Downloaded 54 times
Uploaded byTung Nguyen Thanh
Software architecture for high traffic website
The document presents a case study on the software architecture of Stack Overflow, highlighting its evolution through two major restructures to handle high traffic. It details key components like caching mechanisms, load balancing with HAProxy, and the use of different technologies such as Redis and ElasticSearch, achieving significant performance improvements. The architecture serves over 95 million pageviews a month with a focus on reinvention and scalable solutions.
More Related Content
![[Workshop] API Management in Microservices Architecture](https://cdn.slidesharecdn.com/ss_thumbnails/apimanagementinmicroservicesarchitecture-191029110253-thumbnail.jpg?width=640&height=640&fit=bounds)
![[2019] PAYCO 쇼핑 마이크로서비스 아키텍처(MSA) 전환기](https://cdn.slidesharecdn.com/ss_thumbnails/nhnforward201926-200122093003-thumbnail.jpg?width=640&height=640&fit=bounds)
Similar to Software architecture for high traffic website
More from Tung Nguyen Thanh
![Support, Monitoring, Continuous Improvement & Scaling Agentic Automation [3/3]](https://cdn.slidesharecdn.com/ss_thumbnails/agenticcommunityseries-day3-cfd-251120170304-ddef8112-thumbnail.jpg?width=640&height=640&fit=bounds)
Software architecture for high traffic website
- 1. Software architecture forhigh traffic website Case study - Stack Overflow Presenter: Ngô Xuân Hòa (Novaon Adnetwork - Novanet) Hanoi .Net Meetup
- 2. Contents About Stack Overflow ●Beginning ● Restructure #1 ● Restructure # 2 ● Founders ● Principles SO architecture ● StackExchange.Redis ● Dapper ● Jil Open-source Libs
- 3.
- 4.
- 5. 2008 Stack Overflow 2009 20102011 Server Fault Stack Exchange 1.0 Stack Exchange 2.0 Stack Overflow Carees Rome wasn’t build in a day!
- 6. ● 100+ Q&ASites ● 600+ million pageviews a month ● 3000+ requests per second ● 16+ million users ● 8+ million question ● 40+ million answers
- 7. Principles Perfomance Is aFeature Cache All The Thing! Reinvention is OK
- 8.
- 9. 2 times restructuring StackExchange 1.0 ● ASP.NET MVC ● SQL Server ● LINQ to SQL ● Wikipedia DB Design Stack Exchange Network LINQ to SQL HAProxy Redis Lucene.NET Scale Up ● Cache every things ● Elastic Search ● Reinvention
- 10. Stack Exchange 1.0Structure Windows NLBLoad balancing IIS Server IIS ServerWeb server SQL ServerDatabase
- 11. Window NLB ● Cons: ○Limit to 8 Nodes ○ Cannot detect service failed Web-tier ASP.NET MVC LINQ to SQL SQL Server ● All-in-memory ● Full text search
- 12. ● 16 millionpageviews a month ● 3 million unique visitors a month ● 6 million visits a month
- 13. Follow none butlearn from everyone!
- 14. Pros ● Bottleneck: DatabaseSQL Server ● High cost to scale up ● Simple Cons
- 15. Restructure #1 -Stack Exchange Network HAProxy Redis Cache Lucene.NET Tag Engine
- 16. Stack Exchange NetworkStructure HAProxy Redis IIS Servers Database protobuf sqlhttp http
- 17. Load Balancing ● HAProxy: ○Run in Linux ○ Free Web-tier ASP.NET MVC 3 LINQ to SQL jQuery 1.4.5 Lucene.Net Redis ● In-memory cache ● Master-slave ● Messaging notification
- 18. 3 Type Cache LocalCache Site Cache ● Use Redis ● Cache Site’s data: - Q&As - Acceptance rates - ... Global Cache ● Use Redis ● Cache System Data: - User info - Inbox - ... ● Use HttpRunTime.Cache ● Cache: - User Session - View Count - ...
- 19. Update cache flow- Local cache Local Cache Redis DB Other sites 1 3 2.1 2.2 4 1 - OnStartup - Subcribe invalidation message to Redis 2.1 - Data changed (by other sites, apps…) 2.2 - Send message to Redis 3 - Redis send Notification to Subscribers 4 - Get data from DB - update Local cache
- 20. Deployment flow withHAProxy ● Tell HAProxy to take the server out of rotation via a POST ● Delay to let IIS finish current requests (~5 sec) ● Stop the website ● Copy files ● Start the website ● Local testing, update local cache, etc… ● Re-enable HAProxy via another POST
- 21. ● High performance ●Low-cost Load Balancing (use HAProxy) ● Use Messaging của Redis for cache invalidation Pros ● Too many SQL query Cons
- 22. ● 95 millionpageviews a month ● 800 requests per second ● 16 million users
- 23. Restructure #2 -Scale Up Cache All the Thing Elastic Search Reinvention
- 24. Stack Exchange NetworkStructure Elastic Search Tag Engine Databases Redis HAProxy
- 25. 5 Level cache NetworkLevel Local Cache Redis Cache SQL SV Cache SSD ● Network Level: Browser cache… ● Local Cache: HttpRuntime.Cache - Cache all data in memory ● Redis Cache: Cache all data ● SQL Server Cache: Cache all data in memory (the database servers have 384GB of RAM)
- 26. Cache Flow ● CheckLocal Cache ● Else, check Redis Cache and update Local Cache ● If Cache Redis doesn’t have data, fetch from databases, then update Redis Cache and Local Cache
- 27. Cache All theThings!
- 28. Pros ● Data haslatency ● Very, Very Fast (<400ms) ● Low servers load: ○ IIS: 10-15% CPU usage ○ DB: 10% CPU usage ● 99% request served by cache Cons
- 29. ● 95 millionpageviews a month ● 800 requests per second ● 16 million users
- 30. Open-source Libs • StackExchange.Redis- high perfomance Redis client • Dapper - a micro ORM - very fast • Jil - fast JSON Serializer Reinvention is OK!
- 31. Reference sources ● http://stackoverflow.com ●http://highscalability.com ● http://codinghorror.com ● http://www.joelonsoftware.com ● http://nickcraver.com ● http://josephwoodward.co.uk/2014/02/the-architecture-of-stackoverflow/
- 32. Thank you! Ngô XuânHòa xuanhoa862001@gmail.com