Grab Redesigns Counter Service Storage for 50% Lower P99 Latency
50% latency drop, 3TB to 1TB. Here's how Grab redesigned its counter service to cut costs by 45-50%.
Why it matters
Grab's migration from a wide-column database to Aerospike demonstrates a concrete infrastructure optimization pattern — data model redesign plus storage abstraction can yield substantial latency, capacity, and cost wins at scale. Relevant to practitioners operating high-volume transactional systems and evaluating database trade-offs.
The key facts
14 to knowP99 read latency reduced by approximately 50%
Disk usage reduced from 3 TB to 1 TB
Cost per node reduced by 45-50%
Migration strategy: storage abstraction, shadow traffic, data parity validation, gradual traffic migration
Data model change: time buckets consolidated into map-based records
Source system: wide column database; target: Aerospike
Company: Grab (high-volume ride-hailing/delivery platform)
Published: October 7, 2026 by InfoQ
Grab migrated Counter Service from wide-column database to Aerospike
50% lower production p99 read latency achieved
45–50% lower cost per node
Used storage abstraction, shadow traffic, data parity validation, and gradual traffic migration
Redesigned data model consolidated time buckets into map-based records
Published October 7, 2026 on InfoQ
Go to the source
InfoQ AI/MLinfoq.com
Publisher excerpt: Grab migrated its high-volume Counter Service from a wide column database to Aerospike using storage abstraction, shadow traffic, data parity validation, and gradual traffic migration. The redesigned data model consolidated time buckets into map-based records. Grab reports about 50% lower…