According to JEDEC specifications, MLC NAND flash can perform 10 000 writes before it starts to fail; SLC can perform around 100 000 writes. There’s no need to worry, however. If you have a 256GB MLC solid state drive, and copied around 7 gigs of data to the drive every day, it would take around 360 000 days – or 986 years – before each page has been written to 10 000 times, thanks to some clever wear levelling algorithms that get incorporated into SSDs. Each page on the SSD will get written to first, before the first page gets erased and re-written.

As explained previously, an SSD will get slower due to the pages filling with valid and invalid data. It is possible for a person to claim back performance by performing a secure erase using a tool like HDD ERASE – it will free every page on the drive giving you back your performance. And if you don’t feel like losing all your data, you can use Indilinx’s Wiper Tool.

OCZ’s Solid 2 uses 34nm Intel flash with an Indilinx controller.

OCZ’s Solid 2 uses 34nm Intel flash with an Indilinx controller.

A TRIM is as good as a holiday

What TRIM does is take the delete process one step further. Instead of the delete command ending with the OS, the command goes through to the SSD and the pointers to the data get deleted. So while the data is actually still there, you’ve told your SSD that you no longer care about it. The tool queries the OS for available LBAs, then commands your SSD to TRIM those LBAs, giving you a drive that’s virtually “as good as new”.  At the moment Windows 7 is the only Microsoft OS that supports TRIM. After Windows 7 launches (22/10/09), most SSD manufacturers intend to release TRIM-supporting firmware for their drives.

Indilinx is actually the first to officially support the ATA8-ACS2 TRIM command. Most new Indilinx-based SSDs have a 4KB random write speed of around 13.1MB/s; without TRIM support the 4KB random write speed drops to around 6.93. Once no longer new, with TRIM, you get around 12.9MB/s. For comparison, a conventional hard disk will sport speeds of around 0.3MB/s – 0.7MB/s.

OCZ’s Agility uses an Indilinx controller with either Intel 50nm flash or 40nm Toshiba flash.

OCZ’s Agility uses an Indilinx controller with either Intel 50nm flash or 40nm Toshiba flash.

JMicron controllers suck…

…and it’s all thanks to small file write latencies. The JMicron JMF602 controller suffers from slight stuttering. Your PC will pause for anywhere up to a second, then carry on as per normal. Even with JMicron’s updated JMF602B (with double the onboard cache) stuttering is still prominent. Average latencies are in the 500ms range – that’s around 100 times slower than your average desktop drive. Some manufacturers have put two JMF602B controllers in RAID configuration, to help bring average write latencies down to around 300ms, which is still far higher than Intel’s and Indilinx’s Barefoot controllers that have a 4KB write latency of 0.22ms and 0.34ms, respectively.

While not as bad, Samsung’s RBB controller is also lagging behind the cheaper Indilinx controller with regards to performance. While new 4KB random write performance may be 3x that of Western Digital’s VelociRaptor range, once each page has been written to, its performance drops even more. It’s just not worth it, especially when you take into account that the firmware isn’t user upgradeable.

More stuff like this: