Secret to saving 32% per month with infrastructure optimization

About the company

Our customer is an international company with over 20 years of experience providing digital solutions for the construction and infrastructure sectors. The company specializes in software that streamlines fieldwork, compliance, and asset management for civil engineering projects to achieve high operational efficiency. 

The challenge: increasing infrastructure costs

The company’s infrastructure costs kept rising despite its stable customer base and consistent revenue. 

This was partly caused by its migration to the Google Cloud Platform (GCP). The solutions the company selected were initially effective but not optimal for future use cases. Given the application’s stage and limited resources, the team had to make pragmatic decisions to ensure it stayed operational, even if that meant relying on solutions like NFS volumes for file storage. Due to competing priorities and time constraints, there was limited opportunity to rework specific systems. As a result, infrastructure costs started getting out of hand.

Optimization process

Here’s a breakdown of the steps our team has taken and their financial impact:

Replacing the long-term storage solution

The most significant cost reduction stemmed from migrating “Dossier” files from the Filestore (managed NFS solution) to Google Cloud Storage. The change involved eliminating the rapidly growing Filestore instance of 21.5TiB, with zonal availability and 14 days of backups, and replacing it with regionalized, secure components. 

The result: Storage and backup costs dropped by over $4500 per month while this part of the infrastructure’s performance, durability, and resilience increased.

Reducing the number of GKE nodes by optimizing the app’s configuration

The application received new features that contributed to an increase in resource requirements. Over time, we found ourselves far beyond the Committed Use Discounts frames purchased not long after the migration.

• Optimizing the application’s core configuration, including the Apache web server and PHP interpreter. By fine-tuning their settings, we significantly improved resource efficiency, particularly regarding RAM usage.
• Streamlining the process of starting the application in containers by reducing the number of stages required. This change made running cron jobs much less resource-intensive.

Together, these optimizations led to a substantial reduction in computing costs. The first step allowed us to revert to using resources covered by Committed Use Discounts, while the second reduced resource consumption even further.

The result: Since the client had multiple projects in GCP that didn’t rely on fully committed resources, we were able to reallocate computing power, resulting in savings of approximately $1,300 per month.

Optimizing GPU Instance Usage

An image-recognition tool was introduced as part of the client’s offering to expand the application’s functionality. Based on available observability metrics, The Software House determined that the GPU processing power provided was largely underutilized.

By using GPU time-sharing functionality for Kubernetes nodes, we allowed up to 4 pods to share a GPU, significantly reducing the cost overhead at this point and once the solution requires scaling.

The result: Cost savings of $240 per month.

Reducing Log Volumes in GCP Logs

We have stopped gathering logs already aggregated by the application monitoring solution on the Kubernetes cluster. The redundancy was necessary during the application’s migration effort. Due to the number of environments and traffic size, the flow of logs proved a challenge for a logging implementation provided at an early stage.

The result: While data is limited, this is estimated to save $150-$250 per month.

Other money-saving improvements

While working on the project, we noticed additional areas to reduce infrastructure costs further. If the customer can pay even less, then why not work on it?

• Consolidating non-production database instances

Moving non-production databases into one instance would allow us to improve resource utilization and save $114 per month. 

• Using self-hosted NFS for non-production environments

Switching from Filestore to a self-hosted NFS could save $100-$150 per month.

• Using a more efficient compression mechanism for OpenSearch indices

Internal tests indicated that using the `zstd_no_dict` compression method in place of the default LZ4 could result in a 60% smaller index size at a negligible performance cost. That would allow us to save up to $300 per month.

Total savings

Infrastructure optimization was another initiative in our three-year cooperation with. The effort began in July 2024. By January 2025, The Software House had already achieved monthly savings of $7500, with further reductions expected after planned optimizations are implemented.

The infrastructure bill for June 2024 was $23,500…

…and in January 2025, the infrastructure bill was lowered to just $15,800. Making a total yearly savings of around $90,000, with more to come.

Future cost optimization plans

Even though most of the client’s infrastructure resides on the Google Cloud Platform, the next initiative focuses on a service delivered by another cloud provider: AWS OpenSearch. 

The plan includes migrating the existing cluster to Google Kubernetes Engine and implementing a more efficient data compression method. While the former may cause readers to scratch their heads due to potential maintenance costs, the effort of managing such a cluster is largely mitigated by employing /Kubernetes OpenSearch Operator/.

Optimization is an ongoing process, and we have identified more opportunities to reduce the costs of customer’s infrastructure and provide even more efficient and reliable solutions to support the company’s growth and performance.