The first conference day started a bit slowly with a keynote from the creator of CFEngine, a configuration management software. He provided his thoughts on where he thinks configuration management is going. He proposed that the next step for configuration management comes from a deeper understanding of the interactions and relationships between applications and the people who create them and use them. The bottom line is that we still have some ways before we get there.
Second presentation was from one of the architect who designed a new feature in Powershell, Desired State Configuration. While the new feature was interesting, I found the presentation to be an overlap of a presentation I saw from TechEd. While the presenter was knowledgeable, the presentation was a bit shaky. I had the opportunity to ask a few questions regarding possible tooling in Visual Studio and configuration file manipulation. The answer: not there yet! While DSC is interesting, there are significant gaps as this is a v1 of the feature. Definitely more to come on this.
Also saw a presentation from Ansible but was not impressed. While the software is interesting, it’s not groundbreaking I find.
My favorite presentation of the day was from Eric Brewer from Google, Parallelism in the Cloud. He spent some time explaining a challenge that both Google and Global Excel is facing (lol yes, you read that correctly): mixing oltp and batch processing for maximum resource utilization. His solution to this was Akeros, a new experimental OS that allows batch jobs to be quickly interrupted by more time sensitive workload. See the following for more info:
The last presentation of the day was from a release engineer at Google. One can only be impressed at the scale and pace at which Google releases applications. We’re talking about petabytes of code released DAILY on hundreds of thousands of servers. Google+ releases code every hour in production! While the scale is impressive, the tools used are relatively simple to understand. They’re using a combination of TGZ archives stored in their Colossus file system augmented with metadata stored in a BigTable database. The packages are distributed with a variation of ttorrent.
The poster session at the end of the day gave me some insights as to what kind of research is going on in the parallelism world. A lot of it is about providing abstractions and optimizations to developpers to better leverage multiple cores of CPU and GPU in a power efficient manner. I saw some of the concepts I discussed a few years ago on some of the posters, which made me think my ideas were not so crazy after all! See the following posts:
It was overall an interesting day that was in line with what I’m working on right now, automated build deployment and configuration using Powershell. I’ll be sharing more on this very soon!