GoPay Optimizes Payment Processing Performance

GoPay improved transaction throughput on existing instances without adding costs and scaled capacity using newer instances.

At a glance:

  • GoPay, a part of GoTo Financial, is one of the subsidiaries of GoTo Group, the largest digital ecosystem in Indonesia.

  • GoPay turned to Google Cloud and the Intel Software Center of Excellence (CoE) to explore performance improvement opportunities; the recommended optimizations resulted in immediate improvement in CPU utilization and increased transaction capacity.



Executive Summary 

GoPay, a part of GoTo Financial, is one of the subsidiaries of GoTo Group, the largest digital ecosystem in Indonesia. GoTo uses the Barito logging platform running on Google Cloud to monitor and log all transactions. The Barito platform presented a bottleneck for scaling up transactions for GoPay. The company turned to Google Cloud and the Intel® Software Center of Excellence (CoE) to explore performance improvement opportunities for the platform. 

Developers in the Software CoE recommended optimizations for Barito that resulted in an immediate improvement in Barito CPU utilization and an increase in transaction capacity. The recommendations also included benchmarking Barito on N2 instances with 3rd Gen Intel® Xeon® processors to understand how capacity could be further improved. These optimizations allowed GoTo to increase throughput without adding cluster costs on existing N1 instances and to significantly scale capacity using newer Google Cloud N2 instances. 

Optimizations for the Barito logging platform deliver up to 15% higher throughput on Google Cloud with 2nd Gen Intel® Xeon® processors and up to 87% performance improvement with 3rd Gen Intel® Xeon® processor-based instances.1


GoPay is an e-wallet based in Indonesia. GoPay, a part of GoTo Financial, is one of the subsidiaries of GoTo Group, the largest digital ecosystem in Indonesia. Using the GoPay or Gojek app available on Apple and Android devices, users can make transactions with any merchants that accept GoPay as a payment method.

An intelligent logging platform called Barito (Figure 1) logs transactions across all GoTo’s properties. Barito is critical to maintaining strong and efficient operations across GoTo’s services. 

Figure 1. The Barito logging system logs activities across all GoTo properties including GoPay.

Figure 2. The Barito logging platform integrates several applications and tools.

The Barito cluster integrates several tools and applications, including Apache Kafka, Kibana, and other tools to produce, track, search, and visualize all transactions as they happen (Figure 2). GoTo’s IT and customer support teams can access the logs to monitor services and troubleshoot issues for the company and customers. Barito runs on clusters of N1 instances on Google Cloud using Intel® Xeon® processors.

With thousands of transactions coming every minute, the need to get better visibility and monitoring real-time was crucial. GoPay needed more processing or CPU headroom to process more transactions. Rather than scale instances, which would lead to higher costs, GoTo engineers turned to Google Cloud and the Intel® Software Center of Excellence (CoE) to help isolate issues and offer solutions to improve performance.


The Software Center of Excellence is a service of Intel in collaboration with Google Cloud. The CoE helps customers running Google Cloud instances based on Intel® architecture and technologies to optimize performance of their workloads.

“We’re excited about this collaboration with the Intel CoE program. In the short term, we’re able to identify and improve CPU utilization by up to 10 percent for one of our components. We are looking forward to working closely so we can achieve more optimal utilization and improvement in the future.” — Eka Risky, Principal Engineer, Payments Infrastructure GoTo Financial

Intel Software Center of Excellence (CoE) Overview

Intel Software CoE’s engineers are experts in Intel optimization tools and the technologies and capabilities of Intel processors, chip sets, and other Intel products. Working with Google, these developers have helped Google Cloud customers improve performance by modifying code to optimize for parallelism, reduce time spent in tasks, or take advantage of Intel technologies, such as the built-in Intel® Accelerator Engines integrated in all 4th Gen Intel® Xeon® processors. The CoE services enhance the value customers get from their Intel Xeon processor-based instances. As a Software CoE customer, users get:

  • Direct access to Intel engineers
  • Guidance for improving the price performance and the operational performance of Intel-based Google Cloud instances
  • Code-level recommendations for software running on Intel processors so customers can attain the most benefit possible from their Google Cloud investments

An engagement with the Software CoE includes a Performance Report with optimization recommendations, an action plan, and an implementation plan for customers to execute. Recommendations for potential support from Google Cloud Professional Services Organization (PSO) can also be included.

Figure 3. The Intel Software COE analyzed Barito flame graphs to uncover performance enhancement recommendations.

Barito Analysis

Developers in the CoE analyzed Barito flame graphs provided by GoPay (Figure 3) to understand where latencies occurred during the execution of services. Developers discovered that the Prometheus data scraper feeding Kafka consumed approximately 22 percent of available CPU cycles. This seemed high for the workload. 

The Software CoE Performance Report recommended two actions:

  1. Decrease the Prometheus interval by half, which would reduce CPU utilization for Prometheus.
  2. Test Barito on Google Cloud N2 instances with 3rd Gen Intel Xeon processors.

For testing Barito performance between N2 and N1 instances, the Barito software stack was deployed on Google Kubernetes Engine (GKE) clusters with the client running on N1 instances. Locust was used to drive traffic to the GKE clusters with a varying load to study the characteristics and response of the workload. The usercount was set to a rate of 5000 at 40 users per second. The length of the logs was set at 1100, and Wikipedia was the text source for benchmarking.


For the optimizations on N1 instances, by cutting the Prometheus scraping interval by half, Kafka CPU utilization dropped significantly by 10 percent.2 Throughput of Barito logging increased by up to 15 percent,2 allowing more transactions to occur.

For the N1 vs. N2 benchmarks, the higher performance 3rd Gen Intel Xeon processor-based instances enabled the following improvements:

  • 20 percent overall reduction in CPU usage
  • Fewer nodes needed—from 186 N1 to 150 n2- standard-32 nodes

With N2 instances, Intel showed 87 percent more responses per second compared to N1—nearly twice the response rate of N1 (Figure 4). This allowed 50 percent more users to be added with requests being processed nearly 10 percent faster, while using fewer Google Cloud resources.3

Figure 4. The higher performance 3rd Gen Intel Xeon processor-based instance had nearly twice the response rate compared to the 2nd Gen Intel Xeon processor-based instance.

These optimizations and migrating to N2 instances with a smaller overall configuration due to needing fewer nodes allowed GoTo to improve Barito logging throughput cut costs about 9.8 percent.1

The benchmarks resulted in low CPU utilization on both worker nodes of the GKE cluster, suggesting GoPay could potentially reduce spending by moving to lower core count Google Cloud instances on the worker nodes, while maintaining transaction quality to customers. Intel also recommended GoPay to upgrade Java, Golang, and Ruby to the latest versions to get the best performance from Intel architecture for their Barito software stack. 

Solution Summary

GoPay transaction throughput suffered due to the Barito logging system. Barito runs on Google Cloud N1 instance clusters. Working with the Intel Software Center of Excellence, GoPay implemented the optimization recommendations of the Software CoE developers, resulting in an immediate significant improvement in CPU utilization. The optimizations allowed GoPay to improve transaction throughput without upgrading their cluster instances or increasing vCPU count. 

Software CoE developers also recommended benchmarking Barito on N2 instances with 3rd Gen Intel Xeon processors. The benchmark revealed the N2 instances could support up to 50 percent more users with 87 percent more responses per second, while cutting average response time by nearly 10 percent.2 These improvements could all be achieved on 36 fewer instances, potentially reducing costs for running Barito.

Solution Ingredients

  • Google Cloud N2 instances with 3rd Gen Intel Xeon processors
  • Intel optimizations for Barito logging platform 
  • Optimizations by Intel Software Center of Excellence