Enterprise MLC (eMLC), an emerging class of MLC technology, uses wear leveling, bad-block mapping, improved error correction and write amplification techniques to improve the life of MLC flash storage. But, how do they work and what improvements do they make to the life of solid-state flash storage?
In this interview, SearchStorage.co.UK Bureau Chief Antony Adshead speaks with Chris Evans, an independent consultant with Langton Blue, about eMLC and the improvements vendors are doing to turn MLC into enterprise MLC.
Listen to the podcast on enterprise MLC and solid-state storage for the enterprise or read the transcript.
Download for later:
- Internet Explorer: Right Click > Save Target As
- Firefox: Right Click > Save Link As
SearchStorage.co.UK: What is eMLC?
Evans: Let’s deal with eMLC as an acronym and what it means. The “e” stands for “enterprise,” and “MLC” stands for “multi-level cell.” This is a type of solid-state device where an individual cell of memory is used to store more than one bit of information.
You can compare that to SLC, which is single-level cell, where a single piece of information is stored in a single cell.
The difference between the two is that MLC is much cheaper to make compared to SLC, and consequently people want to try and use MLC where they can because SLC per gigabyte is a lot more expensive to deploy.
However, there’s a major difference between the two. MLC is much less reliable than SLC due to the way the technology works. And one of the features that affects that is if you try and store multiple bit values in a single cell, that has an impact on the quality of those cells.
Now, the “e” part comes in because we’re now getting [MLC] devices classed for enterprise use, where manufacturers are making changes to improve the quality, the lifetime and get them to a position where they can be classed as good enough to use in the enterprise.
SearchStorage.co.UK: What types improvements are vendors making to make MLC more reliable for enterprise use?
Evans: There are a whole host of features being added to these devices. Let’s go through a few of them.
The first one is wear leveling. A solid-state device isn’t written to in an even fashion. If you’ve got an area with a file or block of data that’s continually being hit by the host, then that particular section is going to wear out quicker, so one of the features that gets used is wear leveling. What that does is dynamically relocate blocks to other parts of the solid-state device to make sure that write activity is consistent across the whole unit. It means you get a more consistent lifetime for the device; it lasts longer because you’re writing across the whole device.
Another [one] is called bad-block mapping. There is where if you have an area that fails on a particular SSD, you can remap that bad block to another spare area that’s been overprovisioned in the SSD. That is where you rewrite those blocks of data, so, effectively, you’re giving yourself some extra capacity so if any blocks fail you, can remap those bad blocks to this extra space.
Another technique is to improve error correction. If we’re reading and writing to these devices and we see bit errors, the [error correction] calculation that you would normally expect to be done in memory is done on SSD as well, and that helps to prolong the life a bit further.
Another feature is called write amplification. When you write to an SSD, one of the problems you have is if you are storing multiple bits in the same cell, then you may have to read that cell to get that data out, change the bit that you’re setting, then write that data again. Effectively, what you’re doing is writing more data physically to the SSD than you are writing from the host, and that’s simply because of the way that data is stored.
Write amplification is just a way of caching that data so that when you do write to disk, you’re writing in a more consistent fashion, so you’re not overwriting and you’re doing less I/O to the physical SSD. Again, it’s all about trying to reduce the amount of writing you’re doing to the device to improve its lifetime.
Those are a handful of the features the vendors are putting into their technology, but of course we’re sure to see more developments, including proprietary developments that vendors put into their technology.
This was first published in September 2011