Myce.com Latest Updates

Beyond USB3, with UASP

Author

Wendy Robertson
Senior Administrator and Reviewer
Article posted 05 May 13 16:03

IOMeter test results

I/O Performance

IOMeter is probably the most versatile of all the synthetic benchmarks. Its ability to be configured to generate a multitude of different I/O traffic is unmatched. Another great feature of IOMeter, is the capability to test any storage metric that you can think of, providing you know how to configure the assignments. The reviewer also has complete control over things like queue depth, block size, whether the traffic is random, sequential, or even a mixture of both.

Partition alignment and sector boundaries

Windows 8, Windows 7, and Windows Vista will automatically align a partition to 4k boundaries during partition creation, Windows XP won’t. It is imperative that an SSD’s partition is aligned. Windows XP is also restricted to sector boundaries, while Windows 7 and 8 will use 4k boundaries if they can. The OCZ Vertex 3.20 series is 4k boundary aware, and will use these boundaries if possible. Of course it will also remap LBAs for compatibility with the sector boundaries so that the drive can be used with Windows XP.

IOMeter allows us to set the sector boundaries for conducting the tests, and I have therefore set the sector boundaries at 4K, which means the IOMeter tests are valid for Windows 7, Windows 8 and Windows Vista users. XP users will not be able to obtain such results.

All the IOMeter tests create a 10GB data set on the target drive, and each test is run for a duration of 3 minutes.


IOMeter 4K random write test with repeating data.

The first test involves creating continual 4KB random files on the target drive with IOMeter. I use a 4KB file size, as it is believed that Windows will create and modify many of this size of file constantly in the background during a typical Windows session. It is said that most 4K random writes take place at a queue depth of only one, and I have been requested to include this test in my reviews.

I will show screenshots of USB3 UASP and USB3 BOT mode. If you wish to view the screenshots of SATA 6Gbps results, then you can do so by viewing this page from a recent SSD review.

Queue depth 1

4k random write QD1
USB3 UASP mode – 4K random write (QD 1)

4K random write QD1
USB3 BOT mode (normal USB3) – 4K random write (QD 1)

4K random write QD1

As expected, native SATA 6Gbps is way out in front, and USB3 UASP mode is considerably faster than USB3 BOT mode.

Our next test involves creating continual 4KB random files on the target drive with IOMeter. I use a 4KB file size, as it is believed that Windows will create and modify many of this size of file constantly in the background during a typical Windows session. I will use queue depths of 4 and 16 for these tests.

Queue depth 4

4k random write QD4
USB3 UASP mode – 4K random write (QD 4)

4K random write QD4
USB3 BOT mode (normal USB3) – 4K random write (QD 4)

4K random write QD4

At a queue depth of 4, we can see USB3 UASP NCQ in action, where it is scaling reasonably well. Also notice that USB3 BOT mode without the benefit of NCQ hasn’t scaled at all from queue depth one.

Queue depth 16

4k random write QD16
USB3 UASP mode – 4K random write (QD 16)

4K random write QD16
USB3 BOT mode (normal USB3) – 4K random write (QD 16)

4K random write QD16

Although USB3 UASP mode has risen slightly at a queue depth of 16, the rise from a queue depth of 4 is not significant. USB3 BOT has not risen at all from queue depth one.

4K random write queue depth profile

For this test I used various queue depths from 1 – 16 to give you an idea how the performance varies at different queue depths. For a normal desktop user, with lightweight multitasking, the queue depth will rarely rise above 2. For heavy multitasking, the queue depth is unlikely to rise above a value of 8.

The results are shown below.

4K random write QD profile

As we can see, USB3 BOT mode does not scale at all with increasing queue depths. USB3 UASP mode does scale very well up to a queue depth of four, and from then on performance only increases marginally with increasing queue depths.

Below I present a table of the results in more detail.

4k random write QD table


IOMeter 4K random read test.

If there are many 4k files created, then that must also mean that many 4k files need to be read. This test measures 4k reading performance.

It is said that most 4K random reads take place at a queue depth of only one, and readers have requested that I include this test in my reviews.

Queue depth 1

4k random read QD1
USB3 UASP mode – 4K random read (QD 1)

4K random read QD1
USB3 BOT mode (normal USB3) – 4K random read (QD 1)

4K random read QD1

Again native SATA 6Gbps is way out in front, and this time USB3 BOT mode is a bit faster than USB3  mode.

Queue depth 4

4k random read QD4
USB3 UASP mode – 4K random read (QD 4)

4K random read QD4
USB3 BOT mode (normal USB3) – 4K random read (QD 4)

4K random read QD4

Once again SATA 6Gbps mode is well in ahead, but USB3 UASP mode is scaling well, and has left USB3 BOT mode in its wake.

Queue depth 16

4k random read QD16
USB3 UASP mode – 4K random read (QD 16)

4K random read QD16
USB3 BOT mode (normal USB3) – 4K random read (QD 16)

4K random read QD16

Once again there isn’t a huge leap in scaling from queue depth four in USB3 UASP mode. However, compared to USB3 BOT mode, UASP mode is performing very well.

4K random read queue depth profile.

This test shows how the review drive scales with increasing queue depths.

4K random read QD profile

Below I present a table of the results in more detail.

4k random read QD table

Once again we can see that USB3 UASP mode scales well up to a queue depth of four, and from then on things start to level off. Again we can also see that USB3 BOT mode is static, and doesn’t benefit at all from rising queue depths.


IOMeter 512KB write test with repeating data (dual threaded).

Sequential writing performance is also very important; in this test sequential writing performance is measured.


seq write QD2
USB3 UASP mode – 512K Sequential write

seq write QD2
USB3 BOT mode (normal USB3) – 512K Sequential write

seq write

USB3 UASP mode has reached a very respectable 414.49 MB/s, and USB3 BOT mode has also done reasonably well at 388.65 MB/s. However, SATA 6Gbps mode is more than 100 MB/s faster.

512K sequential write – Queue depth profile

While most sequential writes will rarely rise above a queue depth of two, it has been noted from SATA analyzer traces that with more demanding tasks, queue depths can rise very close to a queue depth of four. This is why I now include queue depth profiles for sequential read and write.

Please note that in the following graph, I do not have the lowest possible score set at zero. This is purely to allow the graphs to be easier to read, but starting with a lowest possible score other than zero, gives the impression that there are large differences between competing SSDs with regard to performance, so please keep this in mind.  

seq write QD profile
512K sequential write – Queue depth profile

Below I present a table of the results in more detail.

seq write QD table

As we can see from the above graph, USB3 BOT mode is static seeing no benefit at all from rising queue depths. USB3 UASP mode does get a boost from NCQ and rising queue depths, and manages a very respectable 416.29 MB/s at a queue depth of four.


IOMeter 512KB sequential read test (dual threaded).

This test measures 512k sequential reading performance.

seq read QD2
USB3 UASP mode – 512K sequential reading test

seq read QD2
USB3 BOT mode (normal USB3) – 512K sequential reading test

seq read

As you would expect, native SATA 6Gbps is well ahead, and once again USB3 UASP mode is faster than USB3 BOT mode.

512K sequential read – Queue depth profile

While most sequential reads will rarely rise above a queue depth of two, it has been noted from SATA analyzer traces that with more demanding tasks, queue depths can rise very close to a queue depth of four. This is why I now include queue depth profiles for sequential read and write.

Please note that in the following graph, I do not have the lowest possible score set at zero. This is purely to allow the graphs to be easier to read, but starting with a lowest possible score other than zero, gives the impression that there are large differences between competing SSDs with regard to performance, so please keep this in mind. 

seq read QD profile
512K sequential read – Queue depth profile

Below I present a table of the results in more detail.

seq read QD table

USB3 UASP mode scales extremely well, and manages 449.07 MB/s at a queue depth of four. USB3 BOT manages a little boost at a queue depth of two, but from there on its static.


IOMeter Workstation simulation (outstanding I/Os = 64).

When running applications you will find that there is a mixture of small random files, and larger sequential files, being created and read. Not only that, it isn’t just one file at a time. In this test I measure a simulated workstation pattern, with a queue depth of 64 (threaded).

Although this is something that you probably wouldn’t run from USB connected storage device, I thought it would be interesting to see how it performs over a USB3 connection.

Workstation
USB3 UASP mode – Workstation simulation

Workstation
USB3 BOT mode (normal USB3) – Workstation simulation

Workstation

Whilst USB3 UASP mode certainly can’t compete with an SSD connected to native Intel SATA 6Gbps, UASP mode is certainly much faster than USB3 BOT mode.


Summary

I think you would agree that USB3 UASP mode certainly has a lot to offer with regard to performance when compared to USB3 BOT mode (normal USB3). I have clearly demonstrated in the IOMeter tests that UASP mode scales pretty well up to a queue depth of four. It also won’t have escaped your notice that native SATA 6Gbps is by far the fastest by quite a long margin. SATA 6Gbps has far greater raw bandwidth to play with, and in most cases will scale well with increasing queue depths all the way up to 32 and beyond.

Of course, it isn’t just about raw bandwidth and queue depth scaling, there are other factors to consider, but I will leave that explanation to the final page of this article.

 

Now let’s head to the next page for some real world copying tests….

 

Get an alert

Receive an e-mail when we publish a new review

Like us

Follow Myce.com