May 05, 2015
Atlantis HyperScale – Virtual Desktops get SuperPowered
Today marks the much anticipated launch of Atlantis HyperScale. In the build up to this date it’s been a busy time here at Atlantis HQ. One of the great things we’ve been doing is validating the new product with a VDI workload.
I guess the first thing I should do is answer the question, what on Earth is HyperScale?
Atlantis HyperScale is an all-flash, hyper-converged appliance that provides all the compute, storage and networking you need to deploy VDI, delivering higher performance and lower cost desktops than traditional PCs. Instead of piecing together costly and complex SAN, NAS or all-flash array storage with servers, Atlantis HyperScale is a low-footprint, fully integrated compute and storage solution that comes preconfigured with Citrix or VMware hypervisors. Within 60 minutes, a simple configuration process can set up the required infrastructure to deploy more than 600 virtual desktops per appliance. Atlantis HyperScale is easily scalable and runs on your preferred hardware platform, minimizing data center complexity. Also, the appliances come with a single point of support and maintenance with 3 year, 24x7x365 coverage and 4 hours parts replacement.
Although you can use Atlantis HyperScale for any workload, VDI is still one of the best use cases for the product and we wanted to validate exactly how good HyperScale is for Virtual Desktops.
Two of the most commonly used tools for synthetic testing of a desktop workload are Iometer and Login VSI, we used both to test.
While Iometer is a superb storage testing tool, it’s not until you configure it correctly for a VDI Workload that it becomes useful. With HyperScale you get multiple datastores to store your desktops on. Each datastore will give you the following performance:
As you can see, the headline performance is great at ~51,000 IOPS per volume; remember you get multiple volumes with HyperScale and you’ll get this performance per volume!
Reading the Iometer result, it’s not just the IOPS we should be looking at, it’s the ability to deliver this performance at such low latency on both read and write (this is 80% write and 80% random workload). Under normal workloads the latency will be even lower as we’ll see further down.
This test is now the industry standard for testing a virtual desktop workload and Atlantis uses it intensively, not just for reference architecture and performance testing but in QA for scale and burn in testing.
There are two main factors to look at when reading a VSI graph: the first is how low the latency is when the system is uncontended, i.e. how fast it is under normal circumstances. This is ‘VSIBase’ and on a normal modern system with shared storage, this might be in the region of 2000 ms. You can see in the below graph the response time on Atlantis HyperScale is under half that at 857ms.
The second factor is how quickly this performance deteriorates under load, giving the ‘VSIMax’ score. In most graphs, the results “hockey stick” quite quickly but, in the graph below we stay very flat. This means that, even when you load up the server to 600+ desktops, performance will be exactly the same to the user as if they were the only one on the hardware.
At no point are we close to crossing the VSI threshold, which means VSIMax isn’t close to being reached.
I have one last graph to share with you and it’s the Latency graph from vCenter. It effectively shows the datastore latency during the Login VSI test.
We saw above that Iometer latency, a storage stress test, is ~2.5ms. Below, the latency is in the microsecond range for almost the whole test.
The combination of low latency, amazing performance combined with a pre-tested BoM which contains standard hardware make Atlantis HyperScale the most compelling solution in terms of both compute and storage around.
Check out this reference architecture using Atlantis HyperScale: http://ow.ly/MR8rL
Current rating: 4.8 (6 ratings)