Do they have a shorter lifetime than regular drives due to limited write cycles?
Generally no.
All SSD controllers use wear leveling routines that prevent certain areas of the NAND from burning out. Also, all SSD drives are over-provisioned by a percentage that varies depending on the controller system used. So, when you buy a 128GB drive, it actually contains 135-160GB of NAND flash, with the remainder reserved by the controller to use for wear leveling (and performance reasons in some cases). Also, the "better" controller (Intel, Sandforce, and Indilinx) based drives use a variety of other compression and caching schemes to minimize writes to the drives. For example, the way NAND Flash is structured requires that the smallest write possible to the "disk" is the block level. This means that if you only need to write 12k of data, you actually have to read and then re-write a full 512k block (this is called
write amplification). So, to counteract this, modern SSD disks employ sophisticated caching schemes where writes are done to onboard DRAM, which stores up writes until block-aligned writes can be done, minimizing the "wasted re-write cycles" (this also speeds writes up, conveniently). Modern controllers also use sophisticated compression schemes and predictive analysis to further improve write performance and wear leveling.
The main concern really only appears in enterprise use-cases, specifically when used in a high-throughput database server, like those used to serve a large database-driven website (like Amazon). The issue with these systems is that, in some usage scenarios, relational databases perform many, many, many
small writes to the drives. In everyday use, these tiny writes are handled seamlessly by the caching system, but if there are thousands of them in a sustained day-in-day-out fashion, the inefficiencies inherent to the caching system designs start to surface (even if you try really hard, you can't always align all of those writes to 512k blocks, so there is still some waste). That's why enterprise class drives have typically used more-expensive SLC NAND, rather than the much more dense MLC NAND (until this coming generation, anyway). SLC Flash is faster and has dramatically improved write endurance. Unfortunately, it is also
way more expensive (because MLC allows for much more storage space in the same amount of silicon). That's why "enterprise-class" SSD drives are often
much more expensive for much smaller capacities. This was once a much more serious problem than it is now. As controllers have improved and new compression and caching techniques have been developed, drive longevity has increased exponentially. That's why most of the manufacturers are moving to use MLC NAND with their next generation drives, even those targeted at the enterprise market.
However, these types of uses are absolutely edge-cases. In the vast majority of drive uses (even enterprise class uses), reads outweigh writes by a dramatic margin (5 or 6 to one for most database servers), and in home uses, the ratio is even larger. Home users also often tend to perform large (easy to cache) writes all at once, rather than lots of tiny writes spread out over time. You write to your disk a bunch when you install a new game or software package, download a file, or similar behaviors, but in general everyday use, writes are
fairly rare (mostly just logging in Windows and updating settings and things like that). Windows' built-in write caching scheme (designed to speed up writes to slow magnetic media) also helps the drive prevent premature burn-out. Sophisticated analysis of modern drives using good controllers now predicts lifespans ranging from 20 - 50 years for most "average" users. In other words, they blow magnetic drives out of the water.
If you want to learn more, here is a great place to start (though he is certainly focused on academic and enterprise use-cases):
http://www.storagesearch.com/ssdmyths-endurance.html#endure
Author
Topic: Does SSD drive improve MC searching? (Read 3203 times)