On Sun, Oct 01, 2000 at 01:48:53PM +0200, Massimo Dal Zotto wrote:
> >
> > What you have here is a case of usage v speed. Your files are broken up
> > by the file system in to blocks and scattered over the disk surface.
> >
> > To store a file which contains just a single character will take up one block.
> > Not very efficient in terms of disk usage.
> >
> > However a large file which requires many blocks also requires the disk heads
> > to seek many time toaccess them. If the block size is small then more head
> > seeking is required and performance goes down. (There's a lot more to block
> > positiong than this.)
> >
> > As a sysadmin it is your job to configure the partion in the best way for
> > the job it has to handle. If the disk is storing emails or news then file
> > size is, on the whole, small and performance not such a big issue. If the
> > file system is holding a database then performance (and a bigger block size)
> > is better.
> >
> > Don't worry about the wasted 10%. The missing space is taken up with the
> > end of files that don't fit into a block. File systems often fragment
> > the little bits of the files left over and place them into one block. The
> > problem with these fragments is that to get them requires more CPU then
> > full blocks.
> >
> > At the end of the day its the old choice of cost against performance. Your
> > system; you decide.
> >
> > Steve
> > --
> > Steve Dobson steve.dobson@krasnegar.demon.co.uk
> >
> > If bankers can count, how come they have eight windows and only four tellers?
>
> Yes, I understant this. The question is if the default debian installation
> should use the mke2fs default (4096) or force a 1024 block size.
A good question
>
> In my opinion optimizing for space would be a better choice for the root
> partition because it contains a lot of very small files that waste a lot
> of disk space and speed optimization is not very important in this case.
But what if / is the only partition? I had this for a time when I was running
my laptop as duel boot.
>
> As a system administrator I usually use different blocksizes based on the
> expected usage of the disk, but debian doesn't give me an option to do
> it in the default installation, unless I open a shell and do the job
> myself, but unfortunately I discovered the problem only after the
> installation was finished.
>
> I suggest therefore that we use the 1K option and maybe prompt the user
> for a better choice if he wants to change the default. Other options which
> could also be changed for non-root partitions are the inode ratio and the
> reserved percentage.
The choice of which block size to use, as you have said, is dependent upon a
number of factors. Most people installing are not as capable as you. I
personally say that installation should be kept as simple as possible. Lets
not confuse the inexperienced installer. How many questions should be asked?
What about those who want to set the inode size, or the number of inodes
in the system. This too have an effect on the performance of the filesystem.
I first installed Debian Slink about 18-months ago. Lots and lots of questions
on packages that were being installed. Potato is much, much better.
The fact that you can just ctrl-alt-F? and create a shell to do any special
set up I believe is a better one.
>
> BTW, my computations of %wasted was wrong because it was relative to total
> disk space instead of used disk space. On my partition the correct number
> is about 11.8% of the used space. That's a lot of space in my opinion.
>
> Regarding the use of fragments, mke2fs doesn't supports them:
>
> # mke2fs -b 4096 -f 1024 /dev/sdc2
> mke2fs 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09
> Warning: fragments not supported. Ignoring -f option
My mistake. I did a quick scan of the code and saw some fragmentation definition
is one of the ext2 headers. Glad you could correct me. Should have checked the
mke2fs manual - it state that fragments aren't supported yet.
Steve
--
Steve Dobson steve.dobson@krasnegar.demon.co.uk
<Knghtbrd> Europe Passes Pro-spam Law
<Knghtbrd> I though only Americans were that fucking stupid =>
<Espy> apparently americans are quite naive :)
Attachment:
pgpiEcwehWwFs.pgp
Description: PGP signature