1. Open API

This is needed to automate the start-up, configuration and termination of cloud instances.  Without automation the cloud infrastructure is no use to you, a manual web page driven administration process is not going to win a provider any points with me.  As a secondary point, it’s even better if they provide tools that integrate with these API’s.  Making me write tools, is again, not going to win any points.

2. Secure IP connectivity

By this I mean the ability to secure the connection between a cloud IP subnet and private infrastructure.  If I need to create instances on demand I need to be able to securely access the subnet they are on and hide those machine from the ‘net,  only being able to access machine securely on an individual basis will not do.

3. Decent guest start-up time

By this I mean under ten minutes guaranteed.  If you only promise between 15 and 45 minutes (RackSpace) then it’s too slow.  Also as billing usually starts from the request not the availability I don’t want to be paying for time I’m not using.  The solution for this would be to move to a billing from availability model, this would motivate providers to get guests up quickly.

4. Support for new Guest versions is quickly adopted

If a new hypervisor, or a new OS version comes out I want to be able to take advantage of those features quickly, I especially don’t want my local infrastructure to be held up by interoperability problems with cloud services if they are behind the upgrade curve.  When you are waiting on a large corporation to upgrade and your business is too small to put pressure on them to make you a special case then you are going to get pretty angry pretty quickly.  There is at least on cloud provider (I’m looking at you Amazon EC2) which doesn’t support Windows 7 or Windows Server 2008 R2 and it’s nine months after RTM.

5. Hypervisor Access

I need to be able to upload my own virtual machine appliances, whether they are from a third party or one I’ve made onsite.  I also need to be able to manage the hypervisor layer with the same tools and using the same skills that I already have in house.

6. Keyboard Video Mouse console access

There is a reason that servers have KVM boards, it’s that not all problems happen after you have RDP or SSH access.  You lose a whole lot of troubleshooting information if you lose visibility of the console.

7. Hourly billing

All instances should be able to be billed hourly, I don’t mind if you have monthly charges as well, but hourly should always be available, if I need a resource permanently, I might as well host it myself.  Give us the option to try out, demo and burst into the full range of your offerings.

So have I found a provider that fits the bill?  The short answer is no.  The slightly longer answer is that I’ve found one who are really close, close enough that I’m happy to use it.  That provider is SoftLayer.

I reserve the right to change my mind at any time as providers change their offerings.

BriForum this year will be running from June 15th to 17th in Chicago, this is a conference that I have wanted to go to for a long time, but never before had the chance.  This year not only will I be going for the first time, but will be speaking alongside my colleague Rick Dehlinger.  The topic is spun out from my blog post Do we have the right tools to cloudburst xenapp into ec2?  and will be a deep dive into what it takes to do this.  If you are planning on attending BriForum, come along and see what we have to show you, or just come up and say hello.

Here is the topic detail:

CloudBursting XenApp – hype or reality?

It seems like every vendor in the world is hyping ‘cloud’ somewhere in their marketing pitch. The noise is so prolific that’s it hard for any astute technologist to ignore. As desktop and application delivery specialists, many of us have been building and running ‘clouds’ for quite some time, albeit ‘private clouds’, private delivery systems, centralized hosting environments, or whatever the term du jour may be.

As we’ve come to expect, the delivery technologies we use and the plethora of available services delivered out of the cloud have evolved at a dramatic pace. As we dive down into the microcosm of our specific niche of the industry, we’re seeing a couple vendors pitching a hybrid approach to cloud service consumption – Citrix and Amazon. The noise they’re making together means that we’ll all likely have to field questions on the topic sometime soon, which begs the question: Is it real, or is it hype?

This session explores this hybrid approach to cloud usage (which has been called ‘cloud bursting’) and seeks to answer some of the key questions on all of our minds. It will seek to answer the following questions:

• What is ‘cloud bursting’?
• Why would anyone want to do it?
• Is it reality or hype?
• What are some of the things we have to consider before adopting such an approach?
• Which vendors provide the right cloud infrastructure?
• What are the infrastructure components we need to achieve the right result?
• How do Citrix and Amazon do it?
• What support does Citrix provide to help?
• What support does Amazon provide to help?
• How do I build it?
• Can I do it with ‘off the shelf components’?
• Can I extend my existing infrastructure?

So what is cloudbursting and why would you want to do it?  Cloudbursting is the ability to expand your existing datacenter infrastructure into the cloud.  This could be useful at times of high demand, for instance seasonal peaks around Christmas, or if your existing infrastructure loses capacity in a disaster recovery situation.

One of the major stumbling blocks on the way to widespread acceptance of utilising cloud infrastructure is the fear in the eyes of many executives of losing control of their data.  What happens when your critical data is stored on someone else’s infrastructure?  Is it secure?  Is it reliable?  Is the support good enough?  What are the response times?  Can you believe the providers when they say you don’t have to worry about your fears?

This is why a XenApp workload is particularly suitable to cloudbursting, there shouldn’t be any data stored on XenApp servers.  Also they should all be identical, making them conducive to fast provisioning.

Imagine the situation of a sales call center, over the xmas period they hire temp staff to cope with extra demand, as their software is provided via XenApp, the company needs to have enough infrastructure to cope with the demand peak, but sits idle most of the year.  As EC2 charges per hour then the ability to use EC2 to cope with the extra demand could save large sums of money.

Now we have decided that cloudbursting is a good idea, can we actually do it?  By this I mean are the tools available off the shelf, without a large development effort.

So what do we need?

• A secure link between the cloud and your datacenter
• The ability to quickly provision and decommission servers in the cloud
• WAN acceleration between the cloud and you
• Monitoring to know when to cloudburst
• Automation to control it all

Lets take these needs one by one:

A secure link between the cloud and you, currently Citrix provide an Amazon Machine Image (AMI) template for Citrix Access Gateway (CAG) with one in the cloud and one in the premise you can have a secure channel between the two. You could also use the Vyatta AMI.  Full marks

With the new workflow studio library we can quickly provision our own saved AMI’s and destroy them when needed.  The question here is why are we not using Provisioning server?  It would be best to provision a ‘bare metal’ server and PXE boot to receive a provisioning server vdisk. So Half marks.

WAN acceleration is possible, you can install the software repeater client on the XenApp servers, but a proper Repeater AMI would be better. Half marks again

Monitoring could be done either with Edgesight or the power and capacity management feature, so full marks.

Automation is the big problem, although workflow studio, now it’s 2.0 with more libraries, is getting there, at the moment it simply doesn’t have enough pre-configured workflows or libraries to cope.  We need a way to join the Servers to the domain and farm and publish the applications. (Although XenApp 6 will let us do this using GPO’s). We could script this, but I want to do it without any dev work. It also needs to be able to take in the output from the power and capacity management feature set.

So where does that leave us?  I’d say it leaves us almost there, in fact with a little powershell knowledge and using the tech preview of XenApp 6 it’s possible right now.

If I have time over the holidays I think I’ll try and set it up and let you know how I get on.