Migrated a data platform processing 2 PB of data daily with zero production outages
Achieved 2 TB/s peak throughput by standardizing on Cloud Storage
Transitioned >95% of compute workloads to cost-effective Spot VMs on Google Kubernetes Engine (GKE)
Migrated 500+ production workflows to Managed Service for Apache Airflow (formerly Cloud Composer) with zero disruption
InMobi seamlessly migrated its 2 PB/day data platform to Google Cloud, achieving zero downtime and transitioning >95% to Spot VMs.
InMobi seamlessly migrated its 2 PB/day data platform to Google Cloud, achieving zero downtime and transitioning >95% to Spot VMs.
Managing petabyte-scale data to match
the pace of advertising
InMobi operates one of the world's largest independent mobile advertising platforms, connecting advertisers with consumers through highly contextual, real-time bidding. At the speed of advertising, latency and data processing capabilities directly dictate revenue.
Having previously leveraged Google Cloud to scale its real-time advertising marketplace to 7 million queries per second with zero downtime, InMobi now set out to modernize its massive backend data platform. Processing approximately 2 PB of data daily, InMobi’s legacy architecture was beginning to show operational strain. Hundreds of Apache Spark workloads were running simultaneously across shared Kubernetes node pools, making it incredibly difficult for the engineering team to optimize CPU, memory, and storage allocations.
Storage I/O bottlenecks had become a critical hurdle. With Spark workloads executing hundreds of millions of object reads and writes, tasks like metadata lookups were actively slowing down pipeline throughput. The shared infrastructure model also created a massive FinOps blind spot—InMobi lacked the granular visibility needed to attribute storage and compute costs back to specific business units.
Our engineers were spending too much time managing infrastructure and untangling shared costs. We needed a platform that allowed us to focus on building data products instead of just keeping the lights on.
Preetham Jayaram
Senior Staff Engineering Manager, InMobi
A zero-disruption modernization strategy
For a platform processing millions of requests per second, a traditional “big bang” migration was deemed too risky. Instead, InMobi architected and partnered with Google Cloud and Ollion on a decoupled, phased migration strategy.
The transition succeeded as a highly synchronized, three-way collaborative effort. InMobi’s engineering team led the core platform architecture and rigorous migration planning, while the Google Professional Services Organization (PSO) team provided critical production readiness validation for each workflow as part of this business-critical migration. Under this shared framework, implementation partner Ollion focused on the heavy lifting of infrastructure provisioning and supporting the migration.
By separating the orchestration, compute, and serving layers, InMobi ensured continuous operation. Ollion’s expertise was instrumental here, accelerating infrastructure readiness by automating the provisioning of Google Kubernetes Engine (GKE), Managed Service for Apache Airflow (formerly Cloud Composer), and Cloud Storage using Terraform.
Our decoupled migration strategy allowed us to validate each stage of the modernization independently before production cutover. Google PSO helped validate our approach at production scale, while Ollion's Infrastructure-as-Code automation accelerated execution, enabling a seamless migration with zero production outages.
Pushpendra Singh Chauhan
Senior Staff Engineer, InMobi
InMobi first migrated 100% of its workflow orchestration. By moving over 500 legacy workflows (DAGs) to a fully managed Apache Airflow service, the team independently validated scheduler scalability, concurrency, and operational readiness. During this phase, Managed Service for Apache Airflow orchestrated production workflows while the legacy compute platform continued executing the Spark workloads.
In parallel, the ad serving (application layer) was progressively validated from 0% to 50% of production traffic, while Bronze data was continuously reverse-synchronized to the legacy platform. This gave the ad serving team the flexibility to scale independently without coupling it to the data migration, until 50% production traffic was successfully achieved. Once this milestone was reached, InMobi migrated its data movement using Storage Transfer Service, establishing Cloud Storage as the single source of truth for the Bronze data layer. Finally, a centralized feature toggle seamlessly redirected all production Spark workloads to GKE.
Gaining unparalleled visibility and efficiency
To decouple downstream dependencies from the platform migration, InMobi continuously synchronized the Gold-layer data across both platforms, using checksum validation and optimistic locking to ensure data integrity and prevent table corruption. This ensured that end-consumer datasets remained consistent and available in both environments, enabling dependent teams to migrate independently at their own pace without disrupting production workloads.
By executing this carefully phased approach, InMobi successfully modernized its entire 2 PB/day platform with zero reported production outages or data integrity issues. This high-stakes transformation was heavily reinforced by Google PSO’s rigorous migration governance and operational health reviews, providing InMobi’s leadership with confidence and clear risk mitigation frameworks at every traffic-cutover gateway.
Google Cloud fundamentally changed our engineering culture. The stability and maturity of the enterprise-grade managed services, combined with Google PSO's validation of our migration strategy and production readiness, gave our leadership the confidence to execute this high-stakes modernization, resulting in a seamless transition with zero production outages.
Anindhya Sankhla
Lead Software Engineer, InMobi
Standardizing on Cloud Storage resolved previous I/O limitations, easily supporting 2 TB/s peak throughput for massive, parallel reads and writes. On the compute side, using workload-specific node pools on GKE has allowed InMobi to run over 95% of its workloads on highly efficient Spot VMs.
Most importantly, the engineering team now has unparalleled FinOps visibility. With access to pod-level Kubernetes metrics and Spark executor telemetry, InMobi can proactively optimize infrastructure. Using Requester Pays features, the company can now accurately attribute costs for shared datasets across business units.
Looking ahead, InMobi is building on this cloud-native foundation to explore Google AI capabilities, aiming to further accelerate engineering velocity through AI-assisted ETL development and predictive workload optimization.
InMobi is a global advertising technology platform that specializes in mobile in-app advertising and real-time bidding. Processing millions of ad requests per second, the company seamlessly connects advertisers with their target audiences in real time.
Industry: Technology
Location: India
Products: Google Kubernetes Engine (GKE), Spot VMs, Managed Service for Apache Airflow (formerly Cloud Composer), Cloud Storage, Storage Transfer Service, Cloud Monitoring
About Google Cloud partner – Ollion
Ollion is a global cloud-native consultancy accelerating digital transformation through cloud migration, security, and AI and data consulting services.
