Re: ot: best filesystem for small files
Phil Brutsche wrote:
> Except that the reiserfs patches wreak havoc on Ingo Molnar's raid patches
> (which are IMO needed to get any decent software raid setup working), or
> they did the last time I tried them together (during the summer).
should work ok with the "stock" raid though right? (the 0.36 not 0.90+)
from what i hear the new raid gives a lot of good stuff for raid 5, but
there isn't a lot of changes for raid 0/1 ? i have been using raid 0.36
for
a couple years now on a production server and am perfectly happy with
sticking to it on raid1 systems.
> All that is moot if you use a hardware raid controller.
maybe we will get hardware raid eventually. at the moment mylex raid
controllers
(what seems to be reccomended as the best for linux) do not like the
latest
supermicro motherboards(they are useless in them actually) which is what
i use
for servers, mylex knows about the problem and will address it in their
next
firmware update(their words), but i don't know when that will be:( (btw,
the
problem is they don't support the (B)ios (B)oot (S)pec which prevents
the system
from booting off the mylex, and the mylex clobbers the onboard adaptec
scsi preventing
it from being able to boot as well.
> IMO the journaling capabilities of reiserfs are one of the biggest reasons
> people use it :)
yeah, the way i see it the only thing journaling does is eliminate fsck
times,
which in an enviornment where we have unexpected crashes once every 2
years or so,
sitting through a 5-10minute fsck every 2 years isn't a problem :) we
have enough
battery backup to ride out even some of the worst power outages. And in
the past
5 years i've only experienced 2 outages that have lasted longer then an
hour ..
> don't be so quick to rule out 2.4, though - test11 is working *really*
> well for me in production environments (once I do stuff like "echo 0 >
> /proc/sys/net/ipv4/tcp_ecn" :)
im sure it is...for me the 2.1 series was really good, but my servers
didn't
see the light of 2.2 until 2.2.10 came out(my first 2.2 production
deployment..)
even if it works well i want to wait to let more debugging happen,
especially
since this is a SMP system and SMP is generally more fragile on
development kernels.
> ext2 will serve you fine. Just be patient if you ever need to fsck the
> filesystem.
cool, for some reason i had this thought in my head that ext2 would
crumble
doing something like this :)
>
> I have >10000 messages in ~/Maildir (qmail's maildir format is very
> similar to the mail folder format of cyrus, from what I hear) here on my
> workstation, and the only performance problem I have is the fact that it's
> on a 5400 RPM narrow Ultra-SCSI HD.
>
> If you get *fast* drives (ie 10k or 15k RPM SCSI) you shouldn't have any
> trouble.
yeah they are 10,000RPM Ultra160 ...i think 4meg of cache on each. can't
wait
to try this out, currently we use UW IMAP, and it is SOOOOOOOOOOOOOOOOO
SLOW!@#
> BTW in such situations I find it's worthwhile to have a fair amount of
> memory in the mail server - it helps performance *a lot* when you can
> cache most (or all) of the mail boxes on the mail server in RAM.
hopefully 256Meg is enough to start, can upgrade to 512 or 1GB later if
needed.
plan to have about 600MB in swap ..
thanks for the suggestions!!
nate
--
:::
ICQ: 75132336
http://www.aphroland.org/
http://www.linuxpowered.net/
aphro@aphroland.org
Reply to: