SSD RAID 0

SoftRAID (not Apple) is optimized for SSDs, and from what I have read, smaller is better, 16k is what I would try using - so just restore from your backup clone.

You probably don't need 300GB for system, most people keep data folders on mechanical drives. Keep it to just OS and applications.

If you use CS4/5 use one for scratch.

It should though be fast no matter which. Not sure what your MBook uses but Apple's tend to be low on the performance tree.

I think the 2 SSD in RAID 0 on a Mac Pro is wasted, as you are limited by the SATA II interface in the Mac Pro. The throughput on both drives is more than the SATA II interface can handle, if you want to get the true potential of the drives you will need a descent dedicated RAID card.

You are better off using the 2 SSD’s for different things like 1 for OS and programs another for say frequently used data that your programs needs.

People talk about different block sizes but I don’t know enough to comment on what is best.

3 SSDs will but all the SATA ports using the 4 drive bays share a single controller for all the bandwidth, and today's SSD maxes out around 250MB/sec or about what SATA II on a dedicated single 3Gb channel would deliver (such as Port Multiplier).

As for needing a dedicated RAID controller, hardware RAID is not, but 6G card might be useful I guess.

http://www.tomshardware.com/reviews/ssd-raid-performance,2798-10.html

http://www.xlr8yourmac.com/IDE/SSDvs_VelociRaptor_vs_Raptor/SSD_vs_VelociRaptorRaptor.html

About block size and other features: (I did mention SoftRAID 4.0, hopefully people would bother to check out the features and why I mention it, no?):

*Support for SSDs and Disks with 4 KB Sectors*
There is now specific support for SSDs (solid state disks) and disks with 4 KB sectors. SoftRAID recognizes SSDs and changes the disk failure prediction algorithm for them. It also allows you to use a 16 KB stripe unit size when creating a stripe with two or more SSDs. This has been shown by Lloyd Chambers to produce better performance than larger stripe unit sizes

http://www.softraid.com/features.html

http://blogs.msdn.com/b/e7/archive/2009/05/05/support-and-q-a-for-solid-state-dr ives-and.aspx

Engineering SSD, why writes are so complex and hard to understand
http://www.diskeeper.com/blog/post/2010/12/31/Inside-SSDs.aspx

RocketRAID 640 didn't even come out smelling like a rose:
http://www.bit-tech.net/hardware/2011/01/04/high-point-rocketraid-640-review/1

a solid state drive (SSD) is an outstanding place to locate a central database or mail server, etc, because an SSD has no seek time or latency, and can handle huge numbers of small I/O requests very efficiently.
http://macperformanceguide.com/index.html

SSDs do great where I/O is mostly doing reads, not writes.

Corair: Restoring SSD performance and imaging Disk
http://forum.corsair.com/v3/showthread.php?t=85344

Write amplification (WA) is a phenomenon associated with Flash memory and solid-state drives (SSDs).
http://en.wikipedia.org/wiki/Write_amplification

Congrats on entering the world of SSD!!! You should keep in mind that not all SSD are created equally. Intel SSD are usually a little slower than some other brands but have a longer lifespan. SSD technology suffers from a limited amount of writes to each cell before the cell fails. If your MacBook SSD is SandForce based you will likely see better performance than the two Intel units, even in a stripe.

When purchasing an SSD do the research on who makes the controller, this is the most important aspect of current SSD based drives. The list of will include Intel, Toshiba (think MBA, Kingston ssdNow series), Indilinx (OCZ Agility), SandForce (OCZ Agility 2, OCZ Vertex 2, OCZ RevoDrive, Corsair F Series, OWC Mercury series), Marvell (Micron/Crucial C300)

As for the block sizes, I would pick a smaller size than 128k for SSD as they really shine in small random access workloads.

I boot my MacPro (1,1 , 2.66ghz) from a Kingston 64 gig SSD. This model is a Toshiba based unit. It performs nicely as a boot device.

I have 2 x 1 TB Western Digital Caviar Black drives in a stripe to hold video and "slower need" data.

I have 2 x 240 GB OCZ RevoDrive (first model, not the updated X2 series) to hold Aperture Libraries. These drives have an onboard stripe which I disabled on a PC during post. Each RevoDrive exposes 2 x 120 GB drives to OS X. I take the "4" drives and stripe them in disk utility using 32K Raid Block Size.

I have a 2 TB Western Digital Caviar Green for Time Machine


I use diglloyd DiskTester for benchmarking. Please note that if you get this tool DO NOT use his Recondition SSD feature as it is only useful on early SSD technology. It slowed my Crucial C300 to the point where I had to re-flash the firmware (which does a low level format)

Results are from Sequential Test command in DiskTester

*Kingston -- (Avg / Fastest / Slowest) -- MB / Sec*
Write - 158 / 176 / 149
Read - 238 / 242 / 232

*WD Caviar Black Stripe*
Write - 186 / 191 / 175
Read - 191 / 202 / 166

*OCZ RevoDrive (Stripe of 4 SandForce based "drives", 2 from each RevoDrive, 32K Block)*
Write - 704 / 717 / 699
Read - 936 / 958 / 865

*OCZ RevoDrive (Stripe of 4 SandForce based "drives", 2 from each RevoDrive, 16K Block)*
Write - 689 / 701 / 655
Read - 943 / 950 / 920

*OCZ RevoDrive (Stripe of 4 SandForce based "drives", 2 from each RevoDrive, 256K Block)*
Write - 696 / 702 / 685
Read - 904 / 916 / 882


*OCZ RevoDrive Single Drive* +(I copied my data off the "Stripe" and tested a single drive, this should be on par with what you can expect from a single SandForce based unit such as a Cosair F series, OCZ Vertex 2, etc)+
Write - 221 / 223 / 217
Read - 266 / 268 / 260

For Comparison I will also post results from my MacBookPro (6,1 , i7 2.66)
*This drive is a Crucial C300*, 256 GB, boot. This drive has been installed for about 5 or 6 months.
Write - 213 / 223 / 209
Read - 280 / 281 / 278

I hope this gives you a better idea of various SSD technologies. As you can see there a great variation in speed when you compare the Kingston (toshiba based controller) to the Crucial (Marvell based controller). The C300 is on par with the Single SandForce Based Unit.

I will add that I have 2 more SSD in non Mac systems. A Corsair F60 and F120. Both are in Dell Laptops and they perform on par with the single SandForce results from above. Also note that the C300 is a Sata3 (6 Gbit) drive that is running on Sata2 (3Gbit) connection. Most online reviews claim that the drive performs better on the higher speed bus, but I cannot confirm.

from all I've read 16 or 32 anything higher is wasted for most users.

Macprorocks:

Great post, I'm pleased to see you did some testing.

I was waffling between 32K and 64K, and your post was very useful in helping me decide.

I'm using a pair of Intel X25 G2 80GB in a MacBook Pro, and they worked well with 32KB blocks, but I too noticed frequent freezes, so I'm hoping moving to 64KB blocks will resolve this Issue.

Also, I would recommend performing a secure erase on the SSD's every 4-6 months, depending on usage. Snow Leopard lacks TRIM support (although, some recent 2011 builds of SL appear to add this support), and SL starts to get a little sluggish, even with RAID 0.

Secure Erase really improves performance on an SSD that's experienced a high rate of change. My SSD RAID set pwrformance improved by around 30% after the secure erase, so Apple deperately needs to add TRIM support to Snow Leopard.

What pattern did you choose for secure erase? Write Zeroes one pass?
...and did you use Disk Utility or another utility?

TRIM isn't supported on RAID volumes.

Secure erase? really?

And 30% doesn't seem like did much good.

Normally I find MPG to be right on the mark, and the money, but not when it comes to reconditioning SSDs

http://macperformanceguide.com/Storage-SSD-Reconditioning.html

I think if you check carefully on the web there is a problem with removing data from SSDs because of the method that is used for leveling the write activity across the device.


.
.
| ___________________

Allan


___________________

Allan-

I am very interested in the subject of keeping SSDs in good shape for fast writing, especially "stupid" ones that do not have access to things like TRIM.

Any links you could provide for further research would be appreciated.

Some additional reading on the complications of data erasure on SSDs.

http://arstechnica.com/ask-ars/2011/03/ask-ars-how-can-i-safely-erase-the-data-f rom-my-ssd-drive.ars

https://docs.google.com/viewer?url=http%3A%2F%2Fwww.usenix.org%2Fevents%2Ffast11 %2Ftech%2Ffull_papers%2FWei.pdf

.
.
| ___________________

Allan


___________________

Toocool4 wrote:
I think the 2 SSD in RAID 0 on a Mac Pro is wasted, as you are limited by the SATA II interface in the Mac Pro. The throughput on both drives is more than the SATA II interface can handle, if you want to get the true potential of the drives you will need a descent dedicated RAID card.

This is, luckily, complete nonsense. I've been using a 2 x Intel 80GB SSD RAID0 for ages now and just today I upgraded to 3 x 120GB Intel (both the classic G2 Postville ones), that's how I found this page, I was looking for some fresh info about the best stripe size.

I just measured sequential speeds, test file size was 500MB. Write speed (the X-25 weak spot) is up to 116MB/s solo and up to 336MB/s with RAID, read speeds are up to 264 MB/s solo and whooping 757MB/s with RAID.

Do the math, the speed gain is nearly linear. And absolutely no slowdown from the SATA 3Gb limit in my 2006 Mac Pro. Actually using more than one SSD is the only way to overcome that limit, current SATA 6Gb models like the Vertex 3 won't reach their potential speed in my Mac Pro.

So to get maximum speed it's a great idea to use an SSD-RAID0. Don't forget to backup, though, if just one SSD dies you will lose everything. But that risk is far lower with quality SSDs (that's why I stayed with the trusty Intels) as it is (was?) with magnetic hard disks and all their mechanics. And you don't have to endure the noise of all those drives.

Have no problem agreeing with the data but in the real world>touch and feel through experience I just don't see the benefit as worth using up valuable sata ports to raid 0. Yes some function are faster but not enough to use up those ports.