A VDI Supercharged on HPE SimpliVity 325 Gen 10

July 26, 2021

Congratulations to HPE SimpliVity!

This latest hyperconverged (HCI) result is based on the HPE SimpliVity 325 platform leveraging 2nd Gen AMD EPYC processors (7702P). This validation confirms that a single, tiny, little, 1U HPE SimpliVity 325 server demonstrates high performance for 200 Knowledge Workers per server. This is a high density solution that demonstrates some of the lowest response times we’ve seen. The validation also confirmed linear scaling, at 200 desktops per server, up to 8 systems in 8U providing high performance for 1600 Knowledge Workers.

Login VSI - Blog - A VDI Supercharged on HPE SimpliVity 325 Gen 10 - Image 1

Think about that… if you lined up the laptops side-to-side, that would be the length of almost 5 soccer fields, and this solution fits in 8U or 14 inches of rack space. Very impressive.

The configuration for this test is a 4-node server block and a 4-node compute block of HPE SimpliVity 325 Gen10 hyperconverged infrastructure host deployment that supports 200 users per host to support 1600 users. The eight-node environment was deployed as a 4+4 cluster. Each 1U host having one 2nd Gen AMD EPYC processor and 1 TB of memory.

Let’s look at the performance results from this validation. With so much compute and memory power in this little package, it’s no surprise that they didn’t hit VSImax (and SimpliVity hasn’t in past Validated reference architectures). This is common with solutions in the HCI category, but let’s look at a couple different Login VSI metrics—VSIbase and VSImax Average (VSIavg). For this validation the VSIbase for the 8-node, 1600 session configuration is 572ms. From the table and chart below, they didn’t hit VSImax of 1573ms even with 1600 sessions… NOT EVEN CLOSE! Impressive results for eight, small 1u HCI servers.

Another reference I like to compare is the difference between the baseline and the ending VSIavg. Below you’ll note that their VSIavg at 1600 sessions is 824ms, and the baseline is 592ms. That means each user only experienced a change in responsiveness of a fraction of a second (only 230ms) with the system completely full. Again, very impressive.