Hi All,
I have an A8n SLI Delixe motherboard. I have 3 250GB drives.
Currently the drives are partitioned and my apps are on one partition.
I want to increase how fast it takes to load the apps as I need to use
them. Someone suggested using RAID 0.
I see that the controllers are built into the motherboard. I have a
few questions regarding this since I am new to this arena:
1) Should I continue to use IDE drives, or should I go with SATA?
2) Is RAID 0 going to give me the increased performance, or some other
RAID type?
3) How difficult is this to set up?
4) I use Western Digital drives. If one goes bad, do I basically lose
the entire array?
5) Can I put 4 drives in the system, and mirror the RAID 0 array to
have a backup at the same time?
Thanks for any helpful input you may have.........
Arthur
1) I was going to point you to the database at storagereview.com ,
but they've only got SATA on there. The PATA were removed.
The media rates for disks are still in the ~70MB/sec range
at the beginning of the disk. It doesn't matter whether you use
100MB/sec PATA, 133MB/sec PATA, or 150MB/sec native SATA, the
sustained rate is still head limited. But the newer drives will
be packaged in SATA format, and perhaps at some point, new
drives won't be coming out in PATA format.
Faster interfaces do allow bursts of data to be held in
cache, until the head can catch up. But, something to be
aware of, is some caches are "multi-segmented", and not
necessarily the entire cache can be used for one locality
of access. I have never seen details for SATA/PATA cache
policy, so I don't know how caches are organized on those
disks. With SCSI disks, apparently it is possible to set up
the cache, to do a better job in either a desktop config or
in a server config.
I/O rate is another parameter. The seek time helps determine
the performance for random access (many small files). The
higher RPM drives (10K SATA Raptor, 10K SCSI, 15K SCSI) have
reduced seek time, which gives improved performance compared
to a 7200 RPM drive.
There are some potential improvements in disks. Like the
ability for the disk to reorder seeks (tagged operation). If
two disks have the same seek time, but one can reorder which
read or write operation to service first, the performance of
the drive can be improved by 10%. That improvement would be
more visible in server environments, as in a single threaded
situation, the application would still be waiting for the
reordered operations to come back, and there would be a
less pronounced improvement.
SCSI has had tagged operation for many years, and SATA now
has the option to do that as well.
2) RAID 0 is excellent for improving STR. With add-in controller
cards, you can put eight drives in RAID 0 and get 400MB/sec.
But then you need system busses on the motherboard that can
handle that data rate. An example, is to stuff a PCI Express x8
controller card, into one of the video card slots on an SLI
motherboard (like products from Areca). Since PCI Express
is so new, there aren't too many (cheap) choices for controller
cards like that, so this is a more expensive option.
3) Any RAID is easy to set up. But, what people don't bother to
experiment with, and learn about, is how to do maintenance when
the array reports certain errors. It is important to learn how
to repair RAID 1, RAID0+1, or RAID 5 arrays, when the RAID BIOS
complains about the array. Simulating failures and repairing them
will build your confidence, when facing real trouble. RAID
manuals are generally woefully lacking in describing what to do.
Waiting for the RAID to break, and having no backups, is the
wrong time to be "experimenting" with the array.
4) RAID 0 is dead, when one drive fails. RAID0+1 or RAID10 are
options with more redundancy.
5) Yes, you can do RAID0+1, and use one two disk RAID array to
mirror a second two disk RAID array. For the cheap controllers
built into motherboards, this is the highest performance option.
RAID5 has pathetic write performance, and needs an expensive
controller card with cache memory and XOR engine, to improve on
that.
You still have to do backups! Say you install four disks in a high
performance redundant configuration. You have a smug look of
satisfaction on your face. Then, the +12V rail on the PSU rises to
+15V and burns the controller boards on all four disks drives. All
your data is instantly inaccessable! There have also been cases,
where a controller failure corrupts information on all disks, and
that will trash the data as well. There was even one report from
a poster, who discovered his RAID 1 mirrored array (on a SIL3112)
was desynchronized - the two disks had different information on
them, and yet he claims to have never had a failure report from
the software. A redundant RAID array is not a replacement for
backups - the main benefit is the ability to defer maintenance
for a few hours, such as in a business environment, where the IT
staff can fix a server array, after the employees go home.
Redundancy there allows a server to continue to operate, when
one disk fails.
Paul
