How eBay Scales to Billions: 7 Proven Practices for Massive Web Systems

At eBay, scalability is a daily architectural pressure, with billions of page views and petabytes of data, making it a make‑or‑break concern.

Best Practice #1: Functional Partitioning

Group related functionality together and separate unrelated parts, whether called SOA, functional decomposition, or engineering secrets. eBay isolates functions into application pools (e.g., sales, bidding, search) across ~16,000 servers and into logical databases spread over 400 physical hosts, enabling independent scaling.

Best Practice #2: Horizontal Sharding

Beyond functional separation, workloads must be broken into many small, manageable units. Stateless application layers use load balancers to route traffic, allowing simple addition of servers. For stateful data, eBay shards databases by primary access paths—e.g., user data across 20 hosts—using techniques such as modulo hashing, range partitioning, lookup tables, or hybrid strategies.

Best Practice #3: Avoid Distributed Transactions

Instead of costly two‑phase commits, eBay relaxes cross‑system transaction guarantees, favoring eventual consistency. Most operations are auto‑committed; where needed, multiple statements are bundled into a single transaction. Techniques like ordered DB operations, asynchronous events, and reconciliation batches achieve the required consistency without sacrificing availability.

Best Practice #4: Decouple with Asynchronous Strategies

Asynchronous communication (queues, multicast, batch jobs) removes tight coupling, allowing components to scale independently and remain available even if peers fail. Internally, architectures such as SEDA provide staged, event‑driven processing while keeping the programming model understandable.

Best Practice #5: Turn Processes into Asynchronous Streams

By moving non‑critical work to background processing, response latency for users drops dramatically. Activities like activity tracking, invoicing, settlement, and reporting are handled asynchronously, smoothing load peaks and reducing infrastructure costs.

Best Practice #6: Virtualize Every Layer

Virtualization abstracts hardware, OS, databases, load balancers, and network endpoints. eBay virtualizes databases so logical hosts map to physical machines, and virtualizes search across partitioned query grids, providing flexibility for scaling, rebalancing, and fault tolerance.

Best Practice #7: Use Caching Wisely

Cache read‑heavy, rarely changing data (metadata, config, static content) using push and pull strategies, while avoiding caching mutable or session data to preserve correctness. Proper cache sizing balances memory pressure against latency gains, and over‑reliance on cache can create new availability risks.

Scalability is not a non‑functional afterthought; it is a prerequisite functional requirement that must be prioritized above all else.