Things to do this week instead of arguing about mixers
Eric Sandeen
sandeen at redhat.com
Tue Apr 28 13:55:29 UTC 2009
Florian Festi wrote:
> Eric Sandeen wrote:
>> - rpm? (file installation and/or db files?)
>
> The rpmdb is a bunch of db4 files. In don't think I want to interfere with
> the db4 engine...
They're often badly fragmented, too :) But whether or not that matters
much for rpm performance, I dunno.
[root at host ~]# filefrag /var/lib/rpm/Packages
/var/lib/rpm/Packages: 844 extents found, perfection would be 1 extent
[root at host ~]# ls -lh /var/lib/rpm/Packages
-rw-r--r-- 1 rpm rpm 52M Apr 27 15:40 /var/lib/rpm/Packages
[root at host ~]# du -h /var/lib/rpm/Packages
52M /var/lib/rpm/Packages
> I am not really sure whether preallocation offers that much benefit for file
> installation. May be if there is a parallel process also writing to the disk
> heavily - but such processes should probably have their own partition/disk.
> Would preallocation make sense for tar?
>
> Let's ask the other way round: What are the costs of preallocation for - say
> 2M files with 50GB overall content?
>
> Florian
I wouldn't use it for every file; it may not even make sense for this
application, I was just throwing things out :) A reasonable test might
be to do a fresh install (maybe a mock chroot install?) and go look to
see if things are fragmented afterwards, and if it's bad enough to
warrant a change. Thinking more about it, it's probably not needed,
most of these files are small and with delayed allocation, layout should
be fine.
I don't want to spur a new generation of perfect-allocation fetishists,
but in places where poor allocation may impact performance (or adversely
affect later allocations by badly fragmenting freespace) it's worth
considering the tool.
-Eric
More information about the fedora-devel-list
mailing list