Re: [gentoo-server] For a 1.5TB server running Samba and NFS ONLY,what's the best kernel?

[Miguel Filipe]

My rule of thumb:
- its a production system -> use a FS that you are acostumed.. don't
try out something new
- its going to store huge/big files -> xfs
- its going to sture lots of small files (tipical unix system) -> reiserfs
- its going to be carelessly used (linux in a deskop system in a
university lab.. for public usage for instance) -> ext3

I have a ftp & samba file server (our gentoo mirror is there) in a
linux 2.6 with reiserfs, we have about 0.5TB of disk space... and a
"regular" usage, since it's a public ftp mirror...

About the filesystems,

-reiserfs is fast & rock solid for what I can say, about on one dozen
servers with diferent workloads (dns server, mail server, www server,
DB server, desktop, firewall..etc..) in fast dual cpu servers and in
slow ancient machines, and we never had problems..
reiserfs its said (and proved) to be the best when it comes to lots of
small files.
However, I do know of system crashes in the university campus proxy
server (proxying a 100mbit uplink) with a raid 5 full of scsi 15krpm
disks that were atributed to the filesystem.. (reiserfs again..) that
machine has a huge workload all year round... cpu usage is allways
above 60-70% ..(dual cpu system)

-ext3 currently, from reiserfs, xfs and ext3 is the one wich offers
higher data-protection...
since it logs metadata & data while xfs, reiserfs and jfs only log metadata..
I use it in linux installs for campus public machines, about one
hundred machines, reinstalled every semester .. all dual-boot
linux/winxp .. some of the machines are quite old and unreliable,
cases are locked but users do tend to use the reset button (no point
in thanking that off, since they would unplug the powercable)
Let me say that in average, 5 to 10% of the machines end up with the
NTFS screwed, and 1% with the linux fs screwed...

-xfs is the most advanced FS in tecnological features (beyond my level
of knowledge to prove it), its also believed to be the best when it
comes to huge file systems/huge files... basically, its the SGI
filesystem they use for SGI supercomputers.. also its the one who
scales best in large SMP and heavy load environments.

-jfs - don't know to much about it, it's porpuse is the same has xfs,
IBM filesystem for IBM big iron with Linux...

-reiser4 (not reiserfs4 .. its name is _reiser4_) its revolutionary
since it does not use logging but it offers better data-protection
through  the implementation on the FS level of transactions... a write
or succeeds.. or fails... never "in between".. but is obviously not
ready for production grade systems..
It also intruduces new ideas/uses and semantics to files... a file is
both a file _and_ directory.. If reiser4 strives in linux, it will be
groundbraking .. since its not a "standart UNIX fs"...

greetings,

On Tue, 21 Sep 2004 15:33:32 -0500, Stuart Stegall
<stuart@×××××××××××××.com> wrote:
> I have been running 2.5 then 2.6 for two years ;)  We've been using XFS as our main FS for about 18months.
>
>
>
> Christian Parpart wrote:
>
> > On Tuesday 21 September 2004 9:16 pm, Stuart Stegall wrote:
> >
> >>Anyone know if all of the x86_64 problems have been resolved? Last time I
> >>checked, it was "working" on the platform, but didn't have performance
> >>improvement over reiserfs3.2.
> >
> >
> > ext3 is known to serve well on x86_64 - although, I'd suggest, to use kernel
> > 2.6 over 2.4 whenever possible.
> > I've two 1.2TB file server racks physically left to my own server at the
> > university, both running in this configuration very well.
> >
> > Greetings,
> > Christian Parpart.
> >
>
>



-- 
Miguel Sousa Filipe