Amazon announced today that it was exiting beta and offering a 99.95% SLA within a region. Hopefully this is going to put some cloud naysayers to rest, at least on the reliability front. Amazon is offering accountability in the form of service credits if it violates the SLA. Now this may not be enough for some of you with mission critical applications, but my guess is that most people out there are going to be just fine with it, considering the cost savings that on demand cloud infrastructure provides. Oh yeah, they’re going to be offering Windows servers too. This should make some dot-net-heads pretty happy.

There are already hundreds (thousands?) of companies taking advantage of Amazon EC2 computing resources, and those that aren’t are going to catch up real quick, especially as they realize how much money they are wasting on static server resources that are mostly sitting around idling. You just can’t afford that, not in this economy. Companies are going to wise up and start cutting costs on non critical infrastructure and pushing it into the cloud. And as they gain trust for the cloud, pieces of critical infrastructure are going to follow.

Rackspace building weapons of mass destruction

Rackspace just acquired Slicehost and JungleDisk in what appears to be an effort to shore up its arms race against Amazon. They are still pretty far behind true cloud infrastructure (by this I mean on-demand api-driven resource allocation) but maybe Slicehost can make this happen for them. I’ve been a loyal Slicehost customer for close to two years now, and they’ve declined to accept uploaded virtual images thus far, but maybe that will change. See below for why Slicehost isn’t really a cloud, yet.

Your VPS ain’t a Cloud

Many ‘cloud’ vendors are still just rebranded VPSs. We’ve had virtualized infrastructure in hosting companies for years. What makes a true cloud like Amazon EC2 is that it only takes a credit card and a minute to get computing resources. The other key is that manual tweaking and hand provisioning are going the way of the dinosaur. You need to be able to get a new server up and running with your latest environment and software in minutes, not hours, days, or weeks.

Hosting solutions that require you to first acquire resources by booting up an image and then installing your software are going to be left in the dust. Amazon lets you upload a virtual image you create, which means you can mange your own image catalog, and if you’re using something like Elastic Server then you can dynamically provision your servers from recipe templates that ensure the quick reproducibility of your stack to any virtual format, whether it’s in your datacenter, or up in the cloud.

In 2009 we’re going to start to see companies moving to virtual and cloud infrastructure and dynamic provisioning to cut costs and gain agility. It’s going to be an interesting year.

Clouds are on the horizon

From the cloud-themed t-shirts and signage of the event, to VMware’s vCloud initiative announcement, to sessions and BOFs on clouds. What’s interesting is that there is still a ton of debate and confusion around what a cloud actually is. See my previous post on the definition of cloud for my own thoughts. Regardless of the lack of definition, many people were nodding in agreement about cloud technology. The promise of provisioning on demand without waiting for weeks from your IT department had many people excited. There was much talk about virtualization and clouds introducing agility in provisioning that would rise to meet the agility software teams require to do their jobs.

Private enterprise clouds are coming

Out on the vendor floor, there were several companies working on enabling private enterprise clouds (by this, I mean virtualized data centers that behave in increasingly cloud-like ways, by offering dynamic provisioning, apis or scriptability, etc). Of these, the one that caught my eye was http://www.qlayer.com/, going so far as to offer their own python-like language for scripting datacenter automation. From what I understood they are also going to be supporting platforms other than VMware, which is important. This will be a company to keep your eye on.

Orchestration and Cloudbursting

We hung out with Enigmatec who work on automation and orchestration. They demoed cloudbursting capabilities (this is a term describing the ability to add cloud resources to a private datacenter during a traffic spike). Cloudbursting was handwave-demoed at VMware CEO Paul Maritz’s keynote, but it was a VMware center to a vCloud, proprietary and locked in. It was also unclear when such capabilities would be widely available. The Enigmatec guys are doing this today and are doing it in a cross-vendor capacity. So you can have your VMware datacenter adding resources from the Amazon EC2 cloud. This is very powerful. Our Elastic Server technology at CohesiveFT powered the servers in the demo. Our concept of build a recipe once and then output to multiple formats is the key to seamless cloudbursting. I really loved this quote by Duncan Johnston-Watt of Enigmatec: “Cloud cover is a great insurance policy.”

Virtual lab automation

I also met the Skytap team, a company doing some really interesting stuff with virtual lab automation, which was one of the topics incidentally covered in a session at the 451 ICE event (see http://twitter.com/elasticserver for coverage). Virtual test labs are an excellent way to increase testing agility while saving a ton of money. Since you may need a large amount of resources for load tests, but only for a brief amount of time, provisioning and tearing them down on demand during test cycles really works.

Virtual image cataloging and visualization

The vendor floor was also heavily populated by companies involved in virtual image cataloging and tracking, with some being VMware specific, while others worked cross platform. Nothing really caught my eye. When it comes to virtual datacenter visualization the key is a really great GUI and most of these companies understandably were made up of hardcore back end engineers who had no idea about UI design. It’s a shame.

One notable exception was bluebear, a small but talented team that built an Adobe AIR client called Kodiak that has a way to go, but looks promising from a GUI perspective. On the other hand, it seems they will need to find a way to integrate with existing management infrastructures to make any headway in customer adoption.

Microsoft guerilla marketing

Not to be outdone, Microsoft had a street team (in costume, no less) outside the Venetian handing out $1 chips and business-card sized anti-vmware propaganda. This really surprised me, because I would expect Microsoft to maybe try some large scale advertising, but this shows them getting more guerilla in their marketing, which a certain part of me really responds to. On the other hand, the wording on the card felt like a smear campaign rather than something touting the benefits of hyper-v. Check out a picture of the propaganda card. The title says “Looking or your best bet? You won’t find it with vmware”. The url given is http://vmwarecostswaytoomuch.com. Yikes.

What about application provisioning?

I did not see many companies working on the application provisioning side (something that we at CohesiveFT are enabling). Most of them assumed the virtual images would come from ‘somewhere’ and they would just manage them. I’m not sure if it’s because we’re in a transition phase from physical to virtual, and that most of current Enterprise usage is from virtualizing existing physical servers, but I see less and less by-hand provisioning in the future because it’s simply a huge pain and time cost for companies right now.

Lots of VI management companies seem to assume you have some static catalog of images. But if computing power continues to grow, and virtualization technology enables easy provisioning, we can expect the number of images to exponentially increase, and to be much more dynamic. During his keynote, Paul Maritz talked about the future being that of custom OS and application-stacks that are customer focused (meaning per-usecase). We’re already doing that today with our dynamic provisioning engine — letting a customer put together a stack just for their current purpose.

I see a lot more images being built, and having shorter lifecycles. Maybe there will be many throwaway images built just for testing and virtual labs. Maybe we will see much more enterprise experimentation because it will be so easy to build a stack and not have your operations guys supporting what’s inside. In fact during the 451 ICE conference, Jim Houghton (CTO, Adaptivity) mentioned that virtualization and rapid provisioning enabled bleeding edge technology experimentation at Wacovia. So I see application provisioning becoming much more consistent, reproducible, and trackable in the future and we’re working on the technology to enable you to go from your software to a provisioned server in the datacenter or cloud in just a couple minutes. I’m curious to see if we get more competitors in this space at the next VMworld event.

Till next year!