Continuous profiling with answers
One number, one line of code: here is what is making your app slow.
Bottleneck profiles your production services continuously and publishes a weekly ranked list of what is slowing them down: the endpoint, line or query, what it costs in milliseconds and compute, and the change to make.
No account. No card. No sales call.
At a glance
- A ranked verdict, not a dashboard
- The endpoint, line and query to fix
- Regressions traced to the pull request
What is one slow endpoint costing you each month?
Put in an endpoint's traffic, its response time now and after a fix, the compute it uses, and the time spent investigating. See the compute and user waiting a faster endpoint would free each month. Runs in your browser. No account, no card, no call.
How it adds up, each month
- Cut in response time
- 63%
- User waiting on this endpoint a month
- 3,378 hours
- Compute this endpoint uses a month
- $1,800
Take this with you
No email required. It is your result.
Bottleneck — slow endpoint cost calculator · 500,000 requests a day, 800 ms now, 300 ms target · Compute freed: $1,125 a month, $13,500 a year · Investigating slowness: $1,900 a month Run it yourself: https://bottleneck.run
Method: compute this endpoint uses = service compute bill x its share. Compute freed = that cost x the cut in response time, as a share of the current time, which assumes compute grows with time spent serving requests. User waiting = requests x response time, over a month. Every input is your estimate; a production profile measures them.
The problem
Observability tools produce beautiful dashboards and no answers.
A team stares at flame graphs and latency percentiles and still argues about what to fix, so performance work gets deprioritised until something breaks.
The information exists but the conclusion does not, and drawing the conclusion requires a skill most teams do not have on staff.
Who it is for
Engineering teams without a performance engineer
Teams at 10 to 300 person companies running production web services.
Tech leads told the app feels slow
Leads who need an answer to what to fix, not another dashboard to argue over.
The conclusion, not just the data
Continuous profiling that tells you what is making your app slow, what it costs, and what to change.
Always sampling production
It samples production continuously at negligible overhead.
Latency attributed precisely
Latency and cost are attributed to specific functions, queries and dependencies.
A weekly ranked list
Each week, an ordered list: this endpoint, this line, this query, costing this many milliseconds at this percentile.
What it costs in compute
Every item shows how many dollars of compute it costs a month.
The change to make
Each item comes with the specific change to make and the expected improvement.
Regressions caught per deploy
Regressions are caught on each deploy and attributed to the pull request that caused them.
How it works
Bottleneck, from the first step to the result.
- 01
Install the agent
It starts sampling production continuously.
- 02
Get the ranked list
Every week, the bottlenecks in order, with cost and code location.
- 03
Make the change
Each item names the change and its expected improvement.
- 04
Catch regressions
New slowdowns are traced to the pull request that caused them.
Pricing
Free for one small service, then priced per service.
Free
Freeone low-volume service
For a single service under a low request volume.
- Continuous profiling
- Ranked bottlenecks
Service
Recommended$199per service, per month
For services in production.
- Weekly ranked verdict
- Compute cost per bottleneck
Team
$899per month, up to 20 services
For teams running many services.
- Per-deploy regression detection
- Cost attribution
Prices in USD.
Questions people actually ask
- How is this different from our observability tools?
- Instead of dashboards, it publishes a ranked verdict: the endpoint, line and query costing the most, with the change to make and the expected improvement.
- Does profiling slow production down?
- It samples production continuously at negligible overhead.
- Does it show cost as well as latency?
- Yes. Each bottleneck shows the milliseconds it costs at a given percentile and the dollars of compute it costs a month.
- How are regressions handled?
- Regressions are caught per deploy and attributed to the pull request that caused them.
- Is there a free plan?
- Yes. It is free forever for one service under a low request volume.
- What do paid plans cost?
- 199 dollars a month per service, or 899 dollars a month for up to 20 services with per-deploy regression detection and cost attribution.
- Why should I trust the slow endpoint cost calculator?
- It is arithmetic on your own estimates, worked out in your browser, and it assumes compute grows with the time spent serving requests. It does not see your system; the free one-time production profile names the top five bottlenecks and their compute cost.
What is one slow endpoint costing you each month?
Put in an endpoint's traffic, its response time now and after a fix, the compute it uses, and the time spent investigating. See the compute and user waiting a faster endpoint would free each month. Runs in your browser. No account, no card, no call.
Open the free toolIt runs in your browser. Bottleneck never sees your inputs.