2 posts • joined 26 Oct 2012
OK – so the VDI market space is definitely moving at a break neck pace. We are in the midst of a perfect storm of data center compute solutions, storage performance with flash technology and user access and mobility with the BYOD revolution. Seems like we are already at VDI 2.0 and most of the implementations I run into are already behind the curve at version 1.0 or nothing at all – or are they?
Pure all flash storage solutions definitely seem to have a good corner case fit for this workload – or at least part of it. The desktop itself – the core OS image and application(s) layer seem to fit well into the all flash and deduplication methods of the all flash solutions. However, I think this comes at a cost – the true cost of capacity for the users behind the desktops.
If you have already migrated all of the user data out of the desktop image leaving only the profile/persona behind, then thin (Linked Clone or PVS) or fat (Full) desktops can be traded off by how you want to manage OS and application updates and pools of desktops. Stateless or Persistent desktops don’t impose that much impact on the infrastructure – but they do affect the user experience.
Question: Do I need to have distinctly different storage platforms for the different parts of my virtualization infrastructure or is there a better, easier to manage model. Traditional spinning disk only solutions have obvious drawbacks of size, cost, and complexity even if the performance measure is easy – just add more spindles and capacity increases too. All flash solutions make performance measures easy – they are fast – all fast – just get enough space to support the workload. How much space that is depends on a lot of variables – namely deduplication and thin provisioning methods and frequency and uniformity of data updates.
I think hybrid solutions provide the best of both technologies. Performance and capacity sizing is still required and although this is a new type of sizing exercise for many storage admins, it is not rocket science. With this hybrid storage (flash + disk) approach you can get the performance (maybe not at screaming levels) needed to support the active workload and the capacity (definitely not at screaming prices) to store the rest of the less-used infrastructure data.
My bet is that all flash solutions will allow IT teams to deploy some interesting VDI solutions, but be prepared for the added cost – the per desktop OS/application layer will be higher than hybrid solutions and there will likely be another storage requirement somewhere else in the picture to hold the user data that lies somewhere between the desktop background image and the corporate file share repository.
Clarifying Confusion on Nimble VDI Solution
It seems our (Nimble Storage) announcement about 1,000 desktops in 3u has caused some confusion. As one of the Nimble storage team involved in this project, let me try to help clear this up a bit.
The 3u footprint is for the storage component only. Nimble Storage appliances are all 3u in size based on a combination of traditional spinning HDD and flash SSD. You can find more information here:
The observation that the $43 per desktop is for the Nimble Storage component only is correct.
For this solution, we combined the Nimble Storage CS220G-X2 with a single Cisco UCS blade chassis (another 8u) with 8 nicely configured B-series blades supporting the virtualization infrastructure and the desktop environments. Add to this a pair of Fabric Interconnect switches (another 3u if you add in some cable management space). The total rack footprint was 14u.
The testing and success criteria were based on the VMware View Planner tool (similar to RAWC in many ways). This test bench performs a number of actual desktop operations (opening, closing, updates, etc.) on all desktops over a test period. This includes MS Office (Word, Excel, Power Point), Outlook, IE, Firefox, Adobe Reader, and 7-Zip applications. There is some Windows Media Player testing too but the majority of the testing focuses on the activity that might get directed to the local disk as part of the infrastructure workload. Internet browsing (e.g., YouTube) and media streaming (video and audio) are not included in this type of VDI workload testing. The responsiveness and latency numbers observed were well within acceptable limits specified by VMware and Cisco.
More information on VMware View Planner can be found in the View Planner Community here:
Detailed test results will be published shortly – including performance measurements from both the storage side and the server/hypervisor side (esxtop).
I hope this helps clear up some of the confusion about the solution …
Mike @ Nimble
- Vid Hubble 'scope snaps 200,000-ton chunky crumble conundrum
- Bugger the jetpack, where's my 21st-century Psion?
- Windows 8.1 Update 1 spewed online a MONTH early – by Microsoft
- Google offers up its own Googlers in cloud channel chumship trawl
- Something for the Weekend, Sir? Why can’t I walk past Maplin without buying stuff I don’t need?