The WebSphere CloudBurst Appliance can bring speed and agility to test organizations by drastically increasing the pace and ease with which users interact with WebSphere Application Server environments. I recently got a chance to catch up with IBM's Robbie Minshall. Robbie is a WebSphere Test Architect, and he is responsible for a team of testers that harness a lab of over 2,000 physical machines to put our WebSphere Application Server product through some pretty rigorous testing.

Toward the beginning of 2009 Robbie’s team started to leverage the WebSphere CloudBurst Appliance in order to create the WebSphere Application Server environments needed for their testing.

Robbie, can you tell us a little bit about what the WebSphere Application Server test efforts entail?

In WebSphere Application Server development and test we have two primary scenarios. The first is making sure that developers have rapid access to code, test cases and server topologies so that they can write code, test cases and then execute test scenarios on meaningful topologies. The second scenario is an automated daily regression where in response to a build, we provision a massive amount of WebSphere Application Server topologies and execute our automated regression tests.

Previously we have supported these scenarios through the deployment of the Tivoli Provisioning Manager for operating system provisioning, some applications for checking out environments, and then a lot of automation scripts for the silent install and configuration of WebSphere Application Server cells.

Given those scenarios and the existing solution, what are your motivations for setting up a private cloud using WebSphere CloudBurst Appliance?

We are supporting these scenarios through a pretty complicated combination of technologies.  These include silent WAS install scripts, wsadmin configuration scripts, a custom hardware leasing application and the utilization of Tivoli Provisioning Manager for OS Provisioning. This solution is working very well for us though as always we are looking for areas to improve, opportunities to simplify and to reduce our dependency on investment in our custom automation scripts. Mainly, there were 3 areas where we wanted to improve our framework: Availability, Utilization and Management. This is why we started looking to the WebSphere CloudBurst Appliance.

Can you expand a bit on what you are looking for in those three areas?

he first focus area we have is availability of environments. We really wanted to lower the entry requirement for the skills and education necessary to get a development or test environment. Setting up these environments has just been too hard, too time consuming, and too error prone. Using WebSphere CloudBurst we can provide an easy push button solution for developers to get on-demand access to the topologies they need.

The second area we are looking for significant improvements on is hardware utilization. Our budgets are tight and in our native automation pools we are only using between 6-12% of the available physical resources. In order to improve this we were looking at leveraging virtualization. WebSphere CloudBurst offers the classic benefit of virtualization with the nice additions of optimized WebSphere Application Server placement and really good topology and pattern management. In our initial experiments we were able to push the hardware utilization up to 90% of physical capacity and consistently were leveraging around 70% of our physical capacity.

Finally we are looking to improve and simplify our management of physical resources and automation. We work in a lot of small agile teams and organizational priorities change from iteration to iteration. Not only does WebSphere CloudBurst allow us to maintain a catalog of topologies or patterns for releases but it also allows us to adjust physical resource allocation to teams through the use of sub clouds or cloud groups.

Basically we felt that WebSphere CloudBurst would improve the availability of application environments, enhance automation, and improve hardware utilization all with very low physical and administrative costs.

What were some of the challenges involved with getting a cloud up and running in your test department?

One of our challenges seems like it would be common to many scenarios, especially in today’s world. Our budget for new hardware to build out our cloud infrastructure was initially very limited. Most cloud infrastructure designs depict very ideal hardware scenarios including SANs, large multicore machines, and private and public networks within a dedicated lab. Quite frankly we did not have the budget to create this from scratch. It was important for us to demonstrate value and data to warrant future investment in dedicated infrastructure. After some performance comparisons we were very happily surprised to see that we could leverage our existing mixed hardware within a distributed cloud. The performance of application environments dispensed by WebSphere CloudBurst on many small existing boxes in comparison to large multicore machines with a SAN was very comparable. This allows us to leverage existing hardware, with minimal investment all the while demonstrating the value and efficiencies of cloud computing. That data in turn has allowed us to obtain new dedicated hardware to iteratively build up a larger lab specifically for use with WebSphere CloudBurst.

Specifically with WebSphere CloudBurst, are there any tips/hints you would offer users getting started with the appliance?

Sure. First, we quickly realized as we added hypervisors to our WebSphere CloudBurst setup it was critical to have someone with network knowledge on hand. This is because the hypervisors came from various sections of our lab, and we really needed people with knowledge of how the network operated in those different sections. Once we had the right people we were able to setup WebSphere CloudBurst and deploy patterns within an hour and a half.

Moving forward we continued to have challenges as we dynamically moved systems between our native hardware pool and our cloud. Occasionally the WebSphere CloudBurst administrator would move a system into the cloud but incorrectly configure the network or storage information. This lead to some misconfigured hypervisors polluting our cloud. We overcame this, quite simply and satisfactorily I may add, by creating some simple WebSphere CloudBurst CLI scripts which add the hypervisors, test them individually, by carrying out a small deployment to that hypervisor, and then move the correctly configured hypervisors into the cloud after verifying success. Misconfigured hypervisors go into a pool for problem determination. This has allowed us to maintain a clean cloud, and we are able to dynamically move our hardware in and out of the cloud to meet our business objectives.

We also use the WebSphere CloudBurst CLI to prime the cloud so to speak. Before using a given hypervisor in our cloud, we execute scripts that ensure each unique virtual image in our catalog has been deployed to each of our hypervisors at least once. When the image is first deployed to a hypervisor, a cache is created on the hypervisor side of the connection, thus meaning subsequent deployments do not require the entire image to be transferred over the wire. This gives us consistent and fast deployment times once we are using a hypervisor in our cloud.

I would assume that like many applications deployed on WebSphere Application Server, your team’s applications have several external dependencies. Some of these dependencies won’t necessarily be in the cloud, so how did you handle this?

You’re right about the external dependencies. Our applications and test cases run on the WebSphere Application Server but are dependent upon many external resources such as databases, LDAP servers, external web services etc. WebSphere CloudBurst allows us to deploy WAS topologies in a very dynamic and configurable way but the 1.0.1 version does not allow us to deploy these external resources in the same manner. This was overcome by using script packages in our patterns. These script packages allow us to associate our test applications with various patterns we have defined. The script package definition also allows us to pass in parameters to the execution of our scripts. We supply these parameter values during deploy time, and these values are used to convey the name or location of various external resources. The scripts that install our applications can access these values and ensure the application is properly integrated with the set of resources not managed by the appliance.

What is your team looking to do next with WebSphere CloudBurst and their private cloud?

The next challenge on our plate is to keep up with the demand of our expanding cloud and to develop a more dynamic relationship between our native pools and our cloud using the Tivoli Provisioning Manager. These are fun challenges to have and we look forward to sharing our progress.

Thanks for your time and insights Robbie.