← All comparisons

Uber (Driver Matching) vs Yelp (Location-Based Search)

Uber (Driver Matching)

Uber (Driver Matching)

Geo-indexing, dispatch, real-time location streams.

Components (9)

  • Rider App
  • Driver App
  • API Gateway
  • Dispatch
  • Trip Service
  • Kafka
  • Geo Index
  • Postgres
  • Cassandra

Headline numbers

  • Location pings / sec~250,000/sec
  • Dispatch queries / sec~5,000/sec
  • Matches / sec~1,000/sec
Yelp (Location-Based Search)

Yelp (Location-Based Search)

Geohash nearby search via a read-only location service, a geohash cache and a business-info cache, with a separate business service for owner edits.

Components (11)

  • Mobile Client
  • Load Balancer / API
  • Location-Based Service
  • Business Service
  • Review Service
  • Geohash Cache
  • Business-Info Cache
  • Geohash Index Table
  • Postgres
  • Cassandra
  • CDN

Headline numbers

  • Search QPS~5K/sec avg, ~25K peak
  • Geohash index sizefits in memory
  • Cache footprint (cell lists + objects)tens of GB

Key differences

Only in Uber (Driver Matching)
  • Kafka
In both
  • Client
  • Api Gateway
  • Service
  • Cache
  • Database
Only in Yelp (Location-Based Search)
  • Geohash Index Table
  • CDN

Flow shape

Uber (Driver Matching) flows
  • Request a ride7 steps
  • Driver location update4 steps
  • Geo Index is down4 steps
Yelp (Location-Based Search) flows
  • Search nearby4 steps
  • Search nearby (cold cells)5 steps
  • Owner edits business info (next-day)4 steps
  • Submit a review4 steps
  • Geohash index is down4 steps