AppsFlyer scales attribution analytics to 400B reads/day with Aerospike
About AppsFlyer
AppsFlyer is the world’s leading mobile attribution and analytics platform, enabling app marketers to measure and optimize their campaigns with unmatched precision. Serving more than 65% of the global mobile measurement market, AppsFlyer supports app developers and marketers across 19 offices, with a data footprint spanning billions of events daily. Its SDK is installed on most mobile devices worldwide, supporting a data-driven ecosystem for app growth.
Scaling attribution at massive global scale
To meet increasing real-time data demands from mobile advertising campaigns, AppsFlyer needed a backend that could support petabyte-scale throughput while maintaining millisecond SLAs.
Legacy database instability and downtime
Their previous installs database ran on ~50 AWS instances, with frequent hardware failures, unreliable replication across datacenters, and backup processes that exceeded 24 hours.
Growing traffic and unpredictable peaks
Events like global sporting tournaments or shopping holidays caused traffic spikes that required rapid, elastic scaling.
Cost and operational inefficiency
Hardware costs and support limitations made the system fragile and expensive. The team operated under constant risk of production-impacting failures.
Blazing-fast infrastructure that scales elastically
Aerospike replaced the legacy NoSQL system with a key-value store purpose-built for low-latency, high-throughput workloads. The migration and rollout were completed in just over two weeks.
Massive throughput at ultra-low latency
Aerospike supports 2.5M writes/sec and 4.5M reads/sec, with p99.9 latency under 1ms—even under peak load. Total reads exceed 400 billion per day.
Elastic scalability powered by AKO on Kubernetes
AppsFlyer runs Aerospike fully as a stateful workload in Kubernetes (EKS) using the Aerospike Kubernetes Operator (AKO). This enables reliable scale-down by 50% and scale-up by 100% on demand—with zero performance degradation.
Cost-efficient architecture with Graviton and Karpenter
Running on AWS Graviton instances and orchestrated with Karpenter, AppsFlyer achieved a 60% reduction in compute costs. Node provisioning completes in under one minute.
Real-time performance with radically lower cost and effort
AppsFlyer achieved industry-leading performance with dramatically lower infrastructure and operational burden.
60% lower compute cost
Total infrastructure cost dropped significantly, even after including the Aerospike license. Internal modeling showed more than $10M in projected savings compared to DynamoDB at similar scale and traffic.
<1ms read latency at p99.9, even at scale
Performance remains consistent across 200+ dynamically scaling nodes.
Production migration in <3 weeks
The full evaluation, benchmarking, and production rollout of Aerospike on Kubernetes was completed in under 18 days.
Reduced operational overhead and tooling
Eliminated fragile scripts and legacy provisioning logic with Kubernetes-native automation and open source tools.
Testimonials
Before Aerospike, our installs database was unstable and costly. Backups took more than a day, and we were in constant fear of failure. With Aerospike, we immediately gained stability and cut hardware costs by two-thirds. The thing I like most? It just works.
Ido Barkan
AppsFlyerSoftware ArchitectAdditional resources
For a deeper understanding and more insights, explore these additional resources.
- Video
AppsFlyer: Large Scale NoSQL DB Migration Under Fire
Appsflyer is an ad tech. We’re a mobile attribution company. We help marketeers on mobile app advertisers to sell their apps to market their apps, to run campaigns. We measure how one campaign can excel comparing to the other. - Video
AWS re:Invent 2024 - Real time meets real world: How AI is making decisions at scale
Aerospike CPO Evan and AppsFlyer’s Victor Gershkovich discuss how AI-powered systems—like mobile attribution at AppsFlyer—depend on high-throughput, low-latency infrastructure. The session explores real-time scaling, Kubernetes, and how ML and deep learning models power billions of decisions per day.
Get started with Aerospike
For high-performance, scalable data management and ultra-low latency, ideal for handling massive datasets and real-time applications.
AppsFlyer scales attribution analytics to 400B reads/day with Aerospike
About AppsFlyer
AppsFlyer is the world’s leading mobile attribution and analytics platform, enabling app marketers to measure and optimize their campaigns with unmatched precision. Serving more than 65% of the global mobile measurement market, AppsFlyer supports app developers and marketers across 19 offices, with a data footprint spanning billions of events daily. Its SDK is installed on most mobile devices worldwide, supporting a data-driven ecosystem for app growth.
Scaling attribution at massive global scale
To meet increasing real-time data demands from mobile advertising campaigns, AppsFlyer needed a backend that could support petabyte-scale throughput while maintaining millisecond SLAs.
Legacy database instability and downtime
Their previous installs database ran on ~50 AWS instances, with frequent hardware failures, unreliable replication across datacenters, and backup processes that exceeded 24 hours.
Growing traffic and unpredictable peaks
Events like global sporting tournaments or shopping holidays caused traffic spikes that required rapid, elastic scaling.
Cost and operational inefficiency
Hardware costs and support limitations made the system fragile and expensive. The team operated under constant risk of production-impacting failures.
Blazing-fast infrastructure that scales elastically
Aerospike replaced the legacy NoSQL system with a key-value store purpose-built for low-latency, high-throughput workloads. The migration and rollout were completed in just over two weeks.
Massive throughput at ultra-low latency
Aerospike supports 2.5M writes/sec and 4.5M reads/sec, with p99.9 latency under 1ms—even under peak load. Total reads exceed 400 billion per day.
Elastic scalability powered by AKO on Kubernetes
AppsFlyer runs Aerospike fully as a stateful workload in Kubernetes (EKS) using the Aerospike Kubernetes Operator (AKO). This enables reliable scale-down by 50% and scale-up by 100% on demand—with zero performance degradation.
Cost-efficient architecture with Graviton and Karpenter
Running on AWS Graviton instances and orchestrated with Karpenter, AppsFlyer achieved a 60% reduction in compute costs. Node provisioning completes in under one minute.
Real-time performance with radically lower cost and effort
AppsFlyer achieved industry-leading performance with dramatically lower infrastructure and operational burden.
60% lower compute cost
Total infrastructure cost dropped significantly, even after including the Aerospike license. Internal modeling showed more than $10M in projected savings compared to DynamoDB at similar scale and traffic.
<1ms read latency at p99.9, even at scale
Performance remains consistent across 200+ dynamically scaling nodes.
Production migration in <3 weeks
The full evaluation, benchmarking, and production rollout of Aerospike on Kubernetes was completed in under 18 days.
Reduced operational overhead and tooling
Eliminated fragile scripts and legacy provisioning logic with Kubernetes-native automation and open source tools.
Testimonials
Before Aerospike, our installs database was unstable and costly. Backups took more than a day, and we were in constant fear of failure. With Aerospike, we immediately gained stability and cut hardware costs by two-thirds. The thing I like most? It just works.
Ido Barkan
AppsFlyerSoftware ArchitectAdditional resources
For a deeper understanding and more insights, explore these additional resources.
- Video
AppsFlyer: Large Scale NoSQL DB Migration Under Fire
Appsflyer is an ad tech. We’re a mobile attribution company. We help marketeers on mobile app advertisers to sell their apps to market their apps, to run campaigns. We measure how one campaign can excel comparing to the other. - Video
AWS re:Invent 2024 - Real time meets real world: How AI is making decisions at scale
Aerospike CPO Evan and AppsFlyer’s Victor Gershkovich discuss how AI-powered systems—like mobile attribution at AppsFlyer—depend on high-throughput, low-latency infrastructure. The session explores real-time scaling, Kubernetes, and how ML and deep learning models power billions of decisions per day.
Get started with Aerospike
For high-performance, scalable data management and ultra-low latency, ideal for handling massive datasets and real-time applications.