Think you can do VDI better than a DaaS provider? How arrogant are you?!?

We've been talking about DaaS quite a bit recently. My belief is that DaaS will soon make sense for all VDI use cases.

We've been talking about DaaS quite a bit recently. My belief is that DaaS will soon make sense for all VDI use cases. (In other words I'm not saying that all Windows desktops will become VDI, but I am saying that whatever subset of Windows desktops become VDI—all of those desktops should be hosted by DaaS providers.

In other words I'm saying that the days of building your own VDI are numbered. You won't be able to build VDI that's cheaper, more reliable, with better performance, and that's more secure than someone like Amazon or VMware. You just won't.

And if you do think that you can build VDI that's cheaper, more reliable, more secure, and has better performance than Amazon, I have to ask you why? Why do you think that you, with your skills, buying your off-the-shelf hardware and ones and twos scale, paying your retail pricing for all your software and hardware and licenses—why are you so delusional to think that you can build this better than Amazon? How arrogant are you?

To be clear, I personally don't believe that I can do it better than Amazon.

This is something that I've been talking about for years. Actually many people have, though generally in the context of more mainstream IT services like email. So let's use Gmail as an example of "cloud versus doing it yourself."

My setup statement is, "I can't for the life of me understand why anyone chooses to run their own email servers in this day and age, when Google [or Microsoft, etc.] offer multi-gigabyte corporate-connected email accounts for a few dollars a month. Google, Microsoft, and others like them are cheaper than what you can do, with better performance, with better security, and with more reliability.

Let's look at "Gmail versus you" in each of these categories:

Pricing & Performance

People say that Gmail is too expensive. That is false. You can't do mail cheaper than Gmail. You just can't.

People say that, "Yeah, but Google has to make a profit on you, so you're paying for the email and you're paying extra for them to have profit." This is true. But Google can deliver Gmail so cheaply that they can still make plenty of profit and sell you email service that's still much cheaper than what you can build on your own.

Look at Gmail's scale. Google's datacenters operated with an average PUE of 1.11 (versus, what, best case 1.8 in your datacenter). You only get that efficiency by scaling to millions of users. And look at the computers in a Gmail or Microsoft hosted mail datacenter. Do you see any Dell or HP logos on them? (Nope, and Nope!) These huge providers design their own hardware with zero overhead. Servers with no metal boxes, no USB and no video ports, SSD and memory chips soldered directly to the main board, fans, power supplies, and backup batteries which service the entire rack, etc. It is literally and physically impossible for you to buy hardware which is so cheap (to both acquire and operate) unless you're ordering servers by the thousands.

Again, these huge cloud providers can sell you their services (and make a profit) for less money than you can run an email system yourself. (And I'm just talking about hardware. Then factor in that they write all their own software and don't have to pay for licenses for all that, and they've got a slam dunk!)


When it comes to security, a common belief is, "Hey, the cloud is not secure!" or "Google has agreements with the NSA and they will snoop everything." Let me tell you what: The NSA will have a much easier time hacking into your home-built on-premises Microsoft Exchange environment than into Google. And where are these supposed "secure" servers of your exactly? In your office? How secure is that? You're set because you have electronic locks, right? Come on! Someone who wants your data will smash and grab their way in and be gone before the police arrive. (Your hard drives are all hot swap, right?) I would trust my data in some random secure Google cloud much more than my office building.

And let's say, just for the sake of argument, that Gmail is somehow hacked. I guarantee that if someone hacks Gmail, they're not going after you. They're going for big dollar Edward Snowden-type stuff they can sell to the Chinese, not your cache of PowerPoints about next year's pricing options.


TechTarget (my employer) is one of the companies who still runs their own Exchange servers. (To my coworkers who will undoubtedly tell me I shouldn't give away private company "secrets" in my blog, I'll tell you that anyone who has ever received an email from a TechTarget employee can view the full details of the message to learn all about our mail infrastructure. Failing that, even a non-geek can randomly type "" into a browser and know we use Exchange 2007.)

Anyway I asked why we didn't use Gmail a few years ago, and the answer I got was, "Gmail has no SLA." Do you know why Gmail has no SLA? Because it's Gmail. They don't need an SLA. When Gmail is down it's in the news. And it's been down for what, like 4 hours once in the past three years? How often are home-grown Exchange servers down? Multiple times per month? Heck, patch Tuesday alone is good for a few hours a month! (Unless you want to build clustered hot spares with failover, which if you do, see Point #1 about Pricing above.)

On top of all that, on the rare occasion when Gmail is down, Google has like 100 PhDs who invented the internet scrambling to fix the problem. Think about that. When Gmail is down, Google literally has the guys who wrote Gmail fixing it. When TechTarget goes down, we have Ray. (And we paged him...)

So all VDI should be DaaS?

Of course all of these "onsite Exchange versus Gmail" arguments are the same arguments that apply to public cloud-based services in general, including cloud-based desktops and DaaS versus on-premises VDI. And they also apply to my argument that you can't build and run your own VDI for less cost, with better performance, with more security, and with higher reliability than Amazon. You just can't.

That said, if you still decide that you'd rather build and run your own VDI, that's ok! It's ok, it's ok, it's ok!

The only catch is if you still decide you want to build and run your own VDI, you have to do it knowing it will be more expensive, less secure, and less reliable than Amazon. If that's ok with you, fine. Go for it! You obviously have reasons that are more important than these which require you to build and run your own VDI. Maybe you have specific needs that cloud providers can't provide. Maybe you're just more "comfortable" with it (whatever that means) or maybe you're eccentric or maybe you're protecting your own job. Whatever. Fine. Embrace your own environment, but don't pretend for a second that you can do it better than Amazon.

Join the conversation


Send me notifications when other members comment.

Please create a username to comment.

Spot on!

But my most recurrent argument against cloud and XaaS model is, at the moment, data sovereignty/locality issue.

As soon Google can make sure (SLA?) customers' data is local and ruled by local (read country) regulations, they will move to cloud and XaaS model for the goodness you cited.

In the meant time is a no-no! :(



While I think DAAS will eventually be a no brainer instead of locally hosted, I know I can build a fully HA environment at half the cost AWS are charging. - But them I'm in education and we get crazy deals on software and hardware.

I also know that we've had 100% up time in 3 years - Amazon/MS can't say the same.

Security - whats that? ;) - Depends how you splice the data, etc.

I don't disagree though - as always, it "depends" on the individuals environment and their requirements.



I agree in principle, but the key is this:

" I'm saying that the days of building your own VDI are numbered"

How high will that number be? After all, the same sort of arguments, if not more so, can be made for Cloud vs. on premise server virtualization. And yet, many (most?) organization still host their virtual servers inside their own datacenters.

Moreover, if that number of days is too high then by that time we may not need  Windows Desktops any more. With Windows future as middleware, I believe we will be Cloud hosting individual Windows applications rather than entire desktops. Will this also be called DaaS?



I want to have most users on a 2012 RDSH desktop and some on Windows 7 desktops. I'll even need a handful of Windows XP desktops, of course I don't really want XP but I just have to because my crappy line of business applications just won't work on anything else.

Amazon can't do this you say? I guess I'll just have to do it myself then, when I've finished I'll argue that I DID do it better than Amazon!

I've got a bunch of App-V packages that I'm going to re-purpose for my environment too. Some of them don't sequence so I'll just have to manually install them somewhere - Ah this application won't co-exist with another, I'll have to silo it off onto something else and publish it in, somewhow….

As we know the vast majority of the time deploying and managing VDI is not setting up the infrastructure it's deploying and maintaining random applications. Amazon has a core list of common applications it can deploy for you but that's the low hanging fruit. Its okay Amazon will let me put on my own applications in my own way, but that's not them doing it, its lil old me! They are just feeding and watering my infrastructure that would have only taken me a week to set up anyway. It's really not that much different to how it works in your own data center just far less flexible.


I'm sure you hear this a lot... but I was at a VDI presentation last week and I asked the presenter about DaaS and he was extremely skeptical because of the bandwidth requirements.  How much bandwidth does a knowledge worker require and then add a tiny bit of multimedia to it?  I know that remoting protocols have gotten much better (mostly by switching to UDP and doing as much client side as possible) but the impression is that a reasonable experience still requires between 100-200k per user... a little bursty with activity... constant... and existing networks aren't used to that.  So the caveat is... if you are willing to upgrade your connection significantly to accomodate DaaS then it is doable... but the bandwidth your LAN has is so much better and will always be much cheaper.

If you have a distributed workforce supplying their own bandwidth... that might be better... but will probalby lead to a more diverse range of experiences as different users have different network properties.


I was recently asked by one of my enterprise customers to do a high level analysis on exactly this topic.  This customer currently runs thousands of desktops for developers within their internal private cloud infrastructure.  Bottom line, to forklift the existing desktop VMs from the internal cloud to Amazon's DaaS would have roughly a $12M/yr recurring cost.  This number was greater than the entire annual operating budget for the broader internal cloud environment at the customer.  So, in this specific case the cost argument did not hold water and there were other operational considerations around support and soforth that ultimately killed the idea of going to Amazon.  

Now that said, if you're starting from scratch or if you're a smaller shop I would absolutely agree with Brian's comments.  Bottom line, you really need to look at the big picture and do some sort of analysis of pros/cons before just jumping into the deep end of the pool.


Where data lives still matters. The number of file shares in the average environment with.... 1000 desktops? the amount of backend data for applications they want to run in their desktop...

You have all local apps, and local data in the desktop or to some USB drive? sure. Walk into the average company and move a desktop to amazon or whatever and test all those apps (not word or local copy of outlook) all the other apps.

There is  still a data issue that needs solving. MS doesnt sell SQL license for no reason. Or Oracle, etc. etc. etc


Lots of truth in what everyone is saying here :-)  To build a bit upon what Ron is saying, location and connectivity to both data _and_ the backend for various apps matters a great deal.  

One of the "fringe benefits" of VDI has been to move desktop workloads closer to the enterprise server apps reside.  Now, for greenfield scenarios like SMB or startups, I tend to agree with Brian that public cloud hosted DaaS makes complete sense: all my application workloads like email, my HR tools, customer database, etc are going to be in the cloud anyway, so if I've got a few legacy Windows apps I need to run, and/or I desire a Windows desktop experience from a low OPEX device like a zero-client, super!  Let me put those things in Windows runtimes in the cloud and be done with it.  However, if I've got a lot of workloads and data for whatever reason sitting in my DC that aren't moving to IaaS anytime soon, I may get much better cost/performance keeping my desktops closer at hand within the cozy walls of my corporate LAN.

Now, this also lends itself to an interesting corollary: the natural linkage between IaaS and DaaS.  There will be a natural gravitational pull between them, and if I'm to make a new year's prediction a couple days early, it will be that by the end of 2014, IT folks in the know will become savvy to this, and start selecting clouds (public, private and our favorite term, hybrid) for their workloads on the basis of their ability to deliver BOTH infrastructure and desktop services.  Neither decision should ultimately be made in a vacuum.

Should be pretty exciting!


I think the most important component has been missed here: apps. I also agree with Ron in terms of the other important piece of the puzzle that Brian's forgotten to mention: data. If I already run an enterprise data centre with all my apps and data in it and then want to deploy VDI, how much will my network cost to get sufficient bandwidth (assuming I can also get low enough latency) to hook up to Amazon? Not everyone has an Amazon data center offering DaaS sitting over the road from where their enterprise apps and data are already hosted. Don't get me wrong, DaaS is great and is only getting better (and cheaper), but if you're calculating delivering an IT service to a user based on the cost of delivering the desktop alone, you're living in a false economy.


I've been meaning to reply to this for sometime - I agree with Ron and Robs comments above - one design principle that has always been the first to be considered is - "position the VDI/TS/RDSH/Desktop workloads as close as possible to the Data/Server Room/Data Centre"

So from this, it tends to indicate that *unless* you are going to position your AD/Servers/Data in AWS then AWS's VDI is not for you? If you have 60 - 90% Server Virtualization in your own Data Centre would you seriously consider moving the VDI loads to AWS?

The other element that this article completely ignores is Latency, now maybe in the continental US this is not so much an issue with AWS PoP's being all over the place, but outside the US this will have a big impact.

While nearly all Desktops can be virtualized, it doesn't necessarily mean they should be?