Login VSI Analysis of HPE SimpliVity 2600 RA
Congratulations to HPE SimpliVity for another successful reference architecture, validated by Login VSI. This hyperconverged (HCI) reference architecture is based on HPE’s latest 2600 architecture, leveraging the SimpliVity XL170r nodes. In just 4U this RA demonstrates high performance for 700 and 1000 Knowledge Workers. Think about that… you could almost fill a room with that many laptops. Moore’s law is definitely in effect here as SimpliVity demonstrates the Intel Xeon 6152 Skylake server architecture, capable of up to 8 nodes in that 4U footprint.
The configuration chosen for these tests is a four-node HPE SimpliVity XL170r hyperconverged infrastructure host deployment to support 700 users and six of the same nodes to support 1000 users. The four-node environment was configured as a 4+0 HPE OmniStack Federation. Similarly, the six-node environment was deployed as a 6+0. Each host has two 22-core Intel® Xeon® Gold 6152 CPUs and 464 GB of usable memory. For more details about the reference architecture check out the document itself: https://h20195.www2.hpe.com/v2/Getdocument.aspx?docname=a00049932enw
Figure 1. HPE SimpliVity 2600 logical design—700 users
Figure 2. HPE SimpliVity 2600 logical design—1000 users
Let’s look at the performance profile in this RA. With so much power in this little package, it’s no surprise that they didn’t hit VSImax. In the HCI community this is common, so we’ll look at a couple different Login VSI metrics—VSIbase and VSImax Average (VSIavg). First, SimpliVity is careful to allocate this infrastructure by not oversubscribing CPU or Memory. With fixed memory you get a fixed number of desktops per host. With a fixed number of desktops per host, on a tiny turbocharged system, you don’t hit VSImax because you have plenty of headroom to offer performance. So, what do you do with that headroom? You offer a great user experience through less latency (more responsiveness).
About the Login VSI Performance numbers
About the VSIbase
For this RA the VSIbase for the 4-node, 700 session configuration is 666ms. The VSIbase for the 6-node, 1000 session system is 638ms. You can immediately see the benefit of the 2 additional nodes as the end-user experienced a reduction in latency. At less than 800, these numbers qualify as a “very good” Login VSI Baseline Score.
About the VSIavg
With VSIavg we look at how that latency changes as the workload increases—more work equals more latency. The following is a table highlighting these factors for this reference architecture:
* The max sessions in this test was 700. VSIavg at 700 sessions was 1498
** VSIavg 1000 here is based on a 999 session value
Lastly, it should be noted that SimpliVity had 0 stuck sessions on 1000 user test and only 1 stuck session on the 700 user test. Nice job.
VSI conclusion and additional notes
Using the latest and greatest in their gear, we certainly see the latency decrease and the slope at which the latency grows is very slight, leaving less than a second’s difference between the baseline and latency at 700 and 1000 sessions. Note that at 4 nodes we see the latency start to grow a little more rapidly than the very gentle slope using 6 nodes.
Some other points of interest:
- Win 10 LTSB – SimpliVity uses Win10 LTSB which helps with some of the PDF interactions and File Type Associations with Microsoft Edge.
- Horizon Client – SimpliVity uses Direct-Desktop Connections for their testing. This helps to scale the launcher environment and stabilize testing. If the Horizon client were to be used there would likely be an increased demand on the system resources. This may or may not have an impact on session count.
- Spectre/Meltdown – It looks like some of the components in this configuration may have mitigations in place for the hypervisor and processor microcode, and the performance is still really good. Windows 10 LTSB wouldn’t have the patches since it is from 2016. With the latest Intel processor technology, hypervisor versions and OS release, the performance impact from Spectre and Meltdown is expected to be very low. I wrote a little more about this in my Spectre & Meltdown blog.
- Win 10 and Office 2016 – Using the latest OS and Office builds are great examples for the enterprises looking to learn from these RAs. These builds carry more of an impact than their predecessors, so it is nice to see such good performance numbers using the latest software.