Performance vs. Transfer Size

ATTO is a useful tool for quickly benchmarking performance across various transfer sizes. You can get the complete data set in Bench. The P3700 does a good job of scaling performance, although Samsung definitely holds an advantage at some of the smaller transfer sizes when it comes to reads.

The story changes a bit if we look at sequential writes:

Note the ultra low performance at really small transfer sizes (512B - 2KB). To better showcase what I was seeing, I cropped out the larger transfers and just focused on the first few datapoints:

The P3700's performance is really low when it comes to ultra small sequential write transfers. Once you hit 4KB the P3700's performance skyrockets, but up until that point it's substantially slower than even a high end SATA drive. As very few workloads actually care about performance down here I suspect it's something that Intel never optimized for.

Random & Sequential Performance Final Words
Comments Locked

36 Comments

View All Comments

  • [-Stash-] - Tuesday, June 10, 2014 - link

    Will be great to see how the 35 and 36 perform in the client workloads – really quite excited about these.
  • romrunning - Tuesday, June 10, 2014 - link

    In the 25% OP 4KB Random Write (QD32) graph, the Samsung XP941 showed a massive drop between its 512GB and 384GB (25% spare area) results. From 30k down to 5k - is that an anomaly?

    Also, what's with the vertical scale showing 1, 10, and 100 ranges? That forces all of the data points into a much smaller range, and it's visually not as informative.
  • Kristian Vättö - Tuesday, June 10, 2014 - link

    Are you sure you are not mixing up the graphs? With 25% spare area, the results are better (~30K IOPS), which is how things should be.
  • romrunning - Tuesday, June 10, 2014 - link

    I see - the two charts on the bottom of page one are same but with different vertical scales (as explained so well by DanNeely below).

    Yes, I suppose instead of calling it a drop, you could say it's actually a rise from 5k to 30k IOPs when you go 25% spare area. It seems Samsung drives especially like extra spare area.
  • DanNeely - Tuesday, June 10, 2014 - link

    It's called a log scale; and generally is the better choice when you've got values spanning more than 2 or 3 orders of magnitude (ie more than 100x or 1000x). A linear scale would crush everything but the largest values into a tiny area at the bottom. With the highly consistent performance of the DC P3700 it's not as big a concern; but with the less consistent behavior of most consumer or older enterprise drives it's the only good way to see the behavior. Look at the Samsung 840; it has peak IOPS of ~100,000; drops to a steady state behavior mostly between ~3000-9000 but with occasional excursions as low as ~110.

    Also, the tests are being done to show consistency of performance, a log scale makes seeing that much easier because the same vertical distance between min and max always represents the same multiple (eg 2x 3x 10x) between the values; a linear scale would mask inconsistency in slower drives while exaggerating it in higher performing ones because because 20000-22000 is a much larger interval on a linear scale than 10-1000 despite the former being a highly consistent drive with 10% variation and the latter having a 100x variation.
  • ryanjacoby2 - Tuesday, June 10, 2014 - link

    I'm enjoying the pivotchart type graphs you guys transitioned to a while ago. Makes readability and comparisons a breeze without the clutter, thanks for the change!
  • uruturu - Tuesday, June 10, 2014 - link

    why don't you test the same drives also with 25% OP in storage bench 2013?
    samsung 840 pro (25% OP) vs sandisk extreme II (25% OP)...factory settings seem to penalize some drives.
  • morganf - Tuesday, June 10, 2014 - link

    I thought the same thing. The 2013 storage bench writes an insane amount of data in a short time. If someone really cares about performance with such a heavy write-load, then they would certainly overprovision their SSD. The P3700 obviously already comes from the factory with a lot of OP. It only makes sense to also OP the Samsung 840 Pro and other drives, as a point of comparison, when running the 2013 Storage bench.

    To be clear, I mean to run the 2013 Storage Bench on each SSD with and without OP.
  • eanazag - Tuesday, June 10, 2014 - link

    When's the Intel giveaway for the P3700?
  • extide - Tuesday, June 10, 2014 - link

    LOL! Yeah, SIGN ME UP!! Haha ;)

Log in

Don't have an account? Sign up now