Skip to content

From NLB Sprawl to Mesh Efficiency: How Skyscanner Handles 60M Requests Per Minute With Istio#

KubeCon + CloudNativeCon Europe 2026 digest Program Core pick 23–26 March 2026

Abstract Recording

John Clark, Skyscanner & Steven Thwaites, Solo.io

Play the recording: From NLB Sprawl to Mesh Efficiency: How Skyscanner Handles 60M Requests Per Minute With Istio

  • Why: The other cost-with-numbers talk, from the opposite direction: removing load balancers rather than routing around them. Also a rare case study that says what it gave up, not just what it saved.
  • What: A network rebuild across dozens of clusters at >60M req/min, saving "millions" on AWS while preserving resilience and compliance.
  • How: Four changes: east-west NLBs removed in favour of pod-to-pod multicluster traffic; gateways added for controlled ingress/egress while using Spot instances; Istio Ambient Mesh deployed for zero-trust without sidecars; OpenTelemetry-based observability built alongside. The abstract is unusually candid about the risks — NLB sprawl, multicluster operational drag, sidecar overhead, coordinated rollout across many clusters, and holding peak-time performance while shifting traffic.
  • Where applicable: AWS-specific in its cost model (NLB pricing is what makes the arithmetic work), but the ambient-mesh and multicluster patterns are general. Most relevant if you are running many clusters with east-west load balancers between them.