Overview

ING Türkiye, a leader in digital banking innovation, embarked on a strategic modernization initiative to migrate monolithic applications to a Java-based microservices architecture. This transition, aimed at increasing agility and scalability, introduced significant performance challenges with their existing Oracle database infrastructure. By implementing Hazelcast Platform as a high-speed, centralized caching layer, ING Türkiye successfully eliminated system latency, enabled massive scalability for over 25 million daily transactions, and significantly enhanced its mobile banking user experience, all while achieving a lower total cost of ownership (TCO).

Business Challenge

As a digital-first institution, ING Türkiye faced rapidly increasing traffic to its mobile and online banking services. Their primary business challenge was to enhance the performance, scalability, and operational efficiency of their systems to support this growth. They needed to process over 25 million business transactions daily with minimal latency to maintain a competitive edge and deliver a superior, responsive customer experience. The goal was to build a future-proof infrastructure that could scale effortlessly with evolving customer demands and accelerate the delivery of new banking services.

Technical Challenge

The strategic migration from C# monolithic applications to 300 Java microservices deployed in Red Hat containers was designed to boost scalability. However, it created a new set of technical hurdles. The 55 development teams found that the new architecture funneled an overwhelming number of requests to the centralized Oracle database, which was used for authorization, session management, and fast data reads. This resulted in significant performance bottlenecks, including high database access latency and slow write-back operations. The Oracle database became the primary limiting factor, preventing the microservices from realizing their full potential for speed and independent scaling.

Technical Solution

To decouple their microservices from the database and resolve the performance issues, ING Türkiye's development teams benchmarked leading caching solutions, including Hazelcast and Redis. They selected Hazelcast Platform Enterprise Edition based on its superior performance at linear scale, ease of implementation, high-quality documentation, seamless integration with OpenShift, and a significantly lighter resource footprint. Hazelcast was implemented as a centralized, distributed in-memory computing and caching layer, providing the high-speed data access required by the 300 microservices. This allowed them to abstract away the Oracle database dependency for read-heavy operations and provide a resilient foundation for their applications.

“After benchmarking leading caching options—including Redis—we chose Hazelcast Enterprise for its superior performance at linear scale, developer experience, seamless OpenShift integration, and lighter footprint. We implemented Hazelcast as a centralized, distributed cache to give our 300 microservices high-speed data access and abstract read-heavy operations away from Oracle. With read-through caching and the CP Subsystem, we resolved our latency hotspots, guaranteeing the strict consistency our banking flows require—without custom code.”

— Efecan Ahmetoglu, Senior Software Design Engineer, ING Türkiye

Solution Architecture

Hazelcast was deployed on VMware virtual machines, creating a central, high-speed data fabric accessible to every microservice. This architecture immediately improved data access patterns and system responsiveness.

Key implementation patterns include:

• Stateful Session Management: To meet banking audit and traceability requirements, the system uses Hazelcast's distributed IMap to store and track user context. A unique UUID is generated at login and persisted in a Hazelcast map, allowing for seamless context tracking across the distributed environment until the user logs out.
• High-Consistency Locking: To manage tokens from external institutions that become invalid after a single use, ING Türkiye leverages Hazelcast's FencedLock. This feature of the CP Subsystem provides strict consistency, preventing race conditions and ensuring that critical operations are performed exactly once across different pods.¹
• Dynamic Configuration via Message Bus: To optimize resource use, the team utilized Hazelcast's distributed ITopic to implement the Spring Cloud Bus component. This allows for dynamic, runtime configuration changes—such as adjusting log levels or starting/stopping Kafka listeners—without requiring application restarts or deployments.²
• Database Offloading with Read-Through Caching: The read-through pattern was implemented to significantly reduce the load on the Oracle database. Microservices query Hazelcast directly for data; if the data is not in the cache, Hazelcast automatically fetches it from the database, effectively abstracting the database away from the applications.³ The team also plans to adopt the write-through pattern to further enhance efficiency.

Business Impact

The adoption of Hazelcast Platform delivered significant and measurable business value, reinforcing ING Türkiye’s position as a digital banking leader.

• Superior Customer Experience: By drastically reducing system latency, ING Türkiye now offers a consistently fast, reliable, and seamless mobile banking application for its millions of customers.
• Enhanced Scalability and Agility: The infrastructure now supports growing transaction volumes with ease and provides capacity for future growth. Microservices can be scaled independently without performance degradation, increasing business agility.
• Improved Operational Efficiency: All services are delivered within their SLAs. Development teams are empowered with a robust platform that simplifies data management and accelerates the development lifecycle.
• Lower Total Cost of Ownership (TCO): Hazelcast reduced the dependency on the expensive Oracle database infrastructure and enabled more streamlined resource utilization, leading to significant cost optimization.

Lessons Learned

1. Adopt an Incremental and Phased Approach:
   • Avoid "Big-Bang" Replacements: A "big-bang" replacement of core banking systems is often out of the question due to the high risk of disruption to mission-critical operations. Instead, an iterative, phased model is recommended, allowing for progress at each step while protecting existing assets.
2. Keep Security, Compliance, and Data Governance into consideration:
   • Regulatory Compliance: New regulations, such as ISO 20022 and PSD2, demand faster data processing, real-time reporting, and greater data retention.
3. Focus on Developer Experience and Innovation:
   • Accelerated Development Cycles: Modern databases and microservices architectures enable rapid iteration of applications, faster time-to-market for new products and services, and improved developer productivity.
4. Reduce Complexity and TCO:
   • Modern platforms aim to consolidate multiple tools and layers into a single, integrated platform, reducing architectural complexity, operational overhead, and Total Cost of Ownership.

Summary

By implementing Hazelcast Enterprise, ING Türkiye successfully resolved the critical performance and scalability challenges within its new microservices environment. The platform’s robust, high-performance caching capabilities have future-proofed their digital banking services, ensuring they can continue to innovate and meet the demands of an expanding customer base. Looking ahead, ING Türkiye plans to expand its use of Hazelcast by transitioning batch processes to event-driven systems and exploring active-active clusters for even greater redundancy, ensuring Hazelcast remains a central pillar of its technology strategy.

¹ Navigating Consistency in Distributed Systems: Choosing the Right Trade-Offs

² Dynamic Configuration: No Restarts Required

³ Database Caching - In-Memory Access to Frequently Used Data