Recently my colleague Jasper used Login PI to test the performance of a well-known cloud provider to see what their out of the box performance would look from the user experience point of view. Quickly we realized that, regardless of the hoster, If you want to move your desktops to the cloud it would entail a bit more work than simply selecting the cheapest desktop you can find + Microsoft Office and let it go from there.
Today I want to show you what I’ve learned from looking at Microsoft Azure’s performance for a week. I took a different approach for these tests as I was interested in seeing how changing the amount of money spend on a desktop would influence its performance (and if it would). Unlike other hosters, in Azure there is no limit on what operating system is available out of the box and there is no 200 desktop limit for more flexibility, although you can’t fault 3rd parties for Microsoft’s licensing.
Configuring the virtual desktops
The configuration of my desktops is as follows:
- They’re both 2vcpu virtual machine but the big difference is that for one the storage is on SSD and the other is HDD.
- Next to that I have chosen to have it running all the time costing me about €75 per month for the VM with HDD and about €80 for the one with SSD.
Comparing Virtual Machines
As you can see the price difference between the two is relatively low, but does it get me additional value? To prove that I ran the Login PI – Native and Microsoft Office applications workload to compare logon times and application start times. Image 1 shows the application start times from the VM with a HDD. As you can see the deviation is quite big; the fastest application is Wordpad whereas the slowest is Paint. Of course, these are not your typical enterprise applications but rest assured if you want to measure your business critical applications that’s also possible.
Now switching to SSD we notice something that we all know from the physical world. Applications start faster: A lot faster. Let’s take Microsoft Word as the baseline. As you can see it starts about 3 times faster on this VM while its only marginally more expensive.
While application start times are nice, there is something else the enterprise user is confronted with every day; the dreadful logon process. In this case my environment is rather clean, there are no policies, hardly any logon scripts and no anti-virus or other agents that typically negatively impact performance. But still we can see clearly that the VM with SSD outperforms the HDD by about 3 times.
The conclusion of this blog is that spending a bit more on your VM’s in Azure does make a difference, whether the difference will be the same in your environment, of course, depends on your own applications. But it’s fair to say it probably will.
Similar to Jaspers discovery’s: Do I think you can just deploy a bunch of desktops in the cloud and be done with it? Absolutely not. Either you spend time and effort yourself on making it perform or you pay a 3rd party to do that for you. Next to that it is interesting to see where this market will move. Citrix has announced they will take over Microsoft RemoteApp business and VMware will start to collaborate more with Amazon.
Are we done now? Certainly not! In our next blogs we will outline the steps we have taken to make our cheap desktops perform as our users have come to expect and if there is interest we might even dive into other comparisons: What’s the impact of latency, bandwidth etc. Keep your eyes on the blog section of our website, so you stay updated.
Thanks to Andrey Korenkov (@akoncloud) I just discovered that we are getting a discount on our Azure subscription as a benefit of the MSDN membership. The impact of this is big enough to warrant an update to this post as you will find below. Some estimated prices are pretty funky as you can see within our plan we can get the same VM with SSD/HDD for the same price.