Gaming companies can ensure a successful launch by having the necessary infrastructure to handle client demands. Gamers are often eager to get their hands on the product as soon as possible—so a business that can’t provide the product because their online infrastructure is too weak to handle the demand will spoil their community’s goodwill. The loss of goodwill can decimate sales momentum and turn what should have been a hiccup into a game ender.
The GameTree Platform: A Case Study on Load Testing
Apica recently worked with Toronto-based TransGaming on the launch of GameTree, an on-demand gaming service for digital TVs and CEs. Hosting the GameTree platform is complicated because traffic spikes hit the server instantaneously when clients turn on the channel. The severity and sudden nature of the spikes makes it difficult to plan out server scaling in a way that’s able to keep up with demand.
Regular load testing throughout the development process helps ensure the game performs smoothly without errors at launch. Load testing simulates virtual users within a program to test how many concurrent users the hardware can handle before performance falters. Load tests produce data that shows how many transactions and how much data the servers are pushing against the number of concurrent users accessing the platform over time. In GameTree’s case, the initial test showed that the infrastructure began to lag behind demand at around 650 users. The throughput peaked at 14 Mbps before nosediving at peak load, while the web transaction rate peaked at about 80 URL/s before dropping out. It’s worth noting that concurrency issues such as these may not even show up in normal QA testing.
The transaction rate and network throughput need to keep up with demand in order to provide a smooth experience. TransGaming’s development team was able to take the load test data and identify bottlenecks within the platform code to improve performance. In this case, Apica’s team worked to simulate a set-top box environment and ran the tests from multiple geographical locations to simulate real world use over the course of three weeks. The various load tests across different hardware configurations found that it made better financial sense for GameTree to use more “large” servers than fewer “extra-large” ones. These tests would not have been exceedingly difficult without real-world load testing scenarios.
Load tests establish optimal machine sizing and elasticity rules for adjusting cloud resource availability to match traffic patterns: the servers need to handle peak loads and adjust to traffic spikes. The initial test showed substantial variance in throughput and web transaction rate as more users accessed the platform; the service encountered hiccups while additional hardware was turning on under the old scaling method. It’s common for infrastructure to scale by adding more server resources as the user-count rises. However, Apica’s test results found that the GameTree system’s traffic spikes were so sudden and massive that the system couldn’t add resources fast enough. Instead, Apica recommended utilizing a scheduling plan based on traffic trends to meet demands.
Load testing provides tangible data that helps troubleshoot performance problems and optimize infrastructure for smoother performance and smarter expenses. After applying Apica’s test data to the platform, the new scaling logic and fixed bottleneck issues improved performance by 400 percent. The infrastructure was able to smoothly scale from 0 to 1,500 concurrent users without service disruptions.
To find out more about how Apica can help your company’s game launch go off without a hitch, check out this white paper on the TransGaming test.