loveholidays

loveholidays doubles site performance while cutting CPU costs with Google Axion

Google Cloud Results
  • 54% reduction in latency, increasing site conversions

  • 50-60% reduction in CPU core usage for faster performance

  • 60% cut in observability costs with no performance loss

  • 2x share of customers load site in under 1 second

  • 20 production apps migrated to Google Cloud Axion

Online travel agency loveholidays shifted key workloads to Axion for improved performance and reduced compute costs.

Fast, cheap, or good—or all of the above

For loveholidays, one of the most loved online travel agencies in Europe, every millisecond matters. Faster performance leads to happier customers, which leads to higher conversions. That’s why performance is embedded into the engineering culture of loveholidays, all with a clear goal in mind: delivering measurable business results through faster, more responsive systems.

The company began its cloud journey in 2018 by moving on-premises workloads to Google Kubernetes Engine (GKE) and was operating fully on Google Cloud within a year. That autoscaling investment helped reduce infrastructure costs by 70% in one month, and as travel demand surged back in 2023, GKE elasticity allowed loveholidays to focus on customer experiences over scaling issues.

When Google Cloud introduced Axion—its first custom Arm-based CPU—in 2024, loveholidays was quick to take interest. The early promise seemed too good to be true: high-quality, fast compute for lower cost?

We weren’t looking to fix anything, but what we were seeing from Axion made us curious. The benchmarks were incomparable, and the results were immediate.

Dmitri Lerko

Head of Engineering, loveholidays

That Google was standing behind their silicon – custom-designed for the cloud – and moving services like BigQuery and Spanner to Axion signaled to loveholidays that this was a bigger change than a new type of instance. It looked like the start of a platform shift.

“We weren’t looking to fix anything, but what we were seeing from Axion made us curious,” says Dmitri Lerko, Head of Engineering at loveholidays. With the support of Google product and engineering teams, loveholidays joined the private preview of Axion in August and began testing “The benchmarks were incomparable, and the results were immediate.”

One of the first applications migrated cut latency by a third, and CPU core usage dropped by as much as 60 percent. While performance was already industry leading, loveholidays couldn't miss out on this seismic shift to Arm-based GPUs. It migrated to Axion early, not to fix a problem but to future-proof its stack and simplify its scalability.

A faster experience leads to better conversions

loveholidays began its Axion migration with the applications that mattered most: those powering its search system. This system serves billions of real-time flight and hotel combinations to customers in milliseconds, and even slight improvements in latency can lead to noticeable jumps in conversion. By moving these workloads to Axion-backed C4A instances using GKE, the team saw page load times drop dramatically. One application alone cut its p99 latency by 54 percent.

We're already seeing latency improvements, cost reductions, and faster builds—and we’ve only migrated 60 percent to Axion. We expect even better results as we continue to complete our full migration.

Dmitri Lerko

Head of Engineering, loveholidays

The speed gains weren’t limited to customer-facing services. Prometheus, the monitoring system used by loveholidays to track performance, saw its CPU requirements drop from 22 to 10 cores—a 60 percent reduction in self-hosted observability costs without impacting performance. And CI/CD pipelines running on Axion have shown early signs of reducing build times from 16 minutes to 5 minutes, giving engineers faster feedback loops and more room to experiment.

Migration itself was planned for efficiency using the tools the team already trusted. loveholidays relied on Artifact Registry, Google Cloud’s container image storage service, to support a side-by-side rollout of x86 and Arm64 workloads.

Artifact Registry’s support for remote repositories and multi-architecture builds meant engineers could continue their regular release cadence without disruption. Engineers only needed a single line of Helm configurations to target Axion machines. 

“We're already seeing latency improvements, cost reductions, and faster builds—and we’ve only migrated 60 percent to Axion,” said Lerko. “We expect even better results as we continue to complete our full migration.” For customers, that means a more responsive search experience getting them from idea to booking faster.

loveholidays is now pushing toward full Axion adoption—including CI/CD pipelines—to simplify infrastructure and standardize Arm. The team is also exploring plans to run generative AI on top of Axion. The ultimate goal? Smarter systems, happier engineers, and a smoother journey for every traveler.

loveholidays delivers simple, affordable online holiday bookings to customers in the UK and beyond.

Industries: Travel and Hospitality

Location: United Kingdom

Products: Artifact Registry, BigQuery, Google Cloud Axion, Google Kubernetes Engine, Spanner

Google Cloud