Re: Rethinking system set-up, would like some feedback
Hello Olaf,
On 25-Feb-00, you wrote:
OM> Dear all,
OM>
OM> Now that I've got two brand new external SCSI drives to play with I
OM> am rethinking my system set-up. If you have got some advice, I'd
OM> love to hear it. Here is the scoop.
OM>
OM> The system in question is an IBM PC 300PL with an internal 8.3GB IDE
OM> drive and 2 external SCSI drives (20GB and 27GB). It also has an MO
OM> drive and CDRW connected to it.
General rules: Put the swap, s/w and data on different spindles. Put
database journals/logs on different spindles to the databases.
Consider mirroring on different spindles for s/w and 'flat' data
partitions. Splitting swap, s/w and data across different spindles
will give a performance benefit. Splitting databases and journals
across different spindles gives recovery. Mirroring across different
spindles gives resilience, but if done in s/w results in a performance
loss.
Think about what data you are dealing with and try to balance/reconcile
the above, then leave yourself enough scratch space so that you can
reorganise it all again in the light of experience.
Set up at least two installations of each version of Linux that you run
i.e. stable & unstable (three is better - total six - your system should
never be down and you can be right up to date ;-/ ;-)) - you can update
one and test it, and if there are problems, you can boot a
back-up/fall-back system.
If you have frequently used CDs, put them on HD, in 'scratch' space.
Depending on your data, and MO capacity, archive to these devices. The
RW capability of CDRW can be a 'double edged blade' - I wouldn't want
to RW it too many times, but it's ok for getting it wrong a few times.
Consider putting the SCSI HDs on separate controllers - performance+,
Resilience+- not h/w mirroring though :-(.
If I were running journalised databases, I would put duplicate Linux &
s/w partitions on each of the SCSI HDs with corresponding swap
partitions on the 'other/opposite' HD. I would put db journals/logs/db
backups on the internal HD. I wouldn't mirror the system, s/w & db
partitions but would update the second db from the journal/log from the
first (dirty fingers bit). If I were setting up a workstation I would
put the swap and backups on the internal HD and mirror everything else
- it would be worth getting a second controller - still not h/w
mirroring though :-(.
And then, in the light of experience, I'd probably 'revise' it ;-)
Summary: multi-spindle disk systems can give benefits in:
1. Performance.
2. Recovery (in the event of h/w, s/w & user errors)
3. Resilience (in the event of h/w failure)
Unless you have 50GB+ of data, you should be able to archive at least
two out of three of these.
Bye,
LeeE
--
http://www.spatial.freeserve.co.uk
...or something.
Reply to: