Hi!

I just got a Dell PERC4 (rebranded LSI 320-1) RAID adapter, and I'm 
somewhat surprised of the performance of a simple RAID 0 with 3x73gb 
15k.3 seagate drives.

Under complete RAID 0, I can get a Winbench 99 transfer rate of about 
70-80MB/sec which is pretty low, since with only 1 drive I can get 
70MB/sec on the outer tracks.

I then looked up on the internet for some benchmarks and I found a site 
(albeit not in english) http://www.tweakers.net/benchdb/testcombo/90 
were one can see that 4 disks in RAID 0 give a throughput of 77MB/sec 
all over the disk.

I'm really amazed that performances are so low!

Is this really what I should be getting?

How can I discover if its my PCI-X bus that is limiting my transfer rate 
(I'm running on a DFI 855GME motherboard with only 1 card in the PCI-X 
slot)? Is it possible that the motherboard chipset is limiting the 
transfer rate?

If its the SCSI card that is the problem, what type of SCSI card should 
I get to have the maximum throughput on a RAID 0 setup?

Thanks in advance!
Daniel

lc  <lachinois@gmail.com> wrote:
>I just got a Dell PERC4 (rebranded LSI 320-1) RAID adapter, and I'm 
>somewhat surprised of the performance of a simple RAID 0 with 3x73gb 
>15k.3 seagate drives.
>
>Under complete RAID 0, I can get a Winbench 99 transfer rate of about 
>70-80MB/sec which is pretty low, since with only 1 drive I can get 
>70MB/sec on the outer tracks.

Note that with most loads the transfer rate isn't that important,
access time is usually far more important (multiple spindles help a
lot here under load). The major exception is certain types of video
editing :-)


>I then looked up on the internet for some benchmarks and I found a site 
>(albeit not in english) http://www.tweakers.net/benchdb/testcombo/90 
>were one can see that 4 disks in RAID 0 give a throughput of 77MB/sec 
>all over the disk.

Well, look at the other statistics on the same page, ATTO for example
shows a "maximum read transfer rate" of ~150MB/s, while WinBench99
gives 77MB/s.

It would be interresting to know exactly what they claim to be
measuring, just "transfer rate" doesn't really say much.

IMNSHO WinBench 99 isn't a particularly good benchmark anyway, I don't
think anyone been using that seriously for a long time (it was pretty
heavily criticized when it came, and 6 years is a long time for
computers).


>How can I discover if its my PCI-X bus that is limiting my transfer rate 
>(I'm running on a DFI 855GME motherboard with only 1 card in the PCI-X 
>slot)? Is it possible that the motherboard chipset is limiting the 
>transfer rate?

The DFI 855GME motherboard appears to have a Intel 855GME chipset,
with the "normal" IO controller replaced with a 6300ESB.

From the documentation on Intels page it appears that the 6300ESB is
connected via a 266MB/s (theoretical) link to the rest of the system
(855GME), this bandwidth will then be shared by ALL I/O in the system
(network, USB, IDE and PCI), presumably excluding AGP.

The 6300ESB can "support" a large number of high-speed PCI/PCI-X
connections, but lack the bandwidth to actually do much with them,
unless you can avoid going via CPU or memory! It support 64-bit/66MHz
PCI-X, but the bandwith from the I/O hub to the rest of the system is
only 50% of that, and that bandwidth is also SHARED with several other
I/O devices!

Also, the base chipset for this is a "Mobile" chipset, which makes me
wonder a bit how good performance one can actually expect, though the
6300ESB isn't an mobile part which should help a bit.

But still, from the tech specs it looks like OUGHT to be able to push
a bit more than 80 MB/s over the PCI-X slot, at least if you don't
push much other data over the same "hub interface" (like the network
interface), perhaps up to 200-230MB/s.

But it's such a uncommon construction that nothing can replace
benchmarking, and I don't think I've ever seen someone benchmark the
6300ESB, much less this specific combination.


>If its the SCSI card that is the problem, what type of SCSI card should 
>I get to have the maximum throughput on a RAID 0 setup?

Maximum throughput, no holds barred?
In Linux I would have said any simple SCSI card plus software RAID,
fast and fragile :-) Not sure how fast/good the software RAID in
Windows is.

Seriously, I would expect higher figures than this from even the older
PERC4's, even thought serial STR isn't usually where RAID card receive
most of their tweaking (it's on the list of important things, but I
expect it to just make it to the top 5).

Torbjorn Lindgren wrote:
> lc  <lachinois@gmail.com> wrote:
> 
>>I just got a Dell PERC4 (rebranded LSI 320-1) RAID adapter, and I'm 
>>somewhat surprised of the performance of a simple RAID 0 with 3x73gb 
>>15k.3 seagate drives.
>>
>>Under complete RAID 0, I can get a Winbench 99 transfer rate of about 
>>70-80MB/sec which is pretty low, since with only 1 drive I can get 
>>70MB/sec on the outer tracks.
> 
> 
> Note that with most loads the transfer rate isn't that important,
> access time is usually far more important (multiple spindles help a
> lot here under load). The major exception is certain types of video
> editing :-)
> 

Ah yes, well access time is great really, I have no problems with that! Few.

> 
> 
>>I then looked up on the internet for some benchmarks and I found a site 
>>(albeit not in english) http://www.tweakers.net/benchdb/testcombo/90 
>>were one can see that 4 disks in RAID 0 give a throughput of 77MB/sec 
>>all over the disk.
> 
> 
> Well, look at the other statistics on the same page, ATTO for example
> shows a "maximum read transfer rate" of ~150MB/s, while WinBench99
> gives 77MB/s.
> 
> It would be interresting to know exactly what they claim to be
> measuring, just "transfer rate" doesn't really say much.
> 
> IMNSHO WinBench 99 isn't a particularly good benchmark anyway, I don't
> think anyone been using that seriously for a long time (it was pretty
> heavily criticized when it came, and 6 years is a long time for
> computers).
> 

Indeed! I ran the ATTO bench, and I can get 145-150MB/sec with only 3 
disks. Thats 2/3 of the expect maximum throughput, so I think thats 
pretty good.

Apparently, the 29320-R or 39320-R from adaptec can do even better with 
HostRAID but I'm sceptical... I would believe the PERC4 or LSI 320-1 
should do better?

Do you know if the original LSI firmware for the PERC4 is any better 
than the dell firmware? I have seen a few people attempt this with 
success (if only for the support which is much easier and more 
accessible on the LSI site than trying to find things on 
support.dell.com). I called up LSI and they told me it would most 
problably kill my card if I flashed it to the LSI firmware.

> 
>>How can I discover if its my PCI-X bus that is limiting my transfer rate 
>>(I'm running on a DFI 855GME motherboard with only 1 card in the PCI-X 
>>slot)? Is it possible that the motherboard chipset is limiting the 
>>transfer rate?
> 
> 
> The DFI 855GME motherboard appears to have a Intel 855GME chipset,
> with the "normal" IO controller replaced with a 6300ESB.
> 
> From the documentation on Intels page it appears that the 6300ESB is
> connected via a 266MB/s (theoretical) link to the rest of the system
> (855GME), this bandwidth will then be shared by ALL I/O in the system
> (network, USB, IDE and PCI), presumably excluding AGP.
> 
> The 6300ESB can "support" a large number of high-speed PCI/PCI-X
> connections, but lack the bandwidth to actually do much with them,
> unless you can avoid going via CPU or memory! It support 64-bit/66MHz
> PCI-X, but the bandwith from the I/O hub to the rest of the system is
> only 50% of that, and that bandwidth is also SHARED with several other
> I/O devices!
> 
> Also, the base chipset for this is a "Mobile" chipset, which makes me
> wonder a bit how good performance one can actually expect, though the
> 6300ESB isn't an mobile part which should help a bit.
> 
> But still, from the tech specs it looks like OUGHT to be able to push
> a bit more than 80 MB/s over the PCI-X slot, at least if you don't
> push much other data over the same "hub interface" (like the network
> interface), perhaps up to 200-230MB/s.
> 
> But it's such a uncommon construction that nothing can replace
> benchmarking, and I don't think I've ever seen someone benchmark the
> 6300ESB, much less this specific combination.
> 

Well, it can go higher than 130MB/sec, now thats for sure. I wonder if 
it would go higher than 200MB/sec? I'll know when I get more disks.

> 
> 
>>If its the SCSI card that is the problem, what type of SCSI card should 
>>I get to have the maximum throughput on a RAID 0 setup?
> 
> 
> Maximum throughput, no holds barred?
> In Linux I would have said any simple SCSI card plus software RAID,
> fast and fragile :-) Not sure how fast/good the software RAID in
> Windows is.

Do you know if the adapte 39320 can do RAID 0 spanning channels? That is 
2 drives on each channel, all in the same RAID 0?

> 
> Seriously, I would expect higher figures than this from even the older
> PERC4's, even thought serial STR isn't usually where RAID card receive
> most of their tweaking (it's on the list of important things, but I
> expect it to just make it to the top 5).

Well, its not bad actually, but it seems like the LSI 320-2X is really 
really better at RAID'ing.

Daniel

Torbjorn Lindgren wrote:
 > lc  <lachinois@gmail.com> wrote:
 >
 >>I just got a Dell PERC4 (rebranded LSI 320-1) RAID adapter, and I'm
 >>somewhat surprised of the performance of a simple RAID 0 with 3x73gb
 >>15k.3 seagate drives.
 >>
 >>Under complete RAID 0, I can get a Winbench 99 transfer rate of about
 >>70-80MB/sec which is pretty low, since with only 1 drive I can get
 >>70MB/sec on the outer tracks.
 >
 >
 > Note that with most loads the transfer rate isn't that important,
 > access time is usually far more important (multiple spindles help a
 > lot here under load). The major exception is certain types of video
 > editing :-)
 >

Ah yes, well access time is great really, I have no problems with that! Few.

 >
 >
 >>I then looked up on the internet for some benchmarks and I found a site
 >>(albeit not in english) http://www.tweakers.net/benchdb/testcombo/90
 >>were one can see that 4 disks in RAID 0 give a throughput of 77MB/sec
 >>all over the disk.
 >
 >
 > Well, look at the other statistics on the same page, ATTO for example
 > shows a "maximum read transfer rate" of ~150MB/s, while WinBench99
 > gives 77MB/s.
 >
 > It would be interresting to know exactly what they claim to be
 > measuring, just "transfer rate" doesn't really say much.
 >
 > IMNSHO WinBench 99 isn't a particularly good benchmark anyway, I don't
 > think anyone been using that seriously for a long time (it was pretty
 > heavily criticized when it came, and 6 years is a long time for
 > computers).
 >

Indeed! I ran the ATTO bench, and I can get 145-150MB/sec with only 3 
disks. Thats 2/3 of the expect maximum throughput, so I think thats 
pretty good.

Apparently, the 29320-R or 39320-R from adaptec can do even better with 
HostRAID but I'm sceptical... I would believe the PERC4 or LSI 320-1 
should do better?

Do you know if the original LSI firmware for the PERC4 is any better 
than the dell firmware? I have seen a few people attempt this with 
success (if only for the support which is much easier and more 
accessible on the LSI site than trying to find things on 
support.dell.com). I called up LSI and they told me it would most 
problably kill my card if I flashed it to the LSI firmware.

 >
 >>How can I discover if its my PCI-X bus that is limiting my transfer rate
 >>(I'm running on a DFI 855GME motherboard with only 1 card in the PCI-X
 >>slot)? Is it possible that the motherboard chipset is limiting the
 >>transfer rate?
 >
 >
 > The DFI 855GME motherboard appears to have a Intel 855GME chipset,
 > with the "normal" IO controller replaced with a 6300ESB.
 >
 > From the documentation on Intels page it appears that the 6300ESB is
 > connected via a 266MB/s (theoretical) link to the rest of the system
 > (855GME), this bandwidth will then be shared by ALL I/O in the system
 > (network, USB, IDE and PCI), presumably excluding AGP.
 >
 > The 6300ESB can "support" a large number of high-speed PCI/PCI-X
 > connections, but lack the bandwidth to actually do much with them,
 > unless you can avoid going via CPU or memory! It support 64-bit/66MHz
 > PCI-X, but the bandwith from the I/O hub to the rest of the system is
 > only 50% of that, and that bandwidth is also SHARED with several other
 > I/O devices!
 >
 > Also, the base chipset for this is a "Mobile" chipset, which makes me
 > wonder a bit how good performance one can actually expect, though the
 > 6300ESB isn't an mobile part which should help a bit.
 >
 > But still, from the tech specs it looks like OUGHT to be able to push
 > a bit more than 80 MB/s over the PCI-X slot, at least if you don't
 > push much other data over the same "hub interface" (like the network
 > interface), perhaps up to 200-230MB/s.
 >
 > But it's such a uncommon construction that nothing can replace
 > benchmarking, and I don't think I've ever seen someone benchmark the
 > 6300ESB, much less this specific combination.
 >

Well, it can go higher than 130MB/sec, now thats for sure. I wonder if 
it would go higher than 200MB/sec? I'll know when I get more disks.

 >
 >
 >>If its the SCSI card that is the problem, what type of SCSI card should
 >>I get to have the maximum throughput on a RAID 0 setup?
 >
 >
 > Maximum throughput, no holds barred?
 > In Linux I would have said any simple SCSI card plus software RAID,
 > fast and fragile :-) Not sure how fast/good the software RAID in
 > Windows is.

Do you know if the adapte 39320 can do RAID 0 spanning channels? That is 
2 drives on each channel, all in the same RAID 0?

 >
 > Seriously, I would expect higher figures than this from even the older
 > PERC4's, even thought serial STR isn't usually where RAID card receive
 > most of their tweaking (it's on the list of important things, but I
 > expect it to just make it to the top 5).

Well, its not bad actually, but it seems like the LSI 320-2X is really 
really better at RAID'ing.

Daniel