Random Read/Write Speed

The four corners of SSD performance are as follows: random read, random write, sequential read and sequential write speed. Random accesses are generally small in size, while sequential accesses tend to be larger and thus we have the four Iometer tests we use in all of our reviews.

Our first test writes 4KB in a completely random pattern over an 8GB space of the drive to simulate the sort of random access that you'd see on an OS drive (even this is more stressful than a normal desktop user would see). I perform three concurrent IOs and run the test for 3 minutes. The results reported are in average MB/s over the entire time. We use both standard pseudo randomly generated data for each write as well as fully random data to show you both the maximum and minimum performance offered by SandForce based drives in these tests. The average performance of SF drives will likely be somewhere in between the two values for each drive you see in the graphs. For an understanding of why this matters, read our original SandForce article.

Iometer - 4KB Random Write, 8GB LBA Space, QD=3

Here you can see the cap on 4KB random writes alive and well. As I've mentioned in previous articles, we're finally good enough when it comes to 4KB random write performance for current desktop workloads - so despite the cap you won't see any real world impact of it in our tests.

Many of you have asked for random write performance at higher queue depths. What I have below is our 4KB random write test performed at a queue depth of 32 instead of 3. While the vast majority of desktop usage models experience queue depths of 0 - 5, higher depths are possible in heavy I/O (and multi-user) workloads:

Iometer - 4KB Random Write, 8GB LBA Space, QD=32

Iometer - 4KB Random Read, QD=3

Sequential Read/Write Speed

To measure sequential performance I ran a 1 minute long 128KB sequential test over the entire span of the drive at a queue depth of 1. The results reported are in average MB/s over the entire test length.

Iometer - 128KB Sequential Write

Iometer - 128KB Sequential Read

Introduction AnandTech Storage Bench 2011
POST A COMMENT

45 Comments

View All Comments

  • wombat2k - Thursday, May 05, 2011 - link

    "The two come with comparable warranties which brings the decision down to pricing, where OCZ currently has a $20 advantage. "

    I don't think that's the complete picture. OWC is more of a Mac shop so provides ways to easily update your firmware from a Mac. OCZ relies on you burning an ISO file. This might have changed for the Vertex 3, but I doubt it. OWC sells directly through macsales.com, and I hear their service is pretty good, so if you're a Mac owner, you would tend to naturally gravitate around the OWC solution, even with the price markup.

    Disclaimer:
    * Not a OWC user yet, but considering them *
    Reply
  • NCM - Thursday, May 05, 2011 - link

    The table on page 1 shows the same specs (other than price...) for all three versions. Reply
  • Shadowmaster625 - Thursday, May 05, 2011 - link

    I paid $95 for my 60GB Agility 2. I cant fathom paying 3 times the money for 2X the capacity and a negligable increase in noticable performance. Reply
  • sor - Thursday, May 05, 2011 - link

    I for one hope Anand will figure out what's up with the OCZ 'max iops' Vertex 3. Seems to me that he was promised that his test sample's performance would be the same as the shipping Vertex 3, and now we have a Vertex 3 with and without a firmware cap, and the firmware capped one was shipped first. Perhaps that's not the case and the 'max iops' is some new tweaks above and beyond, but it's been out a few weeks and so far I've yet to read anything about it anywhere. Reply
  • Mr Perfect - Thursday, May 05, 2011 - link

    Isn't anyone making 64GB drives this generation? With the Z68 chipset reportedly only supporting up to 64G for the SSD/HD hybrid drive configuration, they're going to be in demand. Reply
  • Stargrazer - Thursday, May 05, 2011 - link

    Interestingly enough the 4KB random write cap isn't enough to impact any of our real world tests.


    I would argue that you don't *have* any real world tests.

    Sure, the AnandTech Storage Bench tests are *based on* real world workloads, but since you're playing them back at a faster than normal rate, they stop being real world tests. Just take the case of playing back a movie for example. You have a lot of reads there, but you don't get any increase in performance from being able to perform those reads faster than what is necessary to keep your buffers filled (non-empty really). In addition to this, many of the writes that are performed during the tests should be non-blocking, so increasing the write performance would in many cases not lead to any actual real world performance increases (you'd "free up" the drive faster, but that's mostly a benefit when there's other stuff that needs to be done).

    There is presumably a lot of stuff in there that *is* limited by your IO speed, but it's all mixed up with stuff that *isn't*, so you can't tell which speed increases give a real world improvement, and which do not. You simply can't tell how much real world performance an increase in results represent, the relationship most certainly will not be linear (e.g. a 10 point difference could have different meaning depending on how high the values are), and you can't even conclusively tell that a drive that gets a higher score actually performs better in any real world sense.

    The tests do provide some information, but it's not something that tells you how much you would benefit from upgrading your drive.

    I personally think that it would be interesting if you would provide some *actual* real world tests too, so that people could tell if they would actually see some real world differences between different SSDs. Maybe some program/game loading/zoning tests?

    At least people would then be able to judge if the difference they see would be significant enough (for them) to warrant an upgrade. You know how to value a 10s difference, but how much is a 10 MB/s difference in the storage bench worth?
    Reply
  • Anand Lal Shimpi - Thursday, May 05, 2011 - link

    Note that both the 2011 and 2010 Storage Bench do playback in real time. The 2011 benches in particular even preserve idle times properly so all that's sped up are actual I/O requests. You are correct in that this will speed up things like decoding video, however note that many video players already do a lot of read ahead and pre-decode on frames in order to avoid stuttering. Note that very few of the IOs in these tests are for things like video decode so I don't believe they're biasing the results too much.

    You are correct in that we're focusing exclusively on the I/O aspect of performance. And that a 10% increase in I/O performance won't result in a 10% increase in system performance (except in IO bound tasks).

    I've toyed with doing timing based tests, the issue is that modern SSDs don't show any difference when measuring the launch time of a single application. It's really under heavy multitasking and behavior over time that they differ from one another. Both of these types of tests are very difficult to time in a repeatable fashion, which is why we turn to our trace based performance tools.

    I do agree that there's a need for some perspective on these performance improvements, which is what I tried to do by including disk busy time in our 2011 results. There you can see that over the course of a 3 hour use period (for example in the case of the heavy workload) that one drive may shave off x number of seconds vs another drive. How annoying those seconds are is really up to the end user.

    Personally I find that there are three categories that drives fall into these days. There are those that offer performance around that of an X25-M G2, the next group is around the Vertex 2 and the final group is the 6Gbps 240GB Vertex 3. I feel like there's a noticeable difference going from any one of those groups to the other, but going from an X25-M G2 to something that's slower than the Vertex 2 makes less sense.

    I hope this helps :)

    Take care,
    Anand
    Reply
  • seapeople - Thursday, May 05, 2011 - link

    Considering that it's now a given that someone will ask for a graph of loading an application on the comments of any new SSD article, I think you should just pick 5 or 10 current drives and show everyone a graph that spans from 5.2 seconds to 5.5 seconds (or whatever it is) with a tag line of "There, are you happy? Now stop asking about this!" Reply
  • AnnonymousCoward - Thursday, May 05, 2011 - link

    Real world tests shouldn't be a "There, are you happy?" afterthought. Real world is all that matters.

    Let me refer back to this classic example: http://tinyurl.com/yamfwmg . RAID0 was 20-38% faster in IOPS, and in the time-based comparison it was equal or slightly slower. Anand concluded "RAID-0 arrays will win you just about any benchmark, but they'll deliver virtually nothing more than that for real world desktop performance."

    Do you buy an SSD to use it, or to sit there and run benchmarks?
    Reply
  • Stargrazer - Thursday, May 05, 2011 - link

    The chipset drivers are listed as being Intel 9.1.1.1015. Aren't those drivers from December 2009? Do they work well for H67?

    I can understand if you want to keep the same drivers for consistency in the tests, but have you checked if there are any significant performance benefits from using more recent drivers? If there are any changes, it could be interesting to see how it affects some of the more recent drives.
    Reply

Log in

Don't have an account? Sign up now