Home
| Calendar
| Mail Lists
| List Archives
| Desktop SIG
| Hardware Hacking SIG
Wiki | Flickr | PicasaWeb | Video | Maps & Directions | Installfests | Keysignings Linux Cafe | Meeting Notes | Linux Links | Bling | About BLU |
On 6/3/2012 1:44 AM, Tom Metro wrote: > The Flash chips have commodity volume production in their favor (which > can be significant). If the SSD's with high-end controllers are also > using high-end Flash chips that are produced in smaller volumes, then > this advantage is nullified. There ain't no such thing as "high end" flash chips in the way that you're thinking. High end in the NAND flash arena is a matter of sustained write performance. That's it. Not quality. Not reliability. Speed. The thing is, flash read and write speeds have a technical plateau so you won't see much difference in the raw performance of different manufacturers' chips of a given generation. The difference is how different vendors optimize their controllers. Here is an important point: SandForce doesn't make SSDs. They make SSD controllers. You can stick pretty much any manufacturer's MLC flash chip on a SandForce controller and get reasonable performance out of it. So how do the likes of EMC and Violin go so much faster for so much longer (5-10 years vs. 6-24 months for commodity SSD)? Several things. They use battery-backed DRAM cache on their controllers. One of the reasons why SandForce SSD controllers are relatively inexpensive is because they don't use DRAM cache. Throw a big DRAM cache in front of your flash chips and depending on I/O load you can reap a big performance boost and reduce flash write cycles. The typical consumer-grade SSD has a single big flash chip in it. An enterprise class SSD has banks of chips arranged in something similar to a RAID 0 configuration to reap the benefits of more "spindles". Enterprise-class SSDs also layer error correction onto the write process to ensure data integrity. Typical consumer-grade flash devices are over-provisioned by 7-8% to provide a quantity of spare cells to cover those that are bad at manufacture time and those that die during use. Enterprise-class SSDs can have 100-200% over-provisioning or more. Enterprise-class SSDs are more likely to use that spare capacity before cells fail because a failed cell is a read/write fault is a performance hit. All of this comes at a premium. EMC will gladly sell you a 120GB flash-based SSD for 100 times the cost of the 120GB SSD you'd get at MicroCenter. I leave the rest to the reader, but I suggest comparing STEC's literature to your favorite commodity SSD vendor's literature. -- Rich P.
BLU is a member of BostonUserGroups | |
We also thank MIT for the use of their facilities. |