up vote 7 down vote favorite
2
share [g+] share [fb]

Setup
1.5TB x 8 on a SANS Digital enclosure configured into two sets of 4 drives and connected to a Windows Server 2003 computer over eSATA. They are all identical Seagate Barracuda ST31500341AS drives.

Issue
 I assumed the guy before me set the two sets of 4 drives up as RAID 0 or 5, but he really just set them up as two concatenated volumes. I laughed at the misuse of a perfectly good RAID controller and then proceeded to turn one of them into a RAID 0 array (I do not care if my backups are lost to corruption for one day).

When this was done, I was excited to benchmark my RAID 0 volume vs. the concatenated volume and show it off to my boss, but was disappointed to find out that they are the exact same speed.

So then I switched to a simple JBOD setup and benchmarked an individual disk and got about 70% of the performance and haven't changed anything since then. The r/w speed increases at a diminishing rate when I benchmark multiple drives simultaneously. Does anyone here have any experience solving a problem like this and have any suggestions?

Here's a typical benchmark result for 4 drives in a RAID 0 or concatenated set:

Test File Size: 500 MB
Testing New File Write Speed....
Data Transfer: 30.75 MB/s, CPU Load: 1.0%
Testing Write Speed....
Data Transfer: 73.73 MB/s, CPU Load: 1.4%
Testing Read Speed....
Data Transfer: 75.29 MB/s, CPU Load: 1.4%

Here's a typical single drive benchmark result:

Test File Size: 500 MB
Testing New File Write Speed....
Data Transfer: 23.98 MB/s, CPU Load: 1.6%
Testing Write Speed....
Data Transfer: 54.16 MB/s, CPU Load: 3.3%
Testing Read Speed....
Data Transfer: 50.09 MB/s, CPU Load: 1.4%

Here's the limit I seem to reach as I benchmark more and more individual drives at the same time (I got up to 4 drives at once):

Test File Size: 500 MB
Testing New File Write Speed....
Data Transfer: 73 MB/s, CPU Load: 1.6%
Testing Write Speed....
Data Transfer: 131 MB/s, CPU Load: 3.3%
Testing Read Speed....
Data Transfer: 104 MB/s, CPU Load: 1.4%

Update:
I realized the enclosure brand is SANS Digital, if that matters. The software and internal hardware are Silicon Image. Due to time constraints, I'm leaning towards a 6 disk JBOD setup with a single RAID 1 for truly important data. BackupExec handles multiple backup locations quite well anyway. The tests so far with 3 separate disks are encouraging--over the network, the backup speeds total about 40 MB/s.

Final Update:
I decided that RAID-10 wasn't worth my trouble and followed the plan from the last update. The total speed doesn't seem to be much faster, if at all. So looks like that's the maximum.

link|improve this question
75% accept rate
What are the concurrency limits on the controller itself (IO operations per second, MB/second)? It's possible you're hitting those limits. – voretaq7♦ Aug 29 '11 at 16:06
I believe it has a Silicon Image PCI Express X1 card that maxes throughput at 2.5Gbps, or ~300MB/s. It also has dual eSATA ports. If that were the bottleneck, wouldn't the slower "new file" writes catch up with the regular writes, since the drives would be working in parallel? – erictheavg Aug 29 '11 at 16:41
7  
"Silicon Image" <-- Found the problem. Their products are cheap because they put the absolutely minimum resources in to making it "work". – Chris S♦ Aug 29 '11 at 17:24
Are you using the controller to setup the RAID, or are you setting it up in Windows? If you are using the controller, maybe try setting up a software RAID. Anyway, I think you need to lower your expectations. – Zoredache Aug 29 '11 at 18:03
I have benchmarked it in hardware and software RAID, with similar results. I was just hoping I'd get closer to what some of the comments on Newegg got. What started this whole thing was that the spanned disk was becoming unresponsive about every week, causing me to have to restart the device. I assume that if I don't use the controller it won't fail, so there's that benefit as well. – erictheavg Aug 29 '11 at 18:19
feedback

1 Answer

up vote 1 down vote accepted

If your enclosure is attached to your 2003 server by one single eSATA connection then that's your issue: the maximum throughput of eSATA is the same as a single disk.

If this is the case unfortunately the only real benefit you'll see will probably be seek time improvements.

In order to get a decent speed on the array you'll need an internal controller connected via pci/pcie and either get internal SATA/SAS drives to attach to it or, if you can find one, use a RAID controller that can attach to each disk externally via eSATA.

link|improve this answer
It's connected by dual eSATA, but thanks for the info. That would mean that there's no point in using RAID-0 on more than two volumes, which is helpful. Having dealt with an external USB drive yesterday, I have been reminded what a slow connection really is! – erictheavg Sep 13 '11 at 14:36
1  
Actually no, there is ton of sense in having more volumes if you do a ton of random IO - head movements KILL throughput. I run a 8 x 10k disc raid and sometimes i only get around 20 megabytes / secoind thanks to excessive head movement. – TomTom Sep 13 '11 at 15:15
@TomTom Good point. I should have said "for my purposes," which is large file backups. At any rate, I'm substituting the concurrency of RAID-0 with running a concurrent backup on each drive at night, so it should be a wash. – erictheavg Sep 15 '11 at 13:28
feedback

Your Answer

 
or
required, but never shown
discard

Not the answer you're looking for? Browse other questions tagged or ask your own question.