Clipper 8
$ 49 /month
- vCPU
- 4 × dedicated
- RAM
- 8 GB
- Storage
- 200 GB NVMe Gen4 SSD
- Transfer
- 20 TB
- IPv4 / IPv6
- 1 / /64 routed
Solution
VPS for low-latency API and real-time backends
Clipper 8 · $49/moClipper 16 · $92/mo
Short answer
Sub-10 ms API responses are a placement problem before they are a code problem. Physics puts a floor of about 1 ms per 100 km of fibre, so no amount of optimisation beats being in the wrong city. Deploy Clipper instances in the regions your users are in, keep the hot path free of cross-region calls, and measure at the 99th percentile rather than the mean.
What you need
| Resource | What you actually need |
|---|---|
| CPU | High-frequency dedicated cores; tail latency is a scheduling problem |
| RAM | 8–32 GB depending on working set |
| Disk | NVMe Gen4 |
| Network | Anycast-capable regions for a single global entry point |
Recommended plans
Clipper 16
$ 92 /month
- vCPU
- 8 × dedicated
- RAM
- 16 GB
- Storage
- 400 GB NVMe Gen4 SSD
- Transfer
- 30 TB
- IPv4 / IPv6
- 1 / /64 routed
Clipper 32
$ 174 /month
- vCPU
- 12 × dedicated
- RAM
- 32 GB
- Storage
- 800 GB NVMe Gen4 SSD
- Transfer
- 40 TB
- IPv4 / IPv6
- 1 / /64 routed
Recommended locations
Location is usually the decision that matters most for this workload — either because latency dominates, or because jurisdiction does.
Why OnionVPS for this
- Dedicated high-frequency cores eliminate the steal time that creates p99 outliers.
- Anycast-capable regions let one address route to the nearest healthy instance.
- Locations on six continents, because placement is the largest single lever you have.
- Well-peered core regions on the major exchanges, so paths are short as well as fast.
How to set it up
-
Map where your requests originate
Then deploy to those regions, not to the cheapest one.
-
Deploy Clipper instances per region
High-frequency cores, and pin the process.
-
Keep the hot path region-local
One cross-region database call erases every other optimisation.
-
Enable HTTP/3 and keep connections alive
Connection setup dominates short-request latency.
-
Measure p99, not the average
Averages hide exactly the requests your users complain about.
Frequently asked questions
What is realistic API latency from a VPS?
Under 1 ms of server time is achievable for simple handlers on dedicated cores. Total user-perceived latency is dominated by distance: roughly 1 ms per 100 km each way, so placement decides the outcome.
Does anycast help API latency?
Yes, by routing each user to the nearest healthy instance without DNS propagation delay. It requires instances in several regions and is available in our anycast-capable locations.
Why is my p99 much worse than my average?
Usually CPU steal time on shared cores, or garbage collection. Dedicated cores remove the first; the second is a code problem.