Nimble Storage Performance Testing

As you have read from a previous blog post here I take issue with storage vendors not recognizing the write IO issues that happen with Desktop Virtualization solutions.  I’ve been working recently with a storage vendor Nimble Storage who uses SSD to accelerate the reads and uses inline compression and intelligence to write the write IO sequentially to high capacity lower performing disk.  Nimble Storage refers to their architecture as CASL, which stands for Cache Accelerated Sequential Layout.  If this technology all works what Nimbe Storage provides is much better performance than standard spinning disk solutions but with high amounts of usable capacity usually only found in less performing spinning disk solutions. I spoke to one of the technical marketing folks a few weeks back and what’s interesting is that all write IO goes to spinning disk.  Reads come from SSD, but mainly SSD is there for random reads, sequential reads might still come straight from spinning disk, writes go to spinning disk and are optimized and written down to disk sequentially…some write IO goes to SSD too if the array believes that data might be read soon. Nimble folks…if I mangled this description of CASL I’m sorry.

More information about Nimble Storage was also written up by a fellow BriForum 2011 Chicago speaker Dwayne Lessner here:

While my test is far from scientific, or for that matter even fully simulating a VDI workload it does illustrate that the sequential write IO they preach really does work. VDI disk IO block sizes in real life are variable, with lots of IO happening at the 4K block sizes but it’s also common to see larger block sizes. I did two tests, one at 80% write, 20% read – 75% random and one at 50% read, 50% write, 75% random.  This Iometer test was loosely based on some settings that Jim Moyle outlined in his article here

My disk for all tests was 32000000 sectors as this was the size of the disk in the OpenPerformanceTest32 found here for VMware storage benchmarking that others have used and posted results to these two threads and I used 64 outstanding IO per target and a single worker thread for all tests.

The test environment consisted of a Nimble CS-240 connected via two 1Gb links. A Cisco C200 M1 rackmount server with dual E5520 processors, 24 GB of RAM, and two 1Gb NICs was used. The virtual machine that Iometer was run from was a 4 vCPU (yes i know I was only using a single worker thread) with 4GB of RAM.

This test is far from real world but it does illustrate some significant sustained IO against what is traditionally thought to be slow 7.2K RPM disks…just goes to show what IO optimization can do for you…but i don’t have to tell companies like Nimble Storage or Atlantis Computing that.

OpenPerformanceTest32.icf from

also posted on community site

Test name Latency Avg iops Avg MBps cpu load
Max Throughput-100%Read 15.89 3779 118 0%
RealLife-60%Rand-65%Read 4.44 12898 100 1%
Max Throughput-50%Read 11.50 5027 157 0%
Random-8k-70%Read 3.78 15343 119 0%

VDI workload 80/20 (write/read 75% random)

Test name Latency Avg iops Avg MBps cpu load
VDI 3.02 21193 82 0%

VDI workload 50/50 (write/read 75% random)

Test name Latency Avg iops Avg MBps cpu load
VDI 2.40 26671 104 0%

3 thoughts on “Nimble Storage Performance Testing

  1. Do you mind if I quote a few of your articles as long as I provide credit and sources back to your website?
    My blog site is in the very same area of interest as yours and my users would definitely benefit from some
    of the information you present here. Please let me know if this ok with you.
    Many thanks!

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s