. I also have an interest in operating systems, their concepts and
their development. I completed a course on Operating Systems development
where I and a partner implemented a simple linux-based kernel called “ZeOS”
(using C and, at times, x86 assembler). It’s a very simple kernel that has
a task scheduler, system calls, exceptions, interruptions, memory
management and a file system. We were able to successfully run it on bochs
emulator and it passed all tests needed to successfully finish the course.

If you wish, you may take a look at it [2].

Sadly, it’s documented in Spanish, but I just want to provide it as an
example of code that I can produce.

In spite of it being a kernel implementation, I think that my skills and
knowledge aren’t yet the best suitable for kernel development projects, so
instead I prefer to do something moderately difficult, yet related to
operating systems, like HAMMER2 block compression feature.

My initial outline for project planning is the following:

1. (Starting May 27 or before – June 17) During community bonding period I
would get myself familiarized with the community, Dragonfly BSD system/code
and “hammer2” branch code. I would also refresh my knowledge on compression
algorithms and would complete a design for the future implementation, like
choosing the appropriate algorithm and having defined everything needed to
be done. I have read Matthew's design document for HAMMER2 [3] and, as I
understand, the algorithm #2 is not yet determined, so I believe this
should be decided before the actual coding period will start. Also, more
detailed planning for coding period would be done, with clear milestones
and deliverables.

The deliverables for this phase would be in the form of documents that
would describe the design, the chosen algorithm(s) and the calendar with
milestones.

2. (June 17 – July 29) I would start by coding a prototype application that
would implement the defined design, consisting of algorithms #0, #1 and #2.
I would need it to insure that the algorithm is implemented correctly and
to do all necessary debugging and refining before integrating it in
HAMMER2. I would create a series of tests to prove that the algorithm works
well. If time allows, I would also try to study its behavior statistically
and to improve it, if possible.

The deliverables would be code of the prototype as well as any supporting
applications (to generate tests, for example) and tests themselves.

3. (July 29 – September 16) After confirming that the prototype works well,
I would start actually integrating it into HAMMER2 filesystem. At this
point, the task would be gradually adding prototype’s code without any
conceptual modification into HAMMER2 and testing if it works correctly in
real-life. A series of testing applications with appropriate system calls
would be created and the results checked for correctness. Also, the
appropriate hammer2 utility command to set the compression feature to
directories would be created and tested.

The deliverables would be the code of implemented feature and application
tests for it.

4. (September 16 – End of GSoC 2013) Hopefully, all known bugs would be
corrected at this point, so I would take the time to do final refinements
to documentation, to write a detailed description of what was implemented
for future developers and to finish everything needed to start submitting
my code to Google.

Generally, my philosophy is to start “as soon as possible”, so if some
phase is done before the deadline, I would immediately go to the next
phase. If project goes smoothly and there is some time left, I would be
happy to do small refinements or bugs fixes on general system or HAMMER2.

During the coding period (June 17 – September 16) I don’t have any
commitments (like courses or trips) at all, so I should be able to easily
dedicate 40 hours per week or more to this project. Prior to June 17 I have
my studies and some exams, but generally it should not prevent me from
getting familiarized with code and getting the design/algorithm decisions
done, because my workload is not very big.

I don’t have any past experience with Open Source or generally big
projects, but I believe that I can accomplish this task since it is
generally related to C programming and algorithms, and I believe that I
have sufficient knowledge of those fields from my university courses.

Currently I am getting myself familiarized with Dragonfly 3.2 (for i386)
running it on Oracle VM Virtual Box 4.2.12, but I’m able to install it on
physical machine, if necessary. I’m also trying to learn as much as
possible about BSD in general and particularly Dragonfly.

If you think that my set of skills is more appropriate to some other
project related to HAMMER2 or Dragonfly, I would be grateful if you pointed
at them. I also would consider working on HAMMER2 block encryption feature,
since it is similar in requirements to my current proposal, if there is
already a strong candidate to implement the block compression feature.

Thank you.

[1] http://www.dragonflybsd.org/docs/developer/gsocprojectspage/
[2] https://dl.dropboxusercontent.com/u/56646280/zeos.zip
[3] http://leaf.dragonflybsd.org/mailarchive/users/2012-02/msg00020.html


Daniel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dragonflybsd.org/pipermail/kernel/attachments/20130423/81bec0d5/attachment.htm>