Archive Home arrow Reviews: arrow Storage arrow SandForce SF1200 RAID-0 SSD Performance
SandForce SF1200 RAID-0 SSD Performance E-mail
Reviews - Featured Reviews: Storage
Written by Olin Coles   
Monday, 03 May 2010
Table of Contents: Page Index
SandForce SF1200 RAID-0 SSD Performance
Features and Specifications
SandForce SF-1200 SSD Controller
SSD Testing Methodology
AS-SSD Benchmark
ATTO Disk Benchmark
CrystalDiskMark 3.0 Tests
Iometer IOPS Performance
EVEREST Disk Benchmark
SSD vs Hard Disk Drive
SandForce RAID-0 Conclusion

Iometer IOPS Performance

Iometer is an I/O subsystem measurement and characterization tool for single and clustered systems. Iometer does for a computer's I/O subsystem what a dynamometer does for an engine: it measures performance under a controlled load. Iometer was originally developed by the Intel Corporation and formerly known as "Galileo". Intel has discontinued work on Iometer, and has gifted it to the Open Source Development Lab (OSDL).

Iometer is both a workload generator (that is, it performs I/O operations in order to stress the system) and a measurement tool (that is, it examines and records the performance of its I/O operations and their impact on the system). It can be configured to emulate the disk or network I/O load of any program or benchmark, or can be used to generate entirely synthetic I/O loads. It can generate and measure loads on single or multiple (networked) systems.

To measure random I/O response time as well as total I/O's per second, Iometer is set to use 4KB file size chunks over a 100% random sequential distribution. The tests are given a 50% read and 50% write distribution. While this pattern may not match traditional 'server' or 'workstation' profiles, it illustrates a single point of reference relative to our product field.

Our charts show the Read and Write IOPS performance as well as I/O response time (measured in ms). Iometer was configured to test for 120 seconds, and after five tests the average is displayed in our benchmark results. The first tests included random read and write IOPS performance, where a higher I/O is preferred.

Iometer_Random_4K-IOPS_Results.jpg

In the Random IOPS performance tests the single layer cell (SLC) OCZ Agility EX (3982/3988), Intel X25-E Extreme (3543/3548), and OCZ Vertex EX (3106/3091) outperformed all other products by a wide margin.

Combined into a RAID-0 array using 4KB stripes, the pair of SandForce SF-1200 SSDs only led their individual counterparts by a small margin, and 128KB stripes were virtually the same as a single SSD. The RAID-0 set produced 2205/2204 IOPS, while the individual ADATA S599 SSD produced 2087/2081 I/O's, Corsair F100 delivered 2077/2076, and the RunCore Pro-V recorded 2021/2015.

Indilinx-based MLC SSDs start with the Corsair Nova delivering 1982/1995, then the OCZ Vertex Turbo SSD rendered 1774/1770, while the Corsair recorded 1769/1773; both of which are just slightly ahead of a single Vertex SSD that produced 1702 for read and write IOPS. Finishing out the second-tier IOPS performance is the WD SiliconEdge-Blue with 1625/1632 and OCZ Agility SSD with 1625/1618 IOPS.

The SATA 6Gb/s Crucial RealSSD-C300 produced 1070 read-IOPS with 1069 write, and creates a third-level tier of operational performance suitable for personal computer systems. Kingston's SSDNow V+ SNVP325 offered 826/829 IOPS. Every other product thereafter responded with low IOPS operational performance, and are not suggested for high input/output applications or performance-orientated computer systems.

Iometer_Average_Response_Time_Results.jpg

The Iometer random IOPS average response time test results were nearly an inverse order of the IOPS performance results. It's no surprise that SLC drives perform I/O processes far better than their MLC versions, but that gap is slowly closing as controller technology improves the differences and enhances cache buffer space. The Read/Write IOPS performance for the 64GB OCZ Agility-EX SLC SSD was 0.19/0.06 ms, with the Intel X25-E Extreme SSD measuring 0.22/0.06 ms, while the OCZ Vertex-EX achieved 0.26/0.06 ms.

Stripe size didn't impact response time. Our pair of RAID-0 SandForce SSDs responded to IO commands in 0.23/0.23, which isn't much quicker than a single SF-1200 SSD. The ADATA S599, Corsair F100, RunCore Pro-V (all SandForce SF-1200 SSDs), each produced 0.25/0.25 ms.

The Corsair X256 and OCZ Vertex Turbo SSD both scored 0.50/0.06ms, while the Vertex SSD offered 0.52/0.06ms. The mainstream Agility SSD produced 0.55/0.06ms while Western Digital's SiliconEdge-Blue offered 0.51/0.11ms, the Corsair Nova delivered 0.57/0.09ms, Crucial's 256GB RealSSD C300 responded in 0.87/0.06ms, and the second-generation Kingston SSDNow V+ SNVP325 responded to read requests in 0.27ms while write requests were a bit slower at 0.93ms. The Western Digital VelociRaptor did very well compared against SSD products, producing 6.59/0.82ms. These times were collectively the best available, as each product measured hereafter performed much slower.

Drive Hardware

In our next section, we test linear read and write bandwidth performance and compare its speed against several other top storage products using EVEREST Disk Benchmark. Benchmark Reviews feels that linear tests are excellent for rating SSDs, however HDDs are put at a disadvantage with these tests whenever capacity is high.



 

Comments 

 
# Performance is there but not the priceK Gregory 2010-05-04 15:21
I'm not a SSD owner, and I had written them off as way over-hyped and exponentially over priced. I looked toward the 'PCI-e SSD' solution as the one to wait for,watch and mature. Of course the astronomical prices are completely unjustified as well.

However the SanForce controller is not over-hyped. The products are still exponentially over priced but I must now watch SSD's and view them as a product to consider when the prices reach below SAS at near offered capacities.

Great review again!
Report Comment
 
 
# User experience sells the productOlin Coles 2010-05-04 15:26
Like NVIDIA's GeForce 3D-Vision technology, all you see is another tech gadget... until you try it for yourself. That's what I thought about SSDs three years ago, but when I was loaned one for testing and loaded Windows, I was immediately sold. Had I known how many SSDs I would eventually test (now nearing 50), I wouldn?t have paid money to be an early adopter.
Report Comment
 
 
# Systems Development ConsultantPaul A. Mitchell 2010-05-05 05:31
SATA/3G is NOT the current standard.

Please also test with 2 x Crucial RealSSD C300 in RAID 0,
even if it's necessary to use the software RAID in Windows
+ an inexpensive 6G controller like the ASUS PCIE GEN2 SATA6G
controller.

And, comparing other host controllers would also be a good test
e.g. Highpoint 640, Intel's RS2BL040, and the integrated
6G ports on AMD's latest 890FX chipsets on motherboards by
ASUS, MSI and Gigabyte.

Until more SSDS conform to the current standard,
we will just have to wait and see what happens
to market prices for 3G and 6G products
when both are widely available.

That's why I'm waiting (for now).


MRFS
Report Comment
 
 
# RE: Systems Development ConsultantOlin Coles 2010-05-05 07:22
Paul: SATA-3GB/s may not be the most current standard, but that's what the SandForce controllers are designed for, and that's what the majority of systems can support. I try to reach the largest enthusiast audience possible with my articles.

I only have one Crucial C300 SSD, and there aren't any plans to purchase a second unit. I do have a RAID-capable SATA 6Gb/s controller, if you or anyone else would like to loan one out. Comparing host controllers would be interesting, but since it would be specific to brands and drivers I'm not confident of the articles' shelf life.

In reality, a single SSD of any denomination will deliver virtually the same end-user experience as a RAID-0 array.

Also- I used Intel Rapid Storage Technology Driver 9.6.0.1014 for this article.
Report Comment
 
 
# software RAID?Paul A. Mitchell 2010-05-05 09:37
FYI: I've already written to Olin privately,
so the following comments are for the benefit of
other readers:

We recently added 2 cheap SATA/6G ports with this ASUS
PCIE GEN2 SATA6G controller -- only $20 at Newegg here:

##newegg.com/Product/Product.aspx?Item=N82E16813131610&Tpk=N82E16813131610

Yes, that card does NOT support RAID.

So, we had no trouble enabling XP's software RAID 0
with 2 x new WD 1TB SATA/6G HDDs:

##newegg.com/Product/Product.aspx?Item=N82E16822136533&Tpk=N82E16822136533

Just convert partitions to "dynamic disks" and you're almost there.

Can someone loan Olin one more Crucial C300 SSD, maybe Crucial,
so he can compare software RAID 0 with 2 x SATA/6G SSDs?

It might even work with Intel's latest Rapid Storage Technology
set to JBOD on both devices.

Until then ...
... keep up the good work, Olin!


MRFS
Report Comment
 
 
# Suggesting software Raid 0?K Gregory 2010-05-05 15:11
Software Raid is not the reason for most SSD purchases.(only guessing from all the forums,reviews and the like) I believe most SSD's are purchased to be the boot drive. Whether its a single SSD or a Raid0 SSD boot drive(OS/System), perhaps the standard testing/reviewing should continue to reflect hardware configurations only for SSD articles.

OS Software Raid solutions can alter the testing results when compared to Hardware Raid. This is due to caching, CPU utilization(compute) , increased latency(I/O), and even performance increases in Software configured Raid benchmarks that would not appear in Hardware configured Raid benchmarks(such as queuing/scheduling and the like).

-just some thoughts, but I'm sure Benchmark Reviews will keep having thorough reviews.
Report Comment
 
 
# SSDs and RAIDOlin Coles 2010-05-05 15:16
In my twelves years as an IT professional, I have never used a software-based RAID array. RAID-1/5/10 arrays have always been built and managed through a HBA controller, which removes overhead from the rest of the system. As far as SSDs go, it would be safe to say that at leat 95% of them are used in single-drive configuration... usually as the boot drive.

As for using the the PCIe-based SATA 6Gb/s add-in cards for a RAID array, or even a compatible SAS controller, the PCI-Express bus is still limited to 5GB/s bandwidth.
Report Comment
 
 
# I should've guessed...K Gregory 2010-05-06 17:56
Most SSD's were only single drive configurations. Perhaps as prices decrease there will be more of a shift towards 2 drives in RAID-0.

As I posted earlier, now that controllers like this SandForce are out I am going to focus on SSD's considerably more. I'll scrutinize the prices on SSD's considerably more as well!
Report Comment
 
 
# HBA controller removes overheadPaul A. Mitchell 2010-05-05 16:47
> HBA controller ... removes overhead from the rest of the system

I agree that this has been a major marketing tool
of HBA vendors, but a lot of those vendors began
their HBA marketing when CPUs had only one core.

Now, with dual and quad-core CPUs quite common,
the idle core(s) can do a lot of the same computation
that was previously done by dedicated IOPs.

Multi-cores, in turn, can make software RAID
a very cost-effective proposition, e.g.
for the 8 GB database that we manage and
update cheaply with XCOPY.


MRFS
Report Comment
 
 
# Software 'storage' RAID agreementK Gregory 2010-05-06 17:51
There is full agreement here that software RAID can be both cost effective and offer fantastic performance with modern 4-Core and up CPU's. Software RAID(configured correctly)however is only risk free when its 'storage RAID' and not 'Boot RAID' -at least with Hard Drives.

But back to Olin Coles testing with HBA/Raid Controllers...

Way to go! That's the industry wide accepted standard for most RAID. Software RAID would really change the benchmarks, and probably greatly when considering SSD's. But that would be a great article right there. Since so much technology has changed, what is the performance and benching differences between Software and Hardware RAID today on SSD's and HDD's? Perhaps coming soon to Benchmark reviews?
Report Comment
 
 
# PCI-Express bus is still limited to 5GB/s?Paul A. Mitchell 2010-05-05 16:49
> the PCI-Express bus is still limited to 5GB/s bandwidth

#en.wikipedia.org/wiki/PCI_Express#PCI_Expr ess_2.0

"... a 32-lane PCI connector (x32) can support throughput up to 16 GB/s aggregate."

Thus, PCI-E 2.0 x16 lanes support 8 GB/second bandwidth.


Did you mean 5 Gigabits per second, instead of 5 Gigabytes per second?


I think you must be referring to a single PCI-E 2.0 x1 lane,
in only one direction.

PCI-E 1.0 x1 lane has a bandwidth of 2.5 Gb/sec in each direction;
PCI-E 2.0 x1 lane has a bandwidth of 5.0 Gb/sec in each direction.

Quadruple each for x4 lanes, and double that again for x8 lanes e.g.
Highpoint RocketRAID 2720 or Intel RS2BL080.

The edge connector on the add-in card is a quick indicator
of a RAID card's rated bandwidth.


MRFS
Report Comment
 
 
# RE: PCI-Express bus is still limited to 5GB/s?Andre 2010-08-30 12:53
Actually, the edge connector only determines the maximum possible number of lanes, not the actual used. You could stick a x4 card into a x16 slot, and only get 4 lanes used; or you could stick a x16-sized card (with 4 actual lanes) into a x16 slot and still only use 4 lanes.
Report Comment
 
 
# SSD userAdam Smith 2010-07-11 05:54
Recently bought Ocz Vertex 2 50 GB
Tech at OCZ explained overprovisioning as a MAJOR FACTOR in the usable lifespan of an MLC ssd.

I am using it as my major drive because it is bootable too.

When installing these you must install windows while the bios is set for ACHI for the best performance.

Also turn off indexing and all the "performance" enhancing options in windows.

Windows is SO MUCH FASTER. The internet on the slowest DSL is so much improved.

I have asus ma4a89gtd pro/usb3 (stock athlon II 635 4 core 2.9GHZ processor + windows 7 pro) and the drive set up to usb 6gbs - but I don't think usb3 makes any speed difference.
Report Comment
 
 
# Trim not possible with Raidstw500 2011-02-14 01:58
The article states that Trim is possible...but thats not true when using Raid. Until now there is NO Controller which enables trim support using Raid. So the article lacks of testing the long behaviour of a ssd-raid without trim support...perhaps you should have read some anandtech articles to that theme.
Report Comment
 
 
# RE: Trim not possible with RaidOlin Coles 2011-02-14 07:33
Yes, this article states that TRIM support is possible on SandForce-driven SSDs. But it also states "Most RAID controllers lack pass-through TRIM support" in the Cons section of the conclusion... perhaps you should have read some of this article instead of trying to point out your favorite website.
Report Comment
 
 
# RE: RE: Trim not possible with Raidstw500 2011-02-14 07:53
well, on last page "SandForce RAID-0 Conclusion" there is written, "...Intel's ICH-10 SATA controller allows pass-through TRIM functionality, but only with their latest Intel Rapid Storage Technology (RST) software...". That means for readers, that this controller is able to pass the trim command, but isnt true. The note in the cons, "Most RAID controllers lack pass-through TRIM support" does not balance the pretence the ich10R is capable of trim in raid-mode. Anandtech is not my favourite website, i mentioned it just for an example. When an article about raid-0 with ssds is written, then one of the most important fact is, that trim is not possible. And that has to be mentioned not just as the last note. By the way, it would have been great to test the long behaviour. Some people state that Sandforce dont need trim to stay at full performance after the first slow down which happens all the time.
Report Comment
 
 
# RE: RE: RE: Trim not possible with RaidOlin Coles 2011-02-14 07:57
Please also be aware that Intel originally told consumers that TRIM was supported in RAID arrays with ICH10. See here: intel.com/support/chipsets/imsm/sb/CS-031491.htm
Report Comment
 

Comments have been disabled by the administrator.

Search Benchmark Reviews
QNAP Network Storage Servers

Follow Benchmark Reviews on FacebookReceive Tweets from Benchmark Reviews on Twitter