Amazon Web Services has changed the way applications are developed and deployed, but because of the way Amazon designed its infrastructure, it also created a disconnect in the natural lifecycle of applications and operational infrastructure. A Texas-based company called SoftLayer has taken a more unified approach to operations and supporting the application lifecycle, which has some important advantages for IT practitioners.
CIOs and CTOs who are looking for a coherent cloud strategy that serves the application lifecycle will likely find SoftLayer’s model attractive. In addition, SoftLayer provides a much better path forward to improve and standardize operational processes for both cloud and on-premise applications. I predict SoftLayer’s model will be influential and will eventually be imitated by Amazon and Rackspace.
The Gaps in the Current Lifecycle
The first problem with cloud applications is that most of them shouldn’t stay cloud applications. Here’s a typical progression:
- For a variety of reasons, applications are first built on Amazon Web Services. This could because internal IT couldn’t provide the needed help, because development couldn’t wait, because the architecture of the application demanded it, or many other reasons.
- The application is then developed and deployed in the cloud. For most IT shops, this means creating a new type of operational infrastructure. Tools like RightScale or Puppet are mastered. The ability to diagnose problems and deal with the operational challenges of AWS is conquered. (Netflix’s Chaos Monkey falls into this category.) All the other IT requirements such as security, disaster recovery, and compliance are managed. Eventually the apps can be run in a stable manner.
- Usually, when an application runs in the cloud, it takes a while to understand how many resources are needed to make it scale. It is great to be able to add more servers dynamically during this period, but eventually the amount of resources needed to support an application, including the spikes in traffic, becomes well understood.
Here’s where the gap in the cloud lifecycle appears. At this point, once an application is stable and understood, it may make sense to do two things:
- Move the application to an on-premise data center to reduce costs.
- Provision the application with servers, storage, and networking equipment that are best suited to the performance and operational needs of the application.
“We are building an infrastructure that supports rational and optimal choices,” said Duke Skarda, CTO of SoftLayer. “Your cloud provider’s services shouldn’t necessarily dictate how you architect your application.”
This is the first disconnect that the cloud as implemented by Amazon Web Services causes. Making these moves should be as easy as possible. You shouldn’t have to change the application or its operational procedures. But if you started on Amazon Web Services, you will likely have to change both.
The accuracy of this analysis is supported in a variety of ways. First, Zynga’s application lifecycle works exactly as I suggested. Amazon Web Services is used to support an application once it is launched. Then, once the application is understood, Zynga moves it to its own cloud to run the application cheaply and efficiently.
Second, private cloud software providers like Piston Cloud, which sells an enterprise-class version of Open Stack, and hosting providers like Logicworks find that they have a growing number of clients who are seeking to use their products and services to migrate applications off of Amazon Web Services for the reasons just explained.
SoftLayer’s Unified Approach
SoftLayer has created an approach that takes one set of APIs and one set of operational procedures and allows you to create an application in the cloud and then seamlessly re-provision it with purpose-built hardware without changing anything.
SoftLayer’s service are offered through 13 international data centers and used by customers such as Repsol, Citrix, Struq, and Peak Games. The APIs are abstractions of computing services just like Amazon’s, but because they were built to handle both cloud and dedicated resources, you can choose exactly how you want to provision your application.
In other words, you can start with the cloud, figure out how your application performs, and then provision it with just the hardware, networking, and storage needed.
SoftLayer goes beyond Amazon Web Services in an important way with respect to operations. SoftLayer has created a set of APIs and dashboards that essentially productize operations procedures in a way that accommodates both cloud and dedicated infrastructure in one unified approach. This productization is based on both the cloud and dedicated resources being hosted at SoftLayer.
Amazon partners can provide this sort of productization using a variety of techniques. Rackspace also offers fully productized operations for high performance applications through its Critical Applications Services program, which is based on software from CA Technologies and best practices developed by the Rackspace Fanatical Support team. However, SoftLayer differs from these two approaches in that cloud and dedicated provisioning, operational processes, and supporting software are delivered as one package.
“DevOps has taught us that an application and the way it is operated must be thought of as one unified unit,” said Skarda. “At SoftLayer, that is the scope of the design problem we are addressing. To provide APIs for computing resources without integrated operations best practices and software means you are providing a toolkit, not a complete product.”
“Considerations such as cost and consistent performance also need to be addressed when tackling large workloads,” said Skarda. “A simple virtualized, multi-tenant cloud is inexpensive to start with because performance usually isn't that critical early in the lifecycle. As an application scales, however, it can quickly become costly to acquire the quantity of resources and high performance options necessary to achieve the desired throughput and response times. Additionally, an application at scale typically experiences shortcomings with a virtual machine's reliability or predictability in terms of both uptime and raw performance. Bare metal cloud servers can often address both of these issues.”
In my opinion, the biggest money for cloud computing is just starting to be spent. Companies are realizing that they should not be in the business of operating data centers. Equinix is taking over the large data centers at the high end, Rackspace for the some large customers and large swaths of the mid-market, and Axcient is scooping up lots of hosting for SMBs. Amazon is getting this business through partners like Logicworks who complete the offering by providing an additional on-premise capability.
SoftLayer will be attractive for CIOs who realize they don’t want to invent hybrid cloud/on-premise operational procedures from scratch. Rackspace has realized this is something it has to offer. Amazon doesn’t seem interested in productizing operations for the cloud and on-premise in a standard way. In the next two years, I suspect the market will force them to create such an offering.
Dan Woods is CTO and editor of CITO Research, a publication that seeks to advance the craft of technology leadership. For more stories like this one visit www.CITOResearch.com. Dan has performed writing and research services for Rackspace and Logicworks.
