Sessions at MongoSF 2012 about Scaling on Friday 4th May

Optimizing the database layer for most web applications tends to be mostly about fast reads. That is why the majority of benchmarks and technical blogs focus on read-intensive workloads. At Mailgun, up to 75% of queries are ordered updates and deletes. This presents a number of challenges. This talk is about lessons we learned at Mailgun while building the highly available and fast back-end for our ever-changing message queue on top of MongoDB.

Developed a redundant, scalable database that can replicate naturally to external data centers (for geographic redundancy). As Scene7 SaaS clients have high cache reuse, we implemented a front side cache to the database such that the server footprint is minimized but each individual data center can scale and handle all requests for a geography should the other data center go offline.

Last year craigslist deployed MongoDB for its multi-billion document posting archive, largely due to its schema-free nature and built-in sharding and replica sets. Since then we've looked at it for other projects--specifically high-volume and multi-datacenter. In the process we've learned more about where other features do and don't work so well, including replication, capped collections, and compound indexes. This presentation wil recap what's worked well for us and discuss the other issues we ran into for new projects, as well as possible improvements in the the design or enhancements for MongoDB.

This talk will introduce deployment of MongoDB across multiple data centers. We'll discuss the advantages of a multi data center deployment for read/write locality, the various deployment strategies, and disaster preparedness and recovery. In addition, we'll look at the MongoDB roadmap and planned enhancements around data center awareness.