Grab Migrated Counter Service Backend to Aerospike
The migration cut read latency in half and reduced storage costs by nearly 50 percent for a billion daily requests.
Updated on Oct. 7, 2026 in Data Centers
Live Poll
Do you trust that complex technical upgrades by large platforms lead to better services for you?
Grab has completed the migration of its Counter Service storage backend to Aerospike. This transition for a service handling one billion daily requests reduced p99 read latency by 50 percent and cut on-disk storage requirements from 3 TB to 1 TB.
Why it matters
The migration was driven by inefficiencies in the original wide-column database design, which required four network round trips per increment. Moving to Aerospike optimized both latency and capacity for the company's anti-fraud platform.
The transition reduced on-disk data from 3 TB to 1 TB and lowered the cost per node by 45 percent to 50 percent. Aerospike currently utilizes 64 bytes of primary index metadata per record to achieve these performance gains over the prior system.
The players
Grab
A ride-hailing and delivery super-app that maintains a large-scale distributed backend to support services like anti-fraud and transactional processing.
Aerospike
A high-performance NoSQL database company specializing in real-time data storage and low-latency access for massive-scale applications.
The details
To execute the migration without downtime, Grab implemented a storage facade—a software abstraction layer that decouples storage access from business logic. The engineering team employed shadow reads, where incoming traffic is processed by both the old and new systems simultaneously, to validate the performance of Aerospike before performing a gradual traffic shift. This approach allowed the team to consolidate data, ultimately reducing record counts by more than an order of magnitude.
Timeline
October 7, 2026: Details of the completed storage migration were published.
The Tech Race
Grab's shift reflects a broader trend among high-scale platforms moving away from multi-hop wide-column designs to prioritize lower-latency storage backends. This migration follows a competitive race to reduce compute overhead while maintaining sub-millisecond responsiveness for critical fraud-detection systems.
Users of Grab's services will see no direct interface changes, but the system's increased speed improves the reliability of the platform's anti-fraud processes. The migration ensures the infrastructure can continue to support one billion daily requests with higher efficiency.
The takeaway
This migration demonstrates how decoupling storage logic can yield significant efficiency gains for services operating at massive scale. Engineering teams should watch for similar performance benchmarks as companies look to consolidate data footprints on modern database backends.
Further reading
Learn more about the latest innovations in high-performance infrastructure at Data Centers.
Source note: This article includes information reported by InfoQ.
Live Poll
Do you trust that complex technical upgrades by large platforms lead to better services for you?
