> I would like some input on the following hardware choices:
>
> 3 Adaptec 2940UW SCSI-III controllers.
I've been using a 2940UW and an AIC7880 (same as the 2940UW, but built into
the motherboard) controller without problems, but I've heard plenty of
stories from people who have not been so lucky.
> 3 Adaptec 2940UW SCSI-III controllers.
> - One for root (/) which will include all logging
> - One for mounting just alt.binaries.*
> - One for mounting everything else that goes to the spool.
>
> Single CPU Pentium II 400 Mhz, with 512MB of memory.
>
> The two controllers for /var/spool/news/articles/ will each have about 22
> Gig of drive space, for a total of ~44 Gig. There will be 2 different
> raid groups, something like:
>
> /var/spool/news/articles/
> /var/spool/news/articles/alt/binaries/
>
.....
> We are considering 5 x 4.5 GIG SCSI-III disks for each raid mountpoint.
I'm just wondering, if you want to maximise performance, whether you want to
separate the controllers' functions in this way. I'm assuming you'd put a
roughly equal number of disks on each controller. Is root going to be RAIDed
too? Or have you just got one disk on the first controller? If it's the
latter, wouldn't you be better off reorganizing, so that you put a couple of
the 4.5G disks on the first controller? i.e. have sda-c on controller 1,
sdd-g on cont. 2, and sdh-k on cont. 3. Then make /var/spool/news/articles/
a RAID of sdb, sdd, sde, sdh and sdi, and
/var/spool/news/articles/alt/binaries/ a RAID of sdc, sdf, sdg, sdj and sdk.
That way, however demand for the various newsgroups varies, you spread the
load more evenly across the controllers, and hopefully avoid flooding any
one channel upon heavy demand. The only advantage that I can see of doing it
your way is if the controller for alt.binaries goes down, you don't lose the
rest of the groups (assuming the kernel doesn't go down with it).
Incidentally, have you read the recent posts (see "Kernel/raidtools version
recommendations?" thread) from other users who are choosing RAID-5 rather
than RAID-0 for their newsfeeds, because the increased risk of failure which
RAID-0 entails meant that they could end up rebuilding their spool twice a
year?
Cheers,
Bruno Prior [EMAIL PROTECTED]
