Using DynamoDB and the autosave mechanism allowed GREE to resolve all challenges at a high level. Another Eden was in development for nearly two and a half years—a relatively long time for a mobile social game—and DynamoDB was up and running for roughly two years of that cycle. During this period, the initial 20 data tables grew to several hundred, and this expansion continues with each upgrade. This expansion takes place with almost no involvement from server-side engineers.

"Because the server-side architecture is so simple, there's very little need for client engineers to worry about the server application's characteristics or communication model, and the costs of learning its specifications are greatly reduced. During normal development, they can forget that the server-side application even exists. And this remains just as true now—after more than three million downloads—as it was during the development phase. The result is that we can devote engineering resources to enhancing the fun that directly improves the game's marketability, or enriching the analysis infrastructure that makes this possible," says Suzuki.

For a mobile-social game, the period of growth just after release is critical. How well the game handles this demand can affect the fate of future products as well. However, it is very difficult to predict exactly how strong this demand will be. At GREE, estimates of a product's marketability or a promotion's success are often fluid right up until launch, with efforts to maximize these factors continuing until the very last moment. This meanspredictions can miss the mark in either direction.

"This was true of Another Eden as well, but DynamoDB was just as scalable as we had hoped. In the end, the access figures after the launch of Another Eden were several times what we expected, with more than one million downloads in a very short period of time. But with Amazon EC2 and DynamoDB Auto Scaling, there wasn't a single service interruption due to lack of server resources—and hardly any due to hardware problems," says Yu.

At peak access times, the burden on the servers can be 5–10 times what it is during quiet periods in the middle of the night, but GREE is able to keep costs to a minimum by automatically setting appropriate server capacity based on trends leading up to the previous day. Even when access has spiked during special events or after upgrades, the multiple layers of load-response systems, including advance scheduling and alert-based scale out, have operated so smoothly that it has not once been necessary to assign a dedicated team member in the several months since the game's release.

"Because DynamoDB Streams lets us safely feed data-change logs into the analysis infrastructure without affecting the service, all data-change records are stored on Amazon S3. This lets us aggregate all kinds of KPIs as well as performing individual, ad-hoc analyses of past data whenever necessary," says Yu.