Mountkirk Games Case Study

Non-italicised points represent notes taken from Google’s official case study documentation, with keywords highlighted in bold. Points in italics represent additional insights and possible solutions.

Overview

• Online, session-based multiplayer mobile games
• Already migrated on-prem to Google Cloud
• Starting to expand from mobile to other gaming platforms
• Latest game:
  • Retro first-person-shooter (FPS)
  • Expected to be popular
  • 100s of simultaneous players
    • Not a huge number, but sizable
  • Games take place in a geo-specific ‘arena’
  • Multiple platforms and locations
  • Digital real-time banner displaying global leaderboard
• Planned architecture:
  • Plan to deploy backend to GKE
    • Multiple regional clusters
    • GKE Enterprise for common config management and service mesh capabilities
    • Multi Cluster Ingress
  • Wish to scale rapidly based on demand
  • Usage of global load balancers to route players to the closest regional arena
  • Usage of a multi-region Cloud Spanner cluster to keep global leaderboard in sync
    • Streaming data

Existing Environment

• Recently migrated to Google Cloud
• Five games migrated via lift-and-shift to Compute Engine VMs
  • Suggests Migrate for Compute Engine
• Each new game exists in an isolated project, with a parent folder maintaining permissions and network policies
• Legacy games with lower traffic have been consolidated into a single project
• Dev/test environments also exist

Business Requirements

• Support multiple gaming platforms
  • Potentially multiple back-ends
• Multi-region
  • Consider different sets of compliance requirements
• Rapid iteration of features
  • CI/CD
  • Cloud Build / Cloud Deploy
• Minimize latency
  • Global Load Balancers
  • Premium network tier
  • Cloud CDN
  • Caching with Memorystore
• Optimize for dynamic scaling
  • GKE node and pod horizontal autoscaling
  • Spanner doesn’t autoscale—need to implement custom solution, e.g.
    • Cloud monitoring (CPU), Pub/Sub, Cloud Functions
• Use managed services and pool resources
  • GKE autopilot not an option, as GPUs required
  • Pooled resources: shared project for management and analytics
    • Log buckets, BigQuery, Anthos
• Minimize costs

Technical Requirements

• Dynamically scale based on game activity
  • Horizontal pod autoscaler and node autoscaling
• Publish scores on near real-time leaderboard
  • Streaming data
• Game activity logs stored in structured files for analysis
  • Import to BigQuery from Logging
  • Store logs in Cloud Storage—load as BigQuery external table
• Render graphics service-side via GPUs for multi-platform support
  • GPU support in GKE (but no Autopilot)
• Eventual migration of legacy games to platform

Executive Statement

• Build on existing success of Google Cloud
• Wish to analyse player behaviour and game telemetry
  • Cloud Bigtable for telemetry storage
  • BigQuery, BigQuery ML
• Use cloud native design principles
• Latency is a top priority
• Cost management a second priority
• Expect advanced analytics capabilities to rapidly iterate bug fixes and new functionality

Potential Solution Design

References

• Mountkirk Games case study