“Lithus made our setup much simpler, and it’s now far easier for our team to get things done. Moving off the big US cloud platforms was a bonus, and our costs haven’t changed in two years.”
Toby Hammond, CEO, Futurepump
The Situation
Futurepump manufactures solar-powered irrigation pumps for smallholder farmers across developing countries, and tens of thousands of their pumps in the field stream continuous telemetry data back to a central pipeline for remote monitoring, warranty diagnostics, and fleet analytics.
Their ingestion and analytics platform was running on Google Cloud, where bills were growing by roughly 20% per year with frequent cost spikes as time-series data accumulated. Because Futurepump is a hardware and product company rather than an infrastructure shop, nobody on the team had the specialist cloud engineering time to diagnose why queries were slowing down or why monthly spend kept climbing.
What We Built
MQTT ingestion and high-availability TimescaleDB on local NVMe
We migrated Futurepump’s entire telemetry pipeline from Google Cloud to a multi-node European bare-metal cluster running our managed Kubernetes stack (RKE2 and Cilium). Incoming pump telemetry arrives through an MQTT broker, passes through Python and Django processing workers, and is stored in a high-availability PostgreSQL and TimescaleDB cluster managed by StackGres.
Time-series chunk scans and continuous aggregations across years of pump telemetry are heavily bound by disk I/O and memory. On Google Cloud, shared vCPUs and network-attached persistent disks throttled IOPS as the dataset grew. By allocating dedicated memory and running StackGres directly on local enterprise NVMe storage (OpenEBS ZFS-LocalPV and Mayastor NVMe-oF), we eliminated network-disk I/O wait and cut dashboard and analytics query latency in half, from 600ms down to 300ms, as soon as traffic switched over.
Pipeline observability and EU sovereignty
We deployed our full observability stack (Prometheus, Loki, and Grafana) alongside their Metabase analytics environment so Futurepump’s engineers can see queue depth, MQTT ingestion rates, and database performance in real time. Moving the platform from a US hyperscaler to ISO 27001-certified European data centres also simplified their data governance while locking in a flat monthly price that has not increased in over two years.
Before & After
| Area | Before (Google Cloud) | After (Lithus Bare Metal) |
|---|---|---|
| Dashboard & query latency | ~600ms average query time on network-attached cloud disks | ~300ms (50% faster) on local enterprise NVMe storage |
| Monthly cost trajectory | Growing by ~20% per year with unpredictable billing spikes | Flat, predictable monthly rate with 0% increase over two years |
| Time-series database | Cloud-hosted database throttled by shared IOPS limits | HA StackGres PostgreSQL + TimescaleDB with dedicated RAM and NVMe |
| Pipeline visibility | Limited visibility into MQTT ingestion bottlenecks or slow queries | Full Prometheus, Loki, and Grafana observability with 24/7 engineering alerting |
| Data jurisdiction | US hyperscaler cloud platform | ISO 27001-certified European data centres with Lithus engineers on-call |
More case studies
We migrated XDI Systems from a $25,000 per month AWS setup to a 10-node bare-metal cluster with 200 Gbps storage and custom KubeVirt tooling, cutting costs by 45%.
Read case study →PrepBusiness: SaaS LogisticsWe traced recurring 2-minute database lockups to their source, doubled hardware resources at the same monthly cost, and later consolidated their DigitalOcean cluster onto the same hardware.
Read case study →Talk directly with an engineer who runs production clusters.
- 1Book a 30-minute call with one of our engineers to walk through your current stack
- 2Optionally share a recent cloud bill so we can model your exact hardware and savings
- 3Receive a costed migration plan, a fixed monthly price, and a candid view on whether we are a good fit
Locked into a cloud contract right now? Ask us to check back in 3, 6, or 12 months →