The Cleaning Lady and Write Amplification

Imagine you’re running a cafeteria. This is the real world and your cafeteria has a finite number of plates, say 200 for the entire cafeteria. Your cafeteria is open for dinner and over the course of the night you may serve a total of 1000 people. The number of guests outnumbers the total number of plates 5-to-1, thankfully they don’t all eat at once.

You’ve got a dishwasher who cleans the dirty dishes as the tables are bussed and then puts them in a pile of clean dishes for the servers to use as new diners arrive.

Pretty basic, right? That’s how an SSD works.

Remember the rules: you can read from and write to pages, but you must erase entire blocks at a time. If a block is full of invalid pages (files that have been overwritten at the file system level for example), it must be erased before it can be written to.

All SSDs have a dishwasher of sorts, except instead of cleaning dishes, its job is to clean NAND blocks and prep them for use. The cleaning algorithms don’t really kick in when the drive is new, but put a few days, weeks or months of use on the drive and cleaning will become a regular part of its routine.

Remember this picture?

It (roughly) describes what happens when you go to write a page of data to a block that’s full of both valid and invalid pages.

In actuality the write happens more like this. A new block is allocated, valid data is copied to the new block (including the data you wish to write), the old block is sent for cleaning and emerges completely wiped. The old block is added to the pool of empty blocks. As the controller needs them, blocks are pulled from this pool, used, and the old blocks are recycled in here.

IBM's Zurich Research Laboratory actually made a wonderful diagram of how this works, but it's a bit more complicated than I need it to be for my example here today so I've remade the diagram and simplified it a bit:

The diagram explains what I just outlined above. A write request comes in, a new block is allocated and used then added to the list of used blocks. The blocks with the least amount of valid data (or the most invalid data) are scheduled for garbage collection, cleaned and added to the free block pool.

We can actually see this in action if we look at write latencies:

Average write latencies for writing to an SSD, even with random data, are extremely low. But take a look at the max latencies:

While average latencies are very low, the max latencies are around 350x higher. They are still low compared to a mechanical hard disk, but what's going on to make the max latency so high? All of the cleaning and reorganization I've been talking about. It rarely makes a noticeable impact on performance (hence the ultra low average latencies), but this is an example of happening.

And this is where write amplification comes in.

In the diagram above we see another angle on what happens when a write comes in. A free block is used (when available) for the incoming write. That's not the only write that happens however, eventually you have to perform some garbage collection so you don't run out of free blocks. The block with the most invalid data is selected for cleaning; its data is copied to another block, after which the previous block is erased and added to the free block pool. In the diagram above you'll see the size of our write request on the left, but on the very right you'll see how much data was actually written when you take into account garbage collection. This inequality is called write amplification.


Intel claims very low write amplification on its drives, although over the lifespan of your drive a < 1.1 factor seems highly unlikely

The write amplification factor is the amount of data the SSD controller has to write in relation to the amount of data that the host controller wants to write. A write amplification factor of 1 is perfect, it means you wanted to write 1MB and the SSD’s controller wrote 1MB. A write amplification factor greater than 1 isn't desirable, but an unfortunate fact of life. The higher your write amplification, the quicker your drive will die and the lower its performance will be. Write amplification, bad.

Live Long and Prosper: The Logical Page Why SSDs Care About What You Write: Fragmentation & Write Combining
POST A COMMENT

296 Comments

View All Comments

  • albor - Friday, June 18, 2010 - link

    Hi,
    try RamDisk Plus 11 from SuperSpeed.
    (http://www.superspeed.com/desktop/ramdisk.php)
    I use it on Xp pro 32 bit with 30 GB OCZ Vertex and 8 GB RAM. All above 3.2 GB is configured for swap and temp. Works perfectly and no visible SSD performance degradation after about 10 months.
    Greetings.
    Reply
  • jmr3000 - Thursday, August 23, 2012 - link

    would explain to me how did you install it?

    the ssd as a second drive or did u install all the program on the ssd and use the hhd as a second?

    thanks in advance!

    jm
    Reply
  • marraco - Friday, August 13, 2010 - link

    SWAP file is one of the most important speed bottleneck on windows.

    it writes frequently to disk, so consumes the read write cicles of the disk, reducing his useful life.

    But you are not buying space storage when you buy SSD. You are buying speed, so it makes nosense to buy an expensive SSD, and then remove from it all the activities that need the speed and are bottlenecks.

    you buy a SSD to do the fastest SWAP. keep it on SSD.

    Also, drive indexing permanently does a lot of reads, but it does not matter if the disk is fast. Drive indexing is like a little local google. If you disable it, and then search for all the files with a given text on it, searching the entire disk takes longer than just read an indexing.

    Those activities consume the useful life of the disk, but at the time the disk gonna need replacement, (5 years, maybe), this disk gonna need replacement anyways, and new SSD gonna be dirty cheap, so it makes no sense to disable swap, temp files, and indexing.

    On other side, prefetch, superfetch and defrag most probably are better disabled under SSD.
    Reply
  • jimlocke - Wednesday, June 01, 2011 - link

    Pehu, I know this much after your posting, but I was curios what you ended up doing for swap.
    8GB of RAM almost seems like swap may not be needed, unless you have several memory-hogging apps open.
    Hope you still like your SSD. I'm looking at getting one soon, and agree this was an excellent article!
    -Jim
    Reply
  • krumme - Friday, October 09, 2009 - link

    First: I submit to the importance of random 4k for ssd.
    Second: Over the years I have highly valued the articles of Anand. It is remarkable to see such detailed and enthusiastic information.

    Now I have a few questions, following the general impact of this work.
    Some observations first:
    Following an article at Toms of a ssd article the 6 of September. The author was called a “Moron”, primarily as the random 4k synthetic bm was missing. The author was giving a different opinion on the indilinx vs intel, in the desktop sector, compared to Anand, giving more weight to transfer vs iops.
    In an discussion about a Kingston V-series review, one said that he would take the indilinx ssd any day because it was “750 times faster” – an argument based on iops.
    Another remark I have read several times is: “The Intel x25-M g2 is the only drive to get”.
    Another is: “I would like to buy the Dell xx, but it has an Samsung controller so its of no use”.

    I think it is time to stop, and make sure there is reason in what is happening for normal desktop use.

    Do we have blind test where to tell the difference between the Intel, Samsung and Indilinx?
    What is the actual real world bm fx. Win7 boot times for the 3 controlers?

    There is something called good enough. When is 4k random read/write time enough, to not notice any subjective improvement afterwards in win7? Could it be fx. 10M/s?

    The ssd is the best thing happening since 3d gfx, but I think we should enjoy what is happeing right now, because this time, could be the turning point where we soon are focusing on small differences.

    Anyone knows what´s the next big thing?
    Reply
  • bebby - Friday, October 30, 2009 - link

    Random 4k and its relevance for desktop use is really the main topic for me, too.
    If I assume that I only use the SSD for the OS and software and save my data on other, much less expensive HDDs, I doubt very much that this discussion is worth it. The Samsung SSD then suddenly looks not so bad at all and much cheaper...
    The next big thing for me would be an OS starting up in 5 seconds, like the OS we had in the 90s...making SSD obsolete.
    Reply
  • bebby - Friday, October 30, 2009 - link

    Random 4k and its relevance for desktop use is really the main topic for me, too.
    If I assume that I only use the SSD for the OS and software and save my data on other, much less expensive HDDs, I doubt very much that this discussion is worth it. The Samsung SSD then suddenly looks not so bad at all and much cheaper...
    The next big thing for me would be an OS starting up in 5 seconds, like the OS we had in the 90s...making SSD obsolete.
    Reply
  • marraco - Friday, August 13, 2010 - link

    I agree completely. I think that human beings can nottice the difference between a hard disk, and a non bad SSD, because the difference is too large, but over "good enough", it does not matter much if the SSD is 2X or 4X faster in 4Kb random R/W.

    But mine is just an opinion, and I don't have good data to test it. I would like to read an article with repeatable testing on human perception.
    Reply
  • SimesP - Wednesday, September 23, 2009 - link

    I haven't read all 254 comments (yet) but I'd like to add my thanks to everyone elses for the comprehensive and illuminating article. This, along with the previous AnandTech SSD articles have increased my understanding of SSD's immensely.

    Thanks again!
    Reply
  • ClemSnide - Friday, October 02, 2009 - link

    Anand,

    A couple of guys from HotHardware.com pointed me at your SSD article, and it allowed me to make an informed decision. Thanks!

    I wanted to speed up one game in particular (World of Warcraft) as well as routine OS tasks and web browsing. I think an SSD will do a bang-up job on at least the first two. The one I decided upon was the OCZ Agility 60 GB, which offers some growth room; I currently have 40 GB on my system drive. I know the Intel has better numbers, but I was able to get the OCZ for $156 after a rebate, which translates to decent performance at a price I can justify. (For the curious, it's available from TigerDirect for $184, and OCZ is giving a $30 rebate.)

    Even though my system build is still months away, this should be usable on my old clunker as well. Very nifty!
    Reply

Log in

Don't have an account? Sign up now