What is the difference between a memory request and a memory limit?

In Kubernetes, a request is the amount the scheduler reserves on a node — the pod is guaranteed this memory. The limit is the hard ceiling; exceeding it causes the container to be OOMKilled. Docker's --memory flag sets the limit directly (there is no separate request concept in plain Docker).

How much headroom should I add over observed peak?

A 25–50% buffer over observed peak is typical for stateless services. JVM applications need extra room for GC overhead and off-heap buffers — 50–80% headroom is common. ML inference servers serving large models may need 100%+ to account for input tensor allocations.

What limits-to-requests ratio should I use?

A ratio of 1.25–1.5× is common for memory. Unlike CPU, memory cannot be throttled — a container that exceeds its limit is immediately killed. Keep the ratio modest (1.0–2.0×) to limit blast radius from memory leaks and to avoid surprising OOMKills on overcommitted nodes.

Why does the JVM need special treatment?

The JVM allocates heap (controlled by -Xmx) plus off-heap memory for code cache, metaspace, threads, and native libraries. Total JVM process memory often runs 20–40% above -Xmx. The calculator adds this overhead automatically when you select the JVM app type.

Should I set a memory limit in Docker if there is no Kubernetes?

Yes. Without --memory, a runaway container can consume all node memory and trigger the kernel OOM killer, which may kill unrelated processes. Setting explicit limits contains the blast radius.

What units does Kubernetes use for memory?

Kubernetes accepts plain bytes, or suffixes K/M/G (decimal) or Ki/Mi/Gi (binary). For container memory the binary suffixes Mi (mebibytes) and Gi (gibibytes) are standard and match what monitoring tools report.

Container Memory Limit Calculator

Name: Container Memory Limit Calculator
Availability: InStock
Author: Nham Vu

Enter your app type, observed memory usage, and headroom to get recommended Docker/Kubernetes memory requests and limits.

Workload Parameters

Application Type

Baseline (Idle) Memory (MiB)

Memory used at startup with no active traffic. From docker stats or Prometheus.

Observed Peak Memory (MiB)

Highest memory recorded under production or load-test traffic.

Headroom above Peak (40%)

5%150%

Limit / Request Ratio (1.5×)

1.0×3.0×

Fill in the parameters and click Calculate Limits.

Copied!

Summary

Enter your app type, observed memory usage, and headroom to get recommended Docker/Kubernetes memory requests and limits.

How it works

Select your application type (web server, worker, batch job, JVM, or ML model).
Enter the observed baseline and peak memory usage from your monitoring tool.
Choose a headroom percentage to buffer against traffic spikes.
Set the limits multiplier that controls how far above requests the hard cap sits.
The calculator outputs memory requests and limits in both MiB and GiB, plus ready-to-paste Docker and Kubernetes config snippets.

Use cases

Set Docker --memory flags before deploying a containerized service.
Fill in Kubernetes pod spec memory requests and limits from monitoring data.
Prevent OOMKill events by sizing limits above realistic peak usage.
Reduce wasted cluster memory by avoiding over-provisioned requests.
Compare memory profiles across different application types.
Estimate total cluster memory needed when scaling to N replicas.
Tune JVM heap vs. container limit ratio for Java services.

Frequently Asked Questions

Last updated: 2026-07-22 · Reviewed by Nham Vu