1 SCSI drive outperforms EIDE RAID 0

Raystonn

Distinguished
Apr 12, 2001
2,273
0
19,780
From <A HREF="http://www.storagereview.com" target="_new">http://www.storagereview.com</A>
<pre>Seagate Cheetah 18XL (18.4 GB Ultra160/m SCSI):
</pre><p><pre>All Loads
File Server Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 106.54 149.18 149.37 214.33 268.70
Total MBs per second 1.16 1.63 1.60 2.31 2.88
Avg I/O Response Time (ms) 9.38 26.81 107.11 298.54 952.04
CPU Utilization (%) 0.39 0.54 0.49 0.78 1.10
I/Os per % CPU Utilization 273.18 276.26 304.84 274.78 244.27
</pre><p><pre>Workstation Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 127.60 161.85 161.85 221.46 271.55
Total MBs per second 1.00 1.26 1.26 1.73 2.12
Avg I/O Response Time (ms) 7.84 24.71 98.86 288.95 942.17
CPU Utilization (%) 0.47 0.59 0.65 0.85 1.23
I/Os per % CPU Utilization 271.49 274.32 249.00 260.54 220.77
</pre><p><pre>Database Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 108.48 155.46 155.57 215.96 273.09
Total MBs per second 0.85 1.21 1.22 1.69 2.13
Avg I/O Response Time (ms) 9.22 25.73 102.84 296.30 936.83
CPU Utilization (%) 0.35 0.58 0.56 0.80 1.24
I/Os per % CPU Utilization 309.94 268.03 277.80 269.95 220.23
</pre><p><pre>----------------------------------------------------------------------------
</pre><p><pre>AMI MegaRAID IDE100, RAID 0, 2 Drives:
</pre><p><pre>All Loads
File Server Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 71.52 73.06 84.00 96.24 109.76
Total MBs per second 0.77 0.78 0.92 1.04 1.19
Avg I/O Response Time (ms) 13.98 54.77 190.43 664.66 2328.24
CPU Utilization (%) 1.40 1.47 1.74 1.96 2.68
I/Os per % CPU Utilization 51.09 49.70 48.28 49.10 40.96
</pre><p><pre>Workstation Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 84.13 85.35 96.75 110.39 125.68
Total MBs per second 0.66 0.67 0.76 0.86 0.98
Avg I/O Response Time (ms) 11.88 46.86 165.35 579.55 2034.28
CPU Utilization (%) 1.58 1.56 1.87 2.31 3.03
I/Os per % CPU Utilization 53.25 54.71 51.74 47.79 41.48
</pre><p><pre>Database Access Pattern Linear Very Light Light Moderate Heavy
Total I/Os per second 81.59 82.96 93.29 106.25 120.71
Total MBs per second 0.64 0.65 0.73 0.83 0.94
Avg I/O Response Time (ms) 12.25 48.21 171.47 602.02 2117.99
CPU Utilization (%) 1.64 1.63 1.74 2.16 2.97
I/Os per % CPU Utilization 49.75 50.90 53.61 49.19 40.64
</pre><p><pre>-----------------------------------------------------------------------------
</pre><p>It would seem that a single SCSI drive still outperforms 2 EIDE drives in a RAID 0 configuration. I would like to see motherboard manufacturers place hardware SCSI Ultra160 RAID 5 support on their boards.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
 

Kelledin

Distinguished
Mar 1, 2001
2,183
0
19,780
I'd sure like to see it on server boards, at least...I've seen half-length SCSI RAID adapters, so it should be possible.

FWIW, AMI makes this one crazy two-piece motherboard setup for their 14u rackmount--it's basically a quad Xeon northside part plugging into a backplane with a proprietary connection. The northside part takes up perhaps 11 of those 14u...wonder if it has onboard SCSI RAID? I'll have to check...

Kelledin

"/join #hackerz. See the Web. DoS interesting people."
 

rcf84

Splendid
Dec 31, 2007
3,694
0
22,780
SCSI is alway faster the Standard ATA Raid. Remember the Controller does the work not the CPU "ATA is all CPU work". Also got to remember scsi are smart controllers thx to controllers like Intel i960.

Nice Nvidia and ATi users get a Cookie.... :smile: Yummy :smile:
 
G

Guest

Guest
Do you happen to know which 2 IDE drives were being used in that test? That could make a fairly significant difference.
 

FatBurger

Illustrious
Agreed, it could be two pieces of crap, although they didn't even use the fastest SCSI drive (15,000 RPM), unless the 18XL is the model number, but I thought it was X15.

Also, Promise has a controller for IDE that supports RAID 5. It's the Supertrack 66/Supertrack 100. Just thought somebody might be interested.

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>
 
G

Guest

Guest
SCSI and ATA are two different beasts, and I've never thought it was a fair comparison between the two. They each have their strengths and weaknesses, but comparing the two is like comparing apples to oranges.
SCSI is still okay for servers and the like, but ATA/100 7200rpm drives are more than fast enough for pretty much everything else. 15000rpm is overkill for almost anything, including servers.
Also keep in mind that ATA/133 drives will be shipping towards the middle of October, which will further narrow the gap between SCSI and ATA. Serial ATA is in the works, which is said to be the equivalent of ATA/600, but it won't even begin production until late 2002. Barring any improvements in SCSI's transfer rates, ATA is only 30Mb/s slower than SCSI, with the exception of FibreChannel which is way too expensive even for many businesses.
 

FatBurger

Illustrious
15000rpm is overkill for almost anything, including servers.
Also keep in mind that ATA/133 drives will be shipping towards the middle of October, which will further narrow the gap between SCSI and ATA.
First of all, 15,000 RPM can hardly keep up with many high-end web or program servers. Hard drives are about the biggest bottlenecks in a system.

How can you say that ATA133 will close the gap? We still only have 7200 RPM drives, which will give no more than 35MB/s sustained transfer rate.

ATA is only 30Mb/s slower than SCSI
Did you actually look at the review above? (I'm assuming you mean Bytes, not bits). SCSI drives are 2/3x faster than IDE. And that's for general use. They can be slower or even faster than that in some instances.



EDIT:
I checked that site for info on the 18XL, and it's only a 7200 RPM drive!!! That's really amazing!

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green><P ID="edit"><FONT SIZE=-1><EM>Edited by FatBurger on 10/01/01 05:01 PM.</EM></FONT></P>