Reduce game hosting costs with Spot fleets - Amazon GameLift Servers
How Amazon GameLift Servers works with Spot fleetsDesign considerations
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Reduce game hosting costs with Spot fleets

When hosting multiplayer game servers using Amazon GameLift Servers managed hosting, Spot instances can provide a cost-effective alternative to On-Demand instances. The Spot pricing model offers the same hardware and performance as On-Demand but at potentially significant cost savings (up to 70-90%). However, they come with a limitation: when Amazon needs capacity back, it can reclaim these instances with a two-minute interruption notification.

Amazon GameLift Servers mitigates the risk of interruptions for game server hosting. Amazon GameLift Servers predicts the likelihood of interruptions on Spot instance types and avoids placing game sessions on any instances at risk. If a rare interruption does occur, the notification allows you to gracefully end a game session for players.

How Amazon GameLift Servers works with Spot fleets

When you set up Spot fleets for game hosting, Amazon GameLift Servers continually evaluates your Spot fleet instance types and locations for game hosting viability.

  • The Spot viability algorithm analyzes recent availability patterns and historical interruption rates of Spot instance types by location.

  • Based on this analysis, Amazon GameLift Servers identifies Spot instance types and locations where there's an unacceptable potential for game session interruption. It takes the following actions:

    • It marks the combination of instance type and location as temporarily non-viable.

    • It removes any non-viable Spot fleet location from consideration when placing new game sessions. As a result, game sessions are only placed in Spot fleet locations that have a high likelihood of uninterrupted game server hosting.

    • It drains the Spot fleet location of existing instances, even if Amazon doesn't reclaim them, so you're not paying for instances that can't be used for game hosting. If game session protection is turned on, instances are shut down only after active game sessions are completed.

  • Amazon GameLift Servers continually re-evaluates your Spot fleet instance types and locations for game hosting viability. When a previously non-viable instance type becomes viable again based on updated historical data, you can scale up the Spot fleet again, and Amazon GameLift Servers will resume placing game sessions with it.

Design considerations

When designing your solution to use Spot fleets, consider the following issues:

  • Assess game session length – The average length of your game sessions can impact how well Spot works for your game. With shorter game sessions, the faster turnaround keeps game sessions running on viable instance types based on the latest historical data. Longer game sessions continue to run on instance types without evaluating recent viability data, running a greater risk of interruption over time.

  • Evaluate instance type availability – Not every fleet location offers every instance type as Spot. When choosing an instance type for your Spot fleet, use the Amazon GameLift Servers console fleet creation tool to help you find Spot instance types in the locations you need. With this tool, you can select your fleet locations and then view instance type availability across those locations.

  • Create multi-location Spot fleets – You can create Spot fleets with multiple locations. A single multi-location Spot fleet deploys instances with the same instance type to multiple Amazon Web Services Regions or Local Zones. The Spot viability algorithm evaluates viability based on both instance type and location. If a Spot fleet location is evaluated as nonviable, it doesn't impact other locations in the fleet, which can still be used to host game sessions.

  • Create queues with Spot fleet diversity – If you're using Spot fleets for game hosting, you need to set up a game session placement queue. For each new game session request, the queue looks for available game hosting resources and selects the best possible option. With Spot fleets, you want a queue that can search across multiple fleets that vary in both locations and instance types, and you want to include at least one On-Demand fleet as backup capacity. A well-designed multi-fleet queue that offers diverse placement options is highly resilient against interruptions, slowdowns, and outages. For additional guidance on designing a queue for Spot, see Build a queue for Spot Instances.

  • Handle interruptions gracefully – Set up your game server to minimize player impact in the event of a Spot interruption. When Amazon reclaims a Spot instance, Amazon GameLift Servers passes the termination notification to all affected server processes using the server SDK callback function onProcessTerminate(). Your game needs to implement this callback to end the game session gracefully. For more information, see Respond to a server process shutdown notification.

    Note

    Amazon makes every effort to provide the notification before it reclaims an instance, but it's possible that Amazon reclaims the Spot Instance before the warning arrives. You should also prepare your game server to handle unexpected interruptions.

  • Configure auto-scaling for your backup fleets to maintain service during a Spot interruption. Target-tracking auto-scaling maintains a capacity buffer and scales automatically with demand. With auto-scaling, backup fleets (Spot or On-Demand) will begin increasing capacity whenever they start receiving more game session requests.

    To quickly replace lost capacity when a Spot fleet becomes non-viable, a custom scaling mechanism can use available queue and fleet metrics to initiate a rapid scale-out of backup fleets. Detect when Spot fleets become non-viable with metrics such as FirstChoiceOutOfCapacity, FirstChoiceNotViable, and PercentAvailableGameSessions. Estimate replacement capacity needs by analyzing recent PlacementsStarted metric data. After scaling backup fleets to handle immediate demand, normal auto-scaling can take over.

  • Integration with FlexMatch – If your solution uses a FlexMatch matchmaker, there are no special requirements for Spot fleets. You can configure a matchmaker to use a queue with Spot fleets. Amazon GameLift Servers automatically prioritizes match placements across Spot and On-Demand fleets, including when placing new game sessions and when backfilling empty player slots in existing game sessions.