Gizmodo 
Enter your username and password.
Tip your editors:
Editorial Director:
Brian Lam | | Twitter
Editor:
Jason Chen
| AIM | Twitter
Features Editor:
Wilson Rothman
| Twitter
Senior Contributing Editors:
Jesus Diaz
| AIM | Twitter
Mark Wilson, Reviews
| AIM | Twitter
Contributing Editors:
Matt Buchanan
| AIM | Twitter
Adam Frucci
| Twitter
Sean Fallon
| Twitter
Jack Loftus
| Twitter
John Herrman
| Twitter
Dan Nosowitz
Chris Mascari
Danny Allen
| Twitter
Rosa Golijan
| Twitter
Chris Jacob
Columnist:
Brendan I. Koerner
Interns:
Don Nguyen
Kyle VanHemert
Comment Account Questions:
Original material is licensed under a Creative Commons License permitting non-commercial sharing with attribution.
11/17/09
Until solid-state can compete on price with spinning discs, that seems like a good compromise. At some point, I'm going to update my HTPC that way, so it boots fast from a 60 GB or 80 GB SSD, but has a couple of 2 TB drives for all my... um... legally acquired media. *cough* #fusionioioxtreme
11/17/09
Hell, I wish my laptop had an SSD to boot from... #fusionioioxtreme
11/18/09
11/17/09
11/17/09
11/17/09
Certainly wank worthy, but, I'll still wait for them to be bootable. #fusionioioxtreme
11/17/09
11/17/09
@B-low: Doesn't support RAID, yet. #fusionioioxtreme
11/17/09
@B-low: It barely loses on total price. It's a different story on price per GB, where it gets trounced. #fusionioioxtreme
11/17/09
In Mechanical hard drives, Striping drives decreases seek time by by adding the additional read/write heads of other drives to the same data pool. Actual write time is hardly effected by a RAID0 configuration on most consumer level products, getting your biggest boost with only 2 drives. The effects of adding additional drives to improve write time drops off steeply after than. The theoretical advantage, and the functional advantage are very different. You need a bank of more than 50 drives to boost write times on SATA2 even 10% in real world server benchmarks. Read time on mechanical drives is much improved over write time, as it increases slightly with every drive added until you reach the limit of the bus.
SSDs don't have to mechanically move a head over the drive to find the area the data is stored on. With essentially no seek time, RAID0 doesn't improve performance on SSDs at all.
It is however used to combine data from multiple chips into one logical volume. An 80 GB drive is likely 10x 8GB chips, controlled by a raid controller (inside the drive) to function as one larger drive rather than 10 small drives.
By adding multiple SSDs to yet another RAID array , you can further increase the size of logical volumes, but the performance diminishes, adding additional latency to the data flow with every controller added. This is still notably faster than a single mechanical drive, but it is wrong to assume you can combine several slower SSDs to achieve the speeds of a faster SSD. To do so you would need the BUS speed of the controller to be faster than the system bus, and you'd only see a difference when the system bus is maxed out with a large number of simultaneous requests (like running a large server with hundreds of clients).
Striping SSDs does have its uses, but performance gains are not one of them (at least not yet). #fusionioioxtreme
11/17/09
[www.tomshardware.com] #fusionioioxtreme
11/17/09
Nope.
Their Benchmarks are of dives in RAID0. But the comparison shown in their findings are between different drives all together. To challenge the validity of my claim, you would need to benchmark the same drive connected straight to the SATA controller, and a striped series of that same model drive connected to the same controller.
That being said, there are RAID controllers that use the PCIe bus to connect SATA drives. Then you could push the read/write times of the array faster than that of a single drive. But not for under the cost of just buying a PCIe SSD that runs at the same max speed. You would get no further benefit from a striped series under those conditions than you would from a single drive, though you could get increased performace over simply using the standard SATA bus.
I understand why you would challenge that claim, there are conflicting reports on the subject out there. There are new controller technologies in the works that will provide better access to the system bus, and buffer more requests in faster non-volatile memory to improve performance. In those cases a small (64MB) super fast SSD exists on the controller as a buffer that feeds the attached drives. This wont be available to consumers until mid next year, and initially only on select motherboards. You can buy a card to do it now, but the cost is prohibitive to anyone who gawked at the $900 price tag in the article.
My Claim is this:
Putting 1 SSD on a current-gen SATA interface is just as fast as putting 2+ SSDs in a RAID0 configuration on that same interface.
If we compare different drives to one another, or different controllers, you will see some huge differences in speed. That's why the drive in the above article is priced so highly. #fusionioioxtreme
11/17/09
11/17/09
While an array of slower SSDs may not break the threshhold set by newer models on a faster bus, they can still improve performance-wise in an array up to the limitations set by the rest of the hardware, at which point, attempting further gains on the same configuration becomes moot.
Your point is valid. Most of the SSDs I work with are at the enterprise level, where they tend to perform at higher rates (Due to an internal RAID controller in each drive, managing an array of physical memory). In that case it makes no sense to further stripe them (unnecessary redundancy). But consumer models are still largely operating without that advantage, so they may not reach speeds where bottleknecking with the bus becomes an issue. In that case, you're absolutely right. #fusionioioxtreme
05/05/09
05/05/09
Just saying.
05/05/09
12/11/08
12/11/08
12/11/08
12/11/08
12/11/08
12/11/08
12/11/08
I'll take three!
12/11/08
12/11/08
I love technology. Certainly not as much as Zoltan or Le Trung though.
12/11/08
12/11/08
12/11/08