[NMLUG] RAID5 - ATA vs SATA

[niosop at gmail.com (Matthew Bowie)]

>
>  With a 4 disk raid5 array (safety matters ya know).  Will I notice a huge
> difference in read/write speed from what I have now or just nominal?
>


You probably won't see too much difference, although that's pretty dependant
on the underlying RAID implementation.  You planning on using hardware or
software RAID?

>  Does SATA vs ATA make a big difference?
>

Depends on where the bottleneck is.  I *think* that SATA controllers usually
have higher throughput than ATA controllers, but this is rarely a concern as
the speed of the drives is usually much lower than the controller is capable
of handling.


> I'm pretty sure the bottleneck at this point is the sluggish disk access
> but how can I verify that it's the disks and not a saturated network
> connection?  I've been using a utility called bwm-ng to monitor bandwidth
> and have not seen it come close to the 100mb/s that my NIC is rated.
>

It could be a network issue, a protocol issue, or any of a dozen other
things.  I'd recommend making a file in RAM, and sharing that over the
network (as large a file as possible), then see what kind of transfer rates
you get.  That way you remove the drives from the equation and can narrow it
down to network or protocol issues.  More RAM could also help when you're
doing frequent accesses as Linux buffers pretty aggressively.

Also, what database are you using?  PostgreSQL by default forces a write to
disk after every transaction, which slows down performance, but improves
data integrety.  MySQL just passess it off to the OS, which is faster, but
the results usually aren't written to disk until the OS feels like it which
could be an issue.

First I guess you should see what kind of read performance you're getting
from your RAID 1 array using:  hdparm -tT /dev/md0 (replace md0 w/ whatever
the RAID device is seen as).

Then try a semi-large file copy (>50Meg) via SAMBA and time it.  Then copy
the same file to RAM and do the copy from there (although just doing two
copies might do the same due to Linux's buffering, but you don't have
control of that).  If you want to create a file for testing then:  dd
if=/dev/urandom of=testfile bs=1024 count=50000 will create one for you.
You may have a RAM drive already set up in /dev/shm, the mount command
should let you know.

After doing a SAMBA test, try the same file via FTP or SCP.  If the transfer
times are about the same using different protocols, then it's not a protocol
issue.

Also, a 100mbit connection will usually give you a maximum of around
12MB/sec transfer rate.  If you're getting around 1.2MB/sec, then there may
be a 10 base switch/hub somewhere in there, or there's a problem w/ speed
auto-negotiation and you should look at your network topology and figure
out what's going on.  A hub in your path could seriously degrade performance
if there's a lot of other traffic on your system.

Let me know if you have any questions,
-- 
Matthew Bowie
Programmer/IT Consultant
niosop at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.nmlug.org/pipermail/nmlug/attachments/20061221/2f85ac15/attachment.htm