Scaling for the Expected and Unex
summary
Dealnews.com serves a static Apache front page updated by cron every five minutes to absorb unexpected referrals and uses proxy caches (Varnish/Squid) with front‑page TTLs of 1–2 minutes and max object TTLs of 5 minutes. Architecture denormalizes ~300 production tables into ~30 read tables on replicated MySQL so reads use primary keys only. Goal is to handle three times the Yahoo spike; load testing with Keynote simulates 6,000 concurrent requests from eight locations. Third‑party tags and ads live in bottom iframes.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 0:47What happens when something unexpected occurs
- 3:22What is acceptable service
- 5:23Caching
- 6:42Handling Expected Growth
- 9:26Proxy Servers
- 11:15Cache Stampedes
- 18:08Scaling for the Expected
- 23:20ThirdParty Content
- 24:40Losing Memcache
related talks
45mScaling: from 0 to 20 million users
45mSurviving a Prime Time TV Commercial
56mLarge scale websites performance optimisation tricks
39mFrom Stubbies to Longnecks
41mHigh Performance Kick-Ass Web Apps
11mDatabase Performance in High Traffic Environments
18mHigh-Load Systems: Challenges in Social Network Development
47mScaling: from 0 to 25 million users