We’ve just come back from VMworld 2013 and I want to tell you all about something monumental that took place this year. Every VMworld there is an educational environment available to the show attendees called the Hands on Labs (HoL). We do something similar for EMC World and XtremIO powered the HoL there this year as well. The HoL allows attendees to conduct live training and demonstrations of VMware products, delivered on-demand through a VDI session. It’s an amazing showcase of VMware technology, not just for the students, but because of the infrastructure under the covers that makes it all work. It’s a massive undertaking that is planned out six months in advance, and the VMware HoL data center is the ultimate “dog fooding”, as in “eat your own dog food” because the HoL infrastructure team is doing exactly what VMware enables you to do in your own data center. And they do it at a truly impressive scale.
First, a little background. Each HoL course is a Virtual Application (vApp) – a container of multiple Virtual Machines (VMs). The courses range from about four VMs to as many as 20 in each vApp. Think about the vApp as a self contained environment where you can experience VMware products in action. Maybe you want to try out Horizon View VDI, so the vApp will contain all the infrastructure servers you need (pre-configured) and a bunch of virtual desktops which allow you to run the whole environment on VMware’s virtualized infrastructure that supports the HoL. VMware’s two data centers that support the HoL among other things are based in Wenatchee, Washington and Las Vegas, Nevada.
The HoL environment supports over 400 students at once. Check out this picture of the room – it’s massive! And that’s only the half of it I could fit in the picture. I should have done a panoramic.
In years past, the VMware HoL team leveraged a huge EMC storage infrastructure to support the environment – something along the lines of 14 racks of VNX equipment, heavy on flash to handle the IOPS load. This year, the entire infrastructure was run on four XtremIO X-Bricks (the basic building block of our scale-out clustered design) consuming just a half a rack and with a power budget of 3,000 watts. This being the first time XtremIO would be used at scale for the HoL, we were cautious in sizing the environment and we made sure to provide more than we thought they’d need. As you’ll see, we were way too cautious.
What makes XtremIO ideal for this type of large scale virtualized cloud infrastructure?
- XtremIO has tremendous IOPS performance with sub-millisecond latency. Large scale cloud data centers drive lots of small, highly random I/O to shared storage. This is exactly the workload XtremIO was built for. Each X-Brick the HoL ran is capable of sustaining 150,000 fully random 4K IOPS, measured with a 50% read / 50% write workload. And it can do this all day and all night, regardless of how full the array is.
- XtremIO delivers this performance with true inline data reduction technology. VMs have a lot of common information between them. So we can pack many more VMs onto an XtremIO system than the underlying physical amount of flash would otherwise allow. The HoL got an overall boost of over 6:1 from this technology (note that the HoL used Linked Clones and the space savings with full clone VMs would have been much higher). In other words, the four X-Bricks we supplied with a usable capacity of 30TB were able to support 180TB worth of VMs. And best – zero spaces in the VMs don’t even count toward this utilization.
- XtremIO’s metadata management system (details are beyond our scope here, so stay tuned as we’ll talk more about this in future posts) allows VMware cloning activity to be carried out entirely in memory at extraordinary rates of speed, and without affecting ongoing host I/O to active VMs. The HoL could create new vApps while the labs were active with students – something they could never do in the past!
- The maximum IOPS seen by any of the XtremIO X-Bricks was 50,000 – less than 1/3rd of its overall performance potential.
- Uptime: 100% — the workload is so strenuous that in years past the HoL has experienced outages. Nothing of the sort this year.
- Total labs taken: 9,587
- Total number of VMs delivered: 85,873
The picture below was taken on day 2 of VMworld. There was a real-time dashboard running the entire event. The second picture below is of the real-time vCenter Operations Manager dashboard collecting data from the XtremIO arrays. Given that the arrays were running at far less than their peak capabilities, the ESX datastore latency was about 0.2 milliseconds! Our specs are for sub-millisecond at peak load. When the arrays are loafing latency is just astronomically low.
Now for the best part. Not only did XtremIO help the VMware HoL team achieve success with a massive scale, complex infrastructure and do it with less cost and a vastly simpler setup, the infrastructure ran so well this year that VMware’s customers took notice. Here are some actual quotes from the HoL customer survey forms:
“I have been to VMworld 4 times and in the past I felt the HoL was hit or miss (mostly miss). This has to be one of the best HoL experiences I have had. Excellent work everyone, I would say that this year you knocked HoL out of the park.”
“You guys have the lab thing figured out. Good job! Even with the few issues while trying to get going the staff did a great job keeping us informed on the overall status of the labs availability.”
“Awesome job on the labs this year. I have been to VMworld 5 times and these labs are the best by far. Great job!”
And to wrap up, here’s a video of the live dashboards during the HoL.
VMworld is heading to Europe in October and the HoL guys literally begged us to keep the XtremIO equipment to support that event too. We are happy to oblige. There is simply no better way to run a large scale virtualized data center.