Kakaranet Presentation

Synrc Research Center authored Turkish Kakaranet Social Gaming for classic turn-based board games. Built in pure Erlang environment Kakaranet is easy to manage and support. Perfectly outsourced following Quick and Well strategy.

Transcript

1. KAKARANET Social Gaming Maxim Sokhatsky maxim@synrc.com
2. Technical Input How modern portal should looks like ? • Active Fast Comet Pages • High-load Ready • Reliability and Failover • Ditribution for Speed • Monitoring tools • Facebook-like facilities and integration • Simple and small time-to-market • Single Environment
3. Underlying Products Quick and Well strategy Riak • High Availability • Monitoring Tools • Best In-class • Stable • Manageable RabbitMQ • Connecting People • Deliver Messages • Invented in Banks • Language Opaque • Fast GProc • Distributed In- memory database • Consistent • Complex Queries Erlang • Automatic Fail-over • Purity of code • High Reliability • High Load • Mature Infrastructure Cowboy • Fastest Web Server • SSL Ready • Modular • Young but stable Nitrogen • Best Comet Server • Clustered • Typed HTML • Rich JavaScript • Easy to Develop • Event driven
4. Social Web Application Simple but modern • Users and Groups • Free to register • Direct Messages • Friends • Likes • Feeds • Media Files as Attachments, Music, Video • Drop-down Auto-completion • Comet Pages with Rich JavaScript • Pub/Sub key-routing with RabbitMQ
5. Game Server Nodes Easy to Write new games • AMF listeners and Flex Clients • Tables, Games, Tournaments model • Access to Riak and GProc databases • Matching Players Each Other in ranges • Global Registry of all games in Cluster • Persist Game Data • 40KB per game on server-side
6. Sequential Consistency Guard Processes • All entities that could be changed controlled by Erlang Process spawn on bootstrap • Respawn on failover on another node • Back-end Workers for ordering write requests in RabbitMQ queues • All read operations still available for anyone in cluster • Examples: Feeds, Users, Groups, Lists
7. Cluster Levels Diferrent Tasks Require Different Clusters Database • Riak is present in each node in cluster Memory • Each Process registered in global database Routing • Clustered RabbitMQ to handle routing Filesystem • GlusterFS on each server
8. Cluster Line Erlang/OTP Releases per OS app@srv1 • Workers • Riak R/W • GProc game@srv1 • Games • Riak R/O • GProc web@srv1 • Nitrogen • Riak R/O • Gproc • Cowboy rabbit@srv1 • RabbitMQ • Routing Databse
9. Three Production Lines Continue running after single server down SRV1 • app@srv1 node • game@srv1 node • web@srv1 node • rabbit@srv1 node • SNMP Agent / node • Riak Control / node • 8x Xeon 3GHz • 30GB RAM • 512GB SSD • 3TB RAID SRV2 • app@srv2 node • game@srv2 node • web@srv2 node • rabbit@srv2 node • SNMP Agent / node • Riak Control / node • 8x Xeon 3GHz • 30GB RAM • 512GB SSD • 3TB RAID SRV3 • app@srv3 node • game@srv3 node • web@srv3 node • rabbit@srv3 node • SNMP Agent / node • Riak Control / node • 8x Xeon 3GHz • 30GB RAM • 512GB SSD • 3TB RAID