Re: [GSoC 2013] Defragmentation for FFS

[Manuel Wiesinger <manuel%bsdstammtisch.at@localhost>]

Statusupdate:

I am still working on the analysis part, mixed with features for allocation.

What has been done:

.) fixed (almost all) problems with indirect blocks. There is still one, 
which is worth to be posted on the mailinglist, when I do not find 
anything in the archives, I post it in half an hour or so.

.) Directories can be fully iterated

.) Endianness is fully supported, I have to check all parts precisely, a 
function call might be still missing at somewhere.

.) a lot of debugging of system tools (especially dump), to get a good 
view how certain things are done in a portable and stable manner.

.) A write a precise timelog, which is in the CVS

ad code in CVS:
I got a bit lost while coding, so there is no fully stable testable 
code, to be published on the mailinglist. At the moment I discuss with 
my mentors, if would be better to commit any little success, but this 
would make the CVS, "dirty", because there are many parts, that work, 
but just proof that by a singe printf, with strange output.

Steps to improve:
.) using bread(9) instead of pread(2), it should work but I can replace 
and test that later.

.) I found another possibility to optimize the fs:

.) Dealing with short links, that store their information directly in 
the inode.

.) Test WAPBL better

I found out one more possibility to optimize. When a directory entry is 
deleted, the space of this entry (can be a file, a subdirectory or 
whatever) is just appended to the previous one. I will, when time 
allows, add a feature to optimize that list. This should make it a bit 
more efficient when reading a directory sequentially.

Manuel