dynamic data center
Opscamp Austin: IT Ops and the Cloud
Going to a conference focused on IT on a Saturday sounds like a really geeky way to spend a weekend, but thanks to an “alternative” venue, the “un conference” format, and a group of really good people, it was a great day. I just got back from Opscamp Austin, a meeting of IT professionals, system administrators, tools vendors, cloud providers, and people who spend their time building and keeping the computing infrastructure of business running. Many thanks to those who built this un-conference especially John Willis, Mark Hinkle, Damon Edwards, and all of the sponsors and presenters – you pulled together a great event.
Opscamp was created to allow an “open exchange of ideas around next generation technologies and strategies for IT Operations.” The big topics at this meeting (as measured by both passion and interest) were Monitoring, Configuration Management, and DevOps. I was particularly interested in these topics because we added the phrase “in cloudy environments” to the sessions.
I came away from this conference with a few conclusions:
1. IT operations is currently messy, and “cloud computing” is exposing some of the problems. One problem is that when the cloud delivers on the notion of allocating resources in minutes, the pressure switches from raw provisioning to the speed of configuration; with the timescale compressed from days to minutes, there is nowhere to hide. The other big issue is the dynamic nature of cloud computing – with servers, networks, and storage showing up and disappearing from your environment, it becomes difficult to model and manage. The manual or loosely connected processes that worked because there was time while you waited for hardware to arrive no longer work in a highly dynamic environment.
2. There are very different views about what monitoring in the cloud should be – deep and consistent with current processes, or “just hide the messy parts,” to quote Amazon. Some wanted the cloud providers to give them deep access to the “normal” monitoring data – think cpu temperatures. Others think that the value in the cloud is getting rid of the support and detailed attention to the infrastructure, so who cares about that data, it is for the infrastructure provider to manage. The issue is that the current cloud offerings are quite opaque about what is happening in the infrastructure.
After talking with Bret Piatt from Rackspace it seems that the problem is not collecting the data (they have to do it anyway), but storing it, processing it, and doling it out to the customers. There is real expense here, and for those customers who don’t want it, they would have to carry the burden of the extra cost in a market that is sensitive to price. The fundamental choice here is whether we are going to carry the current detailed monitoring practices into the cloud, or give up on the low level monitoring of the remote resources. In order to carry forward the current models, the cloud providers will have to provide API’s to get at the internal data. In order to change the model, developers and tool providers will have to increase the capabilities of the applications around fault tolerance and providing useful monitoring data that can be measured from the application level.
3. We are still in the early days of cloud computing. This should not be a surprise to anyone, but since at CloudSwitch, we’ve been so fully integrated with the cloud since the beginning – with everything from our build servers, bug database, large number of QA servers, code repository and web site –it’s easy to think that everyone is already comfortable using the cloud. What became clear in the discussions is that many of the organizations are still working though server virtualization and how to take advantage of the advanced features enabled by that technology.
What was really exciting for me was to see the people who actually have to make IT work thinking through what is needed to build a dynamic data center and integrate with cloud computing. With these guys on the case, I know that cloud computing is taking real steps forward.

Digg
Reddit
Delicious
StumbleUpon
Facebook
Twitter
LinkedIn