Southbridge speeds

[Cal Vanize]

I'm considering enhancing the performance of my existing system. I have
an Opteron 175 on an ASUS A8N-VM CSM with a WD Raptor 150ADFD. I want
to rebuild the system using the onboard RAID controller with three or
four identical hard drives in RAID0 (not Raptors, but fast 7200rpm
Seagates or WDs). I recognize that the access time will be adversely
impacted using four drives, but most of what I do is based on high speed
data transfer. I'm not into gaming, I just need fast I/O.

My question is, based on the nVidia 6150/430 chipset, what is the
fastest data transfer rate through the southbridge to the CPU? This
would seem the gating factor in performance. That or the capacity of
the RAID controller.

Any comments?

TIA and happy new year.

 

[kony]

I'm considering enhancing the performance of my existing system. I have
an Opteron 175 on an ASUS A8N-VM CSM with a WD Raptor 150ADFD. I want
to rebuild the system using the onboard RAID controller with three or
four identical hard drives in RAID0 (not Raptors, but fast 7200rpm
Seagates or WDs).

The most common RAID0 (not 1+0 or 0+1) arrangement is a
pair, two drives. What exactly were you wanting to do? If
you meant RAID5 instead of RAID0, it is not much of a
performance improvement, is slower than a pair of drives in
RAID0 config.

I recognize that the access time will be adversely
impacted using four drives, but most of what I do is based on high speed
data transfer. I'm not into gaming, I just need fast I/O.

My question is, based on the nVidia 6150/430 chipset, what is the
fastest data transfer rate through the southbridge to the CPU? This
would seem the gating factor in performance. That or the capacity of
the RAID controller.

Any comments?

TIA and happy new year.

The southbridge - CPU data rate is far higher than you could
ever hope to see from a RAID0 of the fastest drives on
earth. The integral RAID controller supports SATA300 so
that isn't a factor either, likely not even if the drives
were each running at SATA150 since it is a
bandwidth-per-port situation where each of the drives can't
sustain anywhere near 150MB/s alone and when raided it
doesn't matter if individual port bandwidth is lower than
total summed bandwidth.

Your primary bottleneck will be the drives themselves, and
then the other non-related aspects of processing the data
like that your Opteron is not slow by any stretch but now
seeming a bit dated in performance relative to newer
processors, and the DDR(1) 400MHz memory bus is also going
to effect performance some.

 

[Cal Vanize]

The most common RAID0 (not 1+0 or 0+1) arrangement is a
pair, two drives. What exactly were you wanting to do? If
you meant RAID5 instead of RAID0, it is not much of a
performance improvement, is slower than a pair of drives in
RAID0 config.

I'm investigating the use of three or four drives in RAID0 striping IF
(and its a *BIG* if) the additional drives provide improvement in
performance over two drives. (Will four drives in RAID0 come close to
doubling data transfer rates over two drives?)

The southbridge - CPU data rate is far higher than you could
ever hope to see from a RAID0 of the fastest drives on
earth. The integral RAID controller supports SATA300 so
that isn't a factor either, likely not even if the drives
were each running at SATA150 since it is a
bandwidth-per-port situation where each of the drives can't
sustain anywhere near 150MB/s alone and when raided it
doesn't matter if individual port bandwidth is lower than
total summed bandwidth.

Your primary bottleneck will be the drives themselves, and
then the other non-related aspects of processing the data
like that your Opteron is not slow by any stretch but now
seeming a bit dated in performance relative to newer
processors, and the DDR(1) 400MHz memory bus is also going
to effect performance some.

Agreed. But when I recently tried setting up a two-drive RAID0 array
with old 80gb drives on a cheap motherboard with an Ath 64 3700 San
Diego, I was blown away by the performance improvement. A recent
article on Tom's suggest the kind of massive improvements I'm looking
for. Four-drive arrays were pushing read data through at over 300MB/s.
Three-drive arrays read and write at over 200MB/s.

http://www.tomshardware.com/2007/07/02/raid_scaling_charts/

So 200MB/s relates to 1600mhz at serial rates. 300MB/s is 2400mhz.
Plus overhead. If the bus transfers data to/from the CPU (via the
northbridge) at those rates, will I run into a bottleneck downstream of
the hard drives? Or are these bus rates stated in mhz for parallel data
words? (Is this the correct terminology?)

 

[kony]

I'm investigating the use of three or four drives in RAID0 striping IF
(and its a *BIG* if) the additional drives provide improvement in
performance over two drives. (Will four drives in RAID0 come close to
doubling data transfer rates over two drives?)

I doubt 4 would double transfer rate, but this is only
speculation as I have not done this and don't recall seeing
benchmarks of anyone who had. I'm not even sure if that
southbridge controller can support 4 drive RAID0. Even if
it can it seems like a big risk, since you have a much
higher chance of volume loss when 4 drives might
individually fail.

Agreed. But when I recently tried setting up a two-drive RAID0 array
with old 80gb drives on a cheap motherboard with an Ath 64 3700 San
Diego, I was blown away by the performance improvement. A recent
article on Tom's suggest the kind of massive improvements I'm looking
for. Four-drive arrays were pushing read data through at over 300MB/s.
Three-drive arrays read and write at over 200MB/s.

http://www.tomshardware.com/2007/07/02/raid_scaling_charts/

That's with a hardware, 256MB cached raid controller card.
I don't know much about the types of access the test app
causes but the results are suspect because it is reporting
higher read and especially write scores than others have
with even a single drive alone (RAID0 will never double the
score of one drive in most read world uses but maybe in
certain benchmark scenarios?).
Avg single drive score 70MB/s:
http://www.techtree.com/techtree/jsp/article.jsp?print=1&article_id=82880&cat_id=632

So 200MB/s relates to 1600mhz at serial rates. 300MB/s is 2400mhz.
Plus overhead. If the bus transfers data to/from the CPU (via the
northbridge) at those rates, will I run into a bottleneck downstream of
the hard drives? Or are these bus rates stated in mhz for parallel data
words? (Is this the correct terminology?)

The bus supports several GB/s, it will not be a bottleneck.