WhiteSites Blog

SAS RAID 0 VS SCSI RAID 0 VS SATA Benchmarks

Posted on Jan 30, 2009 by Paul White

SCSI vs SAS comparison testsRecently I got into a discussion with another IT about how used servers on ebay are getting dirt cheap.  He told me that for $300 he was able to buy a Dell 2650 rack server with 4 x 146GB SCSI hard drives.  Back in its prime a server like this would have gone for over $4k.  Something else he told me was that if he just parted the thing out, he could make over $100 per SCSI drive.  If true he has found himself quite a nice little investment opportunity.  But this also brought to attention another fact.  SCSI drives are fast, and unlike CPUs and Motherboards,  they don't depreciate in the same way.  The sole fact that there are litterally millions of SCSI drives in use today, means that they will probably be around for quite some time.  Even if a SCSI drive fails, you can just swap the bad one out and slap a new one in, and then let your RAID card Rebuild it.  Its much cheaper to do this than to upgrade your hardware.

Is newer hardware really nescessary?


Considering that for $300 you can buy a rack server off ebay, that could easily run your website server, DB Server, Mail Server, ext.  Why would anyone spend thousands on new hardware.  I have talked with guys who run Hosting company's, and they said that Hard drive failures are rare.  Even said he has some drives that are still running after 10 years with no problems.  If you read up on the detiled specs. Most SCSI drives are rated for several million hours of opporation.  This means they will still be running even when you are long gone.

SCSI vs SAS comparison


The only logical reason a person would have to go with newer technology is because it has some sort of performance or reliability benefit.  This brings me to the question.  How does SCSI and SAS compare.  For this I will need to run benchmarks on two systems.

Our Systems for the benchmark


SAS System


2 x 146 GB 15K RPM SAS drives
running in RAID 0 on the built in ASUS P6T Deluxe RAID controller.
Other hardware ( core i7 2.6, 6GB RAM, Windows XP Pro SP3 )
Full tower ( 4 months old )

SCSI System


2 x 147 GB 10K RPM SCSI drives
running in RAID 0 on a Dell Perc 4/DI RAID controller with 128 MB
Other hardware ( dual Xeon 2.8, 2 GB RAM, Windows 7 Ultimate )
Dell Poweredge 1750 rack server ( over 6 years old )

Points of focus


Notice how the older system has its own dedicated RAID card, while the newer system is using the RAID controller that is built into the ASUS board.  This will help to answer some of those questions about how much performance you really give up by not getting a dedicated RAID controller.

Not a perfect comparision


First I will admit that putting a pair of 10K drives up againts a pair of 15K drives may seem like an unfair comparision.  In fact I don't expect the 10K drives to get the kind of speeds that the 15K drives do, but what is important is how close they do get.  Is it really worth it to spend the extra $$ to support SAS if SCSI is just a little slower.  We will seee.

Initial Testing Issue


It turns out that RoadKil's Disk Speed doesn't run well on Windows 7.  The program would come up but would not find any drives. HD Tack would not run on Windows 7.  Finally I found a program that would run on both systems, HD Tune.

HD Tune Benchmark for SCSI system



SCSI Hard Drive benchmark

HD Tune Benchmark for SAS system




SAS Hard Drive benchmark


Results

SCSI System:


Access Time: 7.6 ms
Max Read: 141.6 MB/s
Min Read: 54.5 MB/s
Average Read: 119.6 MB/s

SAS System:


Access Time: 5.8 ms
Max Read: 196.0 MB/s
Min Read: 117.9 MB/s
Average Read: 172.1 MB/s

Just for fun I decided to also benchmark my SATA WD Caviar black Drive ( 1 TB ).

WD Caviar Black 1TB benchmark

SATA WD Caviar Black 1TB Drive:


Access Time: 12.5 ms
Max Read: 108.6 MB/s
Min Read: 52.9 MB/s
Average Read: 85.1 MB/s

Also tried my SATA Samsung 750 GB Drive

Samsung SATA 750GB benchmark

SATA Samsung 750GB Drive:


Access Time: 14.8 ms
Max Read: 90.0 MB/s
Min Read: 40.5 MB/s
Average Read: 71.9 MB/s

Finally just because I can, I decided to benchmark my Sandisk 8GB cruzer USB drive.

Sandisk USB Cruzer 8GB

USB Sandisk Cruzer 8GB Drive:


Access Time: 0.6 ms
Max Read: 23.4 MB/s
Min Read: 22.1 MB/s
Average Read: 23.3 MB/s

Summary:


The SAS drives perform alot better than their SCSI counter parts.  Even though SAS is faster, SCSI would likely still get the job done if you just need to run a web server.  But if you need something more demanding like a DB server with very large tables, and lots of transactions, SAS might be the better way to go.

Permalink
52868 Visitors
52868 Views

Categories associated with SAS RAID 0 VS SCSI RAID 0 VS SATA Benchmarks

Discussion

Hayden James | Aug 7, 2009 4:14 PM
One thing, the SAS is 15k vs 10K RPM of the SATA. :/ We are on RAIDed SSDs
Darek | Nov 29, 2009 9:28 PM

Heya

I dont want to worry you but you benchmark show nothing. If you want to test RAID you need to use File benchmark this benchmark.

But I can be wrong :) I dunno if HD tune got updated since I used it :]

Bye gl.

matt | Feb 5, 2010 12:10 PM
Just wanted to stop and say thank you for the fantastic post....good man!
Paul | Apr 16, 2010 2:16 PM
Just to Clarify.
The SCSI was a 10K U320
The SAS was a 15K
The Goal of this post was to give people a comparison of the the different technologies.  I am sure 15K SCSI and 15K SAS would be almost neck and neck by comparison.  But for people looking to build a budget server with used parts off ebay this information helps to put things into perspective.
BD | Jul 1, 2010 12:51 AM
This is a good effort and I thank you for this comparison. but it also makes confusion. Will you do the same test with two comparable hds?
15k to 15k?
Scsi hds are a lot cheaper.Is data transfere rate on two spindel-speed identical the same ?
Thanks
Paul | Jul 1, 2010 1:17 AM
SCSI is fine, but eventually the hardware will be phased out, especially as servers make the move to SSD drives which are typically in a 2.5" form factor.  If you are a home user and just want some great linear read and write performance on the cheap then SCSI is fine.  But remember even if you get a bunch of SCSI drives and RAID 0 them, you still will be dealing with latency issues.  If you want to eliminate the Latency get SSD drives
Dan | Oct 15, 2010 2:38 PM
Thank you for posting this. I'll keep it bookmarked. I can't tell you how many times desktop people have argued about how good SATA is and how old and slow SCSI is. A SCSI raid 1 with a hot spare is set and forget. Million hours meantime between failure and great performance it's a no brainer. That doesn't mean I don't do backups, it just means I never have to use them. Once you get to about a 100 drives though reality and statistics start to catchup with you. Even when a million hours is about 80years.
Dave | Mar 16, 2011 3:17 PM
Thanks for doing this comparison. I'm currently weighing up the pros and cons of purchasing some cheap U320 drives vs continuing with SATA. I find SATA great for transferring large files but it really strugles with lots of small files. Of course, if I had the money, I'd be buying SSD :)
Paul | Mar 16, 2011 4:27 PM
Dave,
I highly recommend you give SSD drives a try.
I am running 3 x OCZ 60 GB Vertex 2 Drives, in a RAID 0 Array.  I am pulling 600-700 MB read / writes. The prices on these have come down alot.  I am using them in all my system builds now.  Even just a single drive will blow away any SCSI or SATA drive.  Even better is the built in Intel disk controller can usually do the RAID for you, so you don't need to get any dedicated RAID card.
Dave | Mar 17, 2011 2:39 AM
Paul, I'd love 3 x SSD drives but I'm going to wait until they come down in price a bit more. Although I may buy one for my OS drive soon ;)
Paul | Jul 15, 2014 8:58 PM

Larry,  Keep in mind it wasn't an apples to apples comparison.  It was more for a fun comparrision since i had the machines in my possession.  The SCSI drives had a dedicated RAID controller.  

The truth is IDE <  SATA < SCSI < SAS < SSD.  Considering today's prices, if you were going to pound a DB Server you would want SSD drives.  Most guys I know use only SSD drives,  and then only SATA for their backup drives, and SAS for LTO TAPE.  

name
Email Needed to confirm comment, but not made public.
Website
 
 
When you Post your Comment, you'll be sent a confirmation link. Once you click this link your thoughts will be made public.. Posts that are considered spam will be deleted, Please keep your thoughts and links relavent to this Article