When developing its cloud-based enterprise mobility management (EMM) solution, MobileIron focused on two main principles: simplicity and scale. In a recent blog post, Ojas Rege wrote about the process of making its offering scalable enough to handle the billion new smartphones that IDC (News - Alert) predicts will make their debut in the next three years.

Mobile data is big data, so in crafting the service the company knew it needed to scale.

A typical approach to scaling is just throwing more hardware at the problem, but MobileIron knew it couldn’t go that route.

“That’s expensive and not sustainable,” wrote Rege. “The architectural approach to scale is to identify the bottlenecks in advance and design the system to minimize or eliminate them.”

The approach started with learning to scale a single node. Mobile devices need to communicate regularly with the EMM service to ensure they are in compliance, but the typical model was one thread per connection, and that didn’t scale well. What MobileIron did was enable the ability to park connections using asynchronous processing of HTTP requests until the system is ready to respond, increasing the number of devices a node can support. It also reduced memory overhead by eliminating object-relational mapping that abstracts database communications but increases memory use.

After fixing vertical scaling, the company tackled horizontal scaling by going stateless. The MobileIroin system transparently picks up and distributes load as new nodes join the cluster, according to Rege.

“Being stateless also makes high availability easier, because you don’t have to do complex session replication in order to ensure that a device or user isn’t left hanging when a node or cluster goes down,” he wrote.

When it came to reducing database load, the company opted for an in-memory data grid that all nodes can access. This basically serves as a massive piece of virtual RAM (News - Alert) that is shared across all nodes with a high-speed bus, so there are less database accesses.

For boosting performance, the company also tackled the database issue by denormalizing the database—flattening it, basically.

“However,” noted Rege, “there are other ways to offload your big data such as search engines for contextual search, MapReduce for crunching large datasets, NoSQL for large documents, and key/value stores.”

For scaling authentication, MobileIron designed its authentication system as an independent, highly available module instead of tying it to a specific cluster. That helped support global scaling, making it easier for the company to provide features, such as universal login, so individual users never need to know specific URLs to access the system.

Finally, noted Rege, the company built in flexibility to be able to leverage not only its own data centers, but also those of providers such as Amazon. To do that, however, the company had to make sure that the system was highly automated and required no manual intervention.

“This flexibility needs to be built into the cloud EMM solution from day one,” said Rege.

The result of this effort has been an EMM that scales well enough to handle all one billion of the new smartphones that should be coming on the market soon.

While most of those won’t be leveraging an EMM, if they all do end up on the MobileIron platform, the company will be ready.