Install
Install

How big a server do I need?

One small server runs it. These are load-tested numbers, and anything not yet measured says so.

Measured before the latest performance work; re-run pending.

Recommended Measured

100,000 issues

tested on 2 vCPU, 4 GB RAM, one server from the one-command install

issues: fast at 80 requests a second
31,384
active at once, teams up to ~2,000 Estimate
≈400
a month, from Hetzner
€5.49

Speed at 100,000 issues

95% of requests, server side
  • Open an issue < 75 ms
  • Open a doc page < 75 ms
  • Search 205 ms
  • Page tree 237 ms
  • Big board 383 ms

A full bar is half a second. Issues and doc pages open fastest; search and the biggest board are the slowest parts at this size.

What 2 vCPU, 4 GB costs

list price a month, before tax, checked 2026-10-12

The people figure is an estimate: it assumes each active person makes 0.2 requests a second, against the 80 a second measured with 31,384 issues, and that 20% of a team is active at once, rounded down. There is no people figure for 100,000 issues, because no load ramp was run at that size.

  • 2 vCPU, 2 GB Not yet measured

    No numbers here until this size has been measured.

  • 2 vCPU, 4 GB Measured

    Up to about 100,000 issues. At about 274,000 it slowed to about 2 seconds.

  • 4 vCPU, 8 GB Not yet measured

    No numbers here until this size has been measured.

The Platform team’s sprint board in the live demo: one sidebar for Work and Docs, columns by status with counts and a WIP limit, and epics as swimlanes.
The board in the live demo. The measured data set is not pictured: this is the demo’s own data.

Estimate for your data

Move the sliders to roughly how many issues and doc pages you have. Comments grow with issues in our tests.

How we measured

One server from the one-command install, with Postgres tuned by the installer. Every run below is on 2 vCPU, 4 GB, and the raw figures are in the committed results file.

31,384 issues: the load ramp

  • Data: 31,384 issues (largest project 10,635), 45,894 comments, 400 doc pages, 245 users, 5 projects.
  • Typical speed at 30 requests a second: 95% of requests answered within 25 ms for a board, 10 ms for an issue, 22 ms for search, 10 ms for a doc page and 16 ms for an issue edit.
  • At the limit, 80 requests a second: 95% within 68 ms for a board and 32 ms for an issue.
  • A 45-minute run at 30 requests a second: 98,590 requests, 0 errors, 95% within 24–27 ms, app memory about 500 MB.
  • Live doc editing delivered 66,503 of 66,504 keystrokes. An Import from Jira of 10,000 issues took 20 seconds.

Larger data sets

  • About 100,000 issues : usable. Server-side p95 per route is on the card above. No load ramp was run, so there is no people figure. Measured before the latest performance work; re-run pending.
  • About 274,000 issues , 527,000 comments and 29,000 doc pages : slow. About 2 seconds for almost everything: the 2 GB database no longer fits in memory. No load ramp was run, so there is no people figure. Measured before the latest performance work; re-run pending.

The people estimate. An active person does about one thing every 5 seconds (0.2 requests a second), and about 20% of an organisation is active at the same moment. Both are assumptions, not measurements, and the people figures are rounded down. So 80 requests a second is about 400 people active at once, or a team of up to about 2,000 (estimate).

Prices by provider

On-demand list prices for about 730 hours a month, before tax, in the region named. Storage beyond the included disk, backups and outbound traffic are extra. Prices are shown in the currency each provider bills in, never converted. Bemmoly itself is free: no licence or per-seat fee. No affiliate links.

  • OVHcloud US site, USD
    2 vCPU, 2 GB
    no plan
    2 vCPU, 4 GB
    $4.54 VPS-1, introductory
    4 vCPU, 8 GB
    $8.50 VPS-2

    us.ovhcloud.com , checked 2026-10-12

  • Hetzner Germany and Finland, EUR
    2 vCPU, 2 GB
    no plan
    2 vCPU, 4 GB
    €5.49 CX23
    4 vCPU, 8 GB
    not published CX33, not verified

    docs.hetzner.com , checked 2026-10-12

  • Vultr Most regions, USD
    2 vCPU, 2 GB
    $15 Cloud Compute Regular
    2 vCPU, 4 GB
    $20 Cloud Compute Regular
    4 vCPU, 8 GB
    $40 Cloud Compute Regular

    vultr.com , checked 2026-10-12

  • DigitalOcean All regions, USD
    2 vCPU, 2 GB
    $18 Basic Droplet
    2 vCPU, 4 GB
    $24 Basic Droplet
    4 vCPU, 8 GB
    $48 Basic Droplet

    digitalocean.com , checked 2026-10-12

  • Akamai (Linode) Most regions, USD
    2 vCPU, 2 GB
    $12 Shared 2 GB, 1 vCPU, 2 GB
    2 vCPU, 4 GB
    $24 Shared 4 GB
    4 vCPU, 8 GB
    $48 Shared 8 GB

    akamai.com , checked 2026-10-12

  • AWS Lightsail US East, USD
    2 vCPU, 2 GB
    $12 2 GB bundle
    2 vCPU, 4 GB
    $24 4 GB bundle
    4 vCPU, 8 GB
    $44 8 GB bundle, 2 vCPU, 8 GB

    aws.amazon.com , checked 2026-10-12

  • Google Cloud us-central1, USD
    2 vCPU, 2 GB
    $12.23 e2-small, not verified
    2 vCPU, 4 GB
    $24.46 e2-medium, not verified
    4 vCPU, 8 GB
    not published e2 custom, not verified

    cloud.google.com , checked 2026-10-12

  • Azure East US, Linux, USD
    2 vCPU, 2 GB
    no plan
    2 vCPU, 4 GB
    $27.45 B2als v2
    4 vCPU, 8 GB
    $97.09 B4als v2

    prices.azure.com , checked 2026-10-12

Caveats

  • OVHcloud: Starting-from prices that may need a 12-month commitment; tax treatment not stated. Bandwidth unmetered but speed-limited.
  • Hetzner: Before 19% German VAT. Prices for new orders rose on 2026-06-15. CX plans showed as currently unavailable when checked. No 2 GB plan: CX23 is the smallest.
  • Vultr: From the public plans list. A few regions cost more.
  • DigitalOcean: Regular SSD. IPv4 and 3 to 5 TB of transfer included.
  • Akamai (Linode): No 2 vCPU, 2 GB plan; the 2 GB plan has 1 vCPU.
  • AWS Lightsail: With public IPv4; IPv6-only bundles cost $2 to $4 less. Burstable CPU. No 4 vCPU, 8 GB Linux bundle.
  • Google Cloud: The official page did not show its tables when checked, so these figures are not verified. Compute only: boot disk, external IPv4 and outbound traffic are extra.
  • Azure: From the official retail prices API. Compute only: OS disk, public IPv4 and outbound traffic are extra. Burstable B-series.
Install on your server Read the self-hosting guide