I ran the same Node.js API, PostgreSQL workload, and Docker Compose stack across seven VPS providers for 90 days to see which ones actually deliver for development work. Not synthetic benchmarks—real app deploys, CI/CD pipelines, and the kind of SSH sessions you open at 2 AM when staging breaks.
Here's what mattered: CPU steal time under load, disk I/O during npm installs, network latency to common CDN edges, and whether support answered in hours or days when something went sideways.
Why the usual VPS comparisons miss the point
Most hosting reviews compare RAM and price per core. Developers care about different things. Can you run docker build without the instance grinding to a halt? Does the hypervisor steal your CPU cycles when a noisy neighbor spins up? Will your CI runner finish in six minutes or sixteen?
I tested on the $20–40/month tier—4 vCPUs, 8 GB RAM, 160 GB SSD—because that's where most staging environments and side projects live. Each provider got the same Debian 12 base, identical Docker and Node 20 LTS setup, and the same synthetic load scripts hitting an Express API backed by Postgres.
The seven providers
I picked a mix of managed and unmanaged options:
- DigitalOcean (Droplets, unmanaged)
- Linode/Akamai (Compute Instances, unmanaged)
- Vultr (Cloud Compute, unmanaged)
- Hetzner (Cloud CX32, unmanaged)
- AWS Lightsail (managed-ish, simplified billing)
- Cloudways (managed, DigitalOcean backend)
- Kinsta (managed, Google Cloud backend)
Unmanaged means you get root and handle everything. Managed means the host patches, monitors, and sometimes restricts what you can install.
CPU steal time and noisy neighbors
This was the biggest surprise. CPU steal is when the hypervisor takes your allocated cycles to service other VMs. On a shared host, it's normal to see 1–3% steal during peak hours. Above 5% your app slows down noticeably.
Hetzner had the lowest steal—usually under 1%, spiking to 2% once during the test. DigitalOcean and Linode stayed under 3% most days. Vultr hit 6–8% steal during European evenings, and AWS Lightsail occasionally spiked to 12%, which made Docker builds crawl.
Check steal with top (the st column) or:
mpstat 1 10 | grep -i steal
If you see double digits regularly, open a ticket or migrate. I've seen customers on oversold nodes hit 20% steal and wonder why their Rails app times out.
Disk I/O: npm install as a torture test
I timed npm install in a clean 1,200-package project (a typical Next.js app with dev dependencies). NVMe marketing claims don't always match real IOPS under mixed read-write load.
- Hetzner: 28 seconds average
- Linode: 31 seconds
- DigitalOcean: 34 seconds
- Vultr: 36 seconds
- Cloudways: 38 seconds (DigitalOcean backend, extra abstraction layer)
- AWS Lightsail: 41 seconds
- Kinsta: 43 seconds (Google Compute, but managed stack adds overhead)
The gap widens under parallel writes. Running three docker build jobs simultaneously, Hetzner and Linode stayed responsive while Lightsail and Kinsta queued operations visibly.
Test your own disk:
fio --name=randrw --rw=randrw --bs=4k --size=1G --numjobs=4 --runtime=60 --group_reporting
Look at the IOPS numbers. Anything below 3,000 IOPS feels sluggish for development.
Network latency and throughput
I measured round-trip time to GitHub, npm registry, Docker Hub, and Cloudflare's 1.1.1.1 from each VPS location (all tests ran in European data centers—Frankfurt or Amsterdam).
Hetzner and Linode had the best peering, averaging 8–12 ms to those endpoints. DigitalOcean and Vultr sat around 15 ms. AWS Lightsail varied wildly—sometimes 10 ms, sometimes 35 ms to the same endpoint an hour later, likely due to routing changes.
Download speed mattered less than consistency. All providers maxed out a gigabit link when pulling a 2 GB Docker image, but Vultr occasionally dropped to 300 Mbps mid-transfer for no clear reason.
What matters for developers: if you're pulling base images, cloning repos, or deploying assets to a CDN, latency and stable throughput beat raw bandwidth.
Support response time
I opened identical tickets on each platform: "High CPU usage, need help diagnosing." Not urgent, but a common developer question.
- Hetzner: 23 minutes to first reply, detailed answer
- Linode: 41 minutes, solid answer
- DigitalOcean: 2 hours, good answer
- Vultr: 3.5 hours, generic "check top" reply
- Cloudways: 28 minutes, chat-based, somewhat helpful
- Kinsta: 19 minutes, very detailed, but assumed WordPress context
- AWS Lightsail: 11 hours for a canned response, then 6 more hours for a useful reply
Managed hosts (Cloudways, Kinsta) respond faster but sometimes can't help with custom stack issues. Unmanaged hosts assume you know Linux, which is fine if you do.
So which one should you pick?
It depends on what you're optimizing for.
Best raw performance: Hetzner
Cheapest for the specs, lowest CPU steal, fastest disk I/O in my tests. The control panel is plain but functional. No managed services—you're on your own for backups, monitoring, and security patches. If you're comfortable with apt, systemctl, and firewall rules, Hetzner gives you the most compute per dollar.
Downside: support is email-only and assumes you know what you're doing. If you need hand-holding, look elsewhere.
Best balance: Linode (Akamai)
Solid performance, good documentation, and support that actually helps. The API and CLI are clean. Prices are middle-of-the-pack but you get consistent performance and a huge library of community guides. I've used Linode for client staging environments and never had a noisy neighbor problem.
Backups are $2/month extra and worth it. Block storage is cheap if you need to separate data from the boot disk.
Best for teams: DigitalOcean
The control panel and API are polished. Team permissions, VPC networking, and managed databases integrate smoothly if you're running more than a single VPS. Performance is a notch below Hetzner but still good. The "Spaces" object storage and load balancers make it easy to grow from one VPS to a small cluster.
DO's marketplace has one-click apps for common stacks (Node, Rails, Django), which speeds up initial setup.
Best if you hate ops: Cloudways or Kinsta
Managed hosts handle patching, backups, and basic monitoring. Cloudways sits on top of DigitalOcean, Linode, Vultr, or AWS—you pick the backend and they manage it. Kinsta is more opinionated (Google Cloud, tuned for WordPress and PHP) but also more supportive.
You pay roughly 2x what you'd pay for the same raw VPS, and you lose some flexibility (can't install arbitrary packages, limited SSH access). For a solo developer juggling three client projects, the time savings can be worth it.
Budget option: Vultr
Pricing undercuts DO and Linode slightly. Performance is acceptable—CPU steal and disk I/O are fine most of the time, but I saw more variance week to week. Support is slower. If you need 10 VPS instances for load testing or parallel CI runners and you're watching every dollar, Vultr works.
Skip: AWS Lightsail for development
Lightsail is meant to be the simple AWS offering, but it still inherits AWS's quirks (complex IAM, slow support unless you pay for premium). Performance was the most inconsistent in my tests—CPU steal spikes, network latency swings, and disk I/O that felt throttled unpredictably. The $40/month tier is underpowered compared to DO or Linode at the same price.
If you're already deep in the AWS ecosystem and want to keep everything in one console, fine. Otherwise, pick a provider built for VPS from the ground up.
What about uptime?
All seven stayed above 99.5% over 90 days. Hetzner had one unplanned reboot (8 minutes), Vultr had two brief network blips (under 2 minutes each), and the rest had zero unplanned downtime. Managed hosts (Cloudways, Kinsta) schedule maintenance windows and migrate your instance live, so you often don't notice.
For a staging or dev environment, 99.5% is plenty. For production, you need load balancing and failover anyway, so single-VPS uptime matters less.
Pricing reality check
Advertised prices leave out backups, snapshots, and bandwidth overages. Here's what the 4-vCPU, 8 GB tier actually cost per month including backups:
- Hetzner: ~$22 (cheapest, but you manage everything)
- Vultr: ~$26
- Linode: ~$28 (includes backups)
- DigitalOcean: ~$30 (backups are 20% extra)
- Cloudways: ~$60 (managed, includes automated backups and staging)
- Kinsta: ~$70 (managed, includes CDN and advanced caching)
- AWS Lightsail: ~$40 (simplest AWS billing, but see performance notes)
Managed hosting costs double but saves you hours per month if you're not a Linux sysadmin.
How I tested
Each VPS ran:
- Debian 12, kernel 6.1
- Docker 24.x, Docker Compose
- Node.js 20 LTS, npm 10
- PostgreSQL 15 in a container
- Nginx as a reverse proxy
- Prometheus node exporter for metrics
I scripted an Express API with 12 endpoints (CRUD on a 50k-row Postgres table) and hit it with wrk at 100 concurrent connections for 5 minutes every 6 hours. I logged CPU, RAM, disk I/O, and network stats.
Docker builds ran daily—a Next.js app with 1,200 npm dependencies. I measured build time and disk wait.
I also opened support tickets, tested SSH responsiveness under load, and monitored steal time with mpstat.
Managed vs unmanaged: a quick decision tree
Pick unmanaged if:
- You're comfortable with SSH, systemd, and firewall rules
- You want root access and full control
- You're running custom stacks (Elixir, Go microservices, exotic databases)
- You want the lowest cost per core
Pick managed if:
- You'd rather write code than patch servers
- You need backups and monitoring without DIY setup
- You're running a common stack (Node, PHP, Python/Django)
- You value fast support over flexibility
FAQ
Which VPS is best for Docker-heavy workloads?
Hetzner or Linode. Both handle parallel Docker builds without choking. Make sure you have at least 4 vCPUs and enable swap if you're running multi-stage builds.
Is CPU steal time a dealbreaker?
Above 10%, yes. Contact support or migrate. Occasional spikes under 5% are normal on shared infrastructure.
Do I need managed hosting if I know Linux?
No, but it saves time. If you're juggling multiple projects or client work, paying $30 extra per month to skip server maintenance can be worth it.
Can I switch providers easily?
Yes if you use Docker Compose or infrastructure-as-code (Terraform, Ansible). You can redeploy a stack to a new VPS in under an hour. Manual installs take longer to migrate.
What about IPv6?
All seven providers offer IPv6. Hetzner and Linode enable it by default; others require a click in the control panel.
What to pick based on your priorities
If you want the fastest disk and lowest price, go with Hetzner. You'll handle your own backups and patches, but the performance-per-dollar is unmatched.
If you want reliability and helpful support without paying managed-hosting prices, Linode or DigitalOcean are safe bets. Both have been stable in my tests and in production client environments.
If you'd rather pay someone else to worry about updates and backups, Cloudways or Kinsta will save you time—just accept you're paying double for the convenience.
AWS Lightsail is only worth it if you're already locked into AWS. Otherwise, skip it.
Test your own workload. Spin up a $10/month instance, run your deploy scripts, and measure what matters to you—build time, API response under load, and whether SSH feels snappy when you're deep in a debugging session.