1

With the new SSDs available there is the concept of wear leveling because they have limited write cycles (are read cycles limited too?)

How many write cycles should a drive achieve before wear leveling can be declared obsolete?

0

3 Answers 3

1

This question is difficult to answer because each SSD is different, especially comparing enterprise drives with home use drives. There are many factors that can affect the number of cycles.

In theory, wear is not affected by read cycles, so if a location cannot be written to, you should still be able to read it.

The following article says this:

We all have heard that there is a limit to the number of times you can write to a flash cell. This number varies dramatically depending on the flash vendor, what generation the flash technology is, and other issues. The claimed range is from more than 100,000 writes per cell to more than a million, but flash cell reliability is not the whole story for SSDs.

2
  • How many write cycles are needed for enterprise drives then?
    – John Smith
    Nov 26, 2010 at 23:06
  • I've updated my answer.
    – user3463
    Nov 26, 2010 at 23:25
4

So long as flash cells can wear out from over use, wear-leveling will be with us. As the lifespan of the cells increase it may get less and less complex, but it will still be present.

To give you an idea as to what kind of wear a typical enterprise-grade rotational magnetic drive can undergo, take the case of a 15 RPM drive being used in the heaviest storage environment: an OLT database. This thing gets written to all day long for years at a whack. Assuming random writes, this drive can undergo 500 write operations a second. This translates to 43.2 million write operations a day. Give it a 3 year operational lifetime, and it'll handle 47.34 billion writes.

We haven't talked sizes yet. Make it a 300GB drive (actually 271GB), and give it a sector size of 512B, and you get a total of 568,328,192 sectors. Assuming 100% random writes, each sector will have to withstand 83 writes over those 3 years. 83 writes.

However, writes are never 100% random. The real endurance of a single sector on a drive will be somewhere in the range between 83 and 47.34 billion writes, which covers the range from 100% random to a single sector getting written to continually for 3 years straight. Until such time as SSDs can endure billions of write operations per cell, complex wear-leveling will still be with us.

Of course, you don't go SSD for the size, you get it for the latency and the horribly huge IO-Ops you can pump to them. I've seen enterprise drives claiming to handle 30,000 I/O Ops per second, which is two orders of magnitude better than rotational media. Even then, for that hypothetical 300GB drive, for 100% random writes each cell will get written to only 4997 times over 3 years of operation vs 2.84 trillion times assuming only a single cell gets written to.

For Enterprise SSDs, we'll have wear-leveling so long as flash cells can wear out at all. For consumer grade SSDs it may go away once lifetimes hit the billions-of-writes range.

1

Are read cycles limited too?

Yes. It's just that the drive totals from writing first in which case reading the drive becomes kind of pointless since most filesystems are set up in a way that require you to log what you read or whatnot.

How many write cycles should a drive achieve before wear leveling can be declared obsolete?

Drives are worth their money if they hit 100k write cycles. They are insanely high-build if you can hit 200k cycles. If you want to find out how many yours can take, nuke it.

The EEEUser Wiki has posted a very useful guide showing how long 100 thousand cycles can last:

Table describing how long 100 thousand write cycles can last, apparently

So even if the effectiveness of the wear leveling is halved and were to write to it constantly at 10 MB/s it will still last a minimum of eight months theoretically.

Practically, you'd write to it on and off at say 10 MB/s from normal download rates which would effectively double the life of it from 8 months to maybe 30.

But wait, there's more: most filesystems have write caching so about one third of the data (the repetitive stuff) doesn't even have to be written.

So a good SSD would last you about four or five years if you don't abuse it.

But on the other hand, SSDs do still suck at writing. Intel has some great optimization tools for making SSDs last longer.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .