Data @Scale Seattle- Naresh Sundaram
summary
Outlook Service chooses a single sharding axis — shard by user or group — and runs every component (cache, index, blobs, compute) on the same server hosting that shard, blending co-located hot and cold data on the same spindles. That design yields simple horizontal scaling, microsecond cross-service calls, efficient cache use, and cheap cold storage (under $0.01/GB·yr claimed). At scale it runs across 50+ data centers, ~100k servers, ~300+ PB logical (≈1+ EB physical), ships daily updates, and trades many small failures for far fewer catastrophic outages.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:55Outlook Service
- 3:13Scale
- 5:35Design Choices
- 5:53Service Composition
- 8:51Request Routing
- 10:53Advantages
- 16:26Colocation
- 24:02Questions
- 25:57Use Cases
- 27:12Data Duplication
related talks
32mSharding for mission-critical, internet-scale apps
23mSharding for hyperscale and geographically distributed relational databases
47mWeb-Scale IT Engineering.
44mScaling Horizons Effective Strategies for Wix's Scaling challenges
29mDesign of a Composable Infrastructure Platform
51mScaling yout Architecture with Services and Events
47mApplying “web scale” patterns (...)
19mDataEngBytes 2023 - SYD-T1-04 - Yash Sharma