Service Level Agreements and Cloud Computing

There is lots of discussion today about “cloud computing”. You can’t see a mention of cloud computing without some reference to Service Levels and Service Level Agreement. The “SLA” term is thrown around and yet has many definitions. Some would argue it is simply a measure of up or down time, while others would argue it is not just availability but performance of the application from an end user’s perspective.

Defining and measuring SLA’s in the cloud (both public and private) are vital to ensuring that your investment in time, effort, and money is yielding an appropriate return and that the end users are receiving the services that they were promised. Whether you’re serving up an ad-hoc test/development environment in a private cloud or leveraging a CRM application on the public cloud you have to be able to measure and quantify application performance and availability.

Knowing the performance of cloud hosted applications and having an insight as to why there are degradations in performance can be particularly challenging. Cloud computing relies heavily on virtualization where, unlike physical environments, resources are shared. So simply looking at resource utilization or up/down status alone isn’t going to yield much value. There are simply too many moving parts and dependencies and again there is that sharing of resource issue.

What’s needed is to aggregate key performance and operational indicators together with user experience measurements (aka response time) in an integrated view of application health. Is the application running? Is it responsive when measured from the user’s perspective? If response is poor, where is the issue and why?

For private clouds this includes not only the health/performance of the virtualized environments, but also the the health/performance of the applications (database, web, etc.), the health of the network, and how all of this impacts the response time of the application. If there is degradation where is it? Why and how long did it last?

Public cloud computing presents it own unique challenge; a cloud provider can certainly measure their resources as well as the performance and availability of their internal infrastructure, but they can’t measure the end user’s experience. Anyone implementing an application in the public cloud should have an independent way of looking at application performance in general and end user experience in particular, ideally with both real time and historical charting of end user experience. It allows consumers of public cloud applications to hold their providers accountable for the services that they are promised.

For a quick look at how Longitude aggregates performance, operational and user response metrics, check out this screen shot of the Longitude SLA Dashboard.

I invite people to comment on my blog -

• What kind of challenges are you facing with public cloud providers, especially when it comes to measuring availability and performance? Are you able to measure what was promised to you?
• For those tasked with delivering private cloud applications, what kind of reporting demands are being made of you to show that the investment is yielding the appropriate benefits to the organization?