Memory Resource Units
What you'll learn: How Kubernetes measures memory so you can accurately request and limit RAM for your containers.
Why Memory Units Matter
In the previous lesson you learned about CPU units (millicores). Memory works differently—it's measured in bytes, which are straightforward but can get unwieldy when dealing with gigabytes of RAM. Kubernetes lets you express memory using convenient suffixes instead of writing out enormous numbers.
Understanding Memory Suffixes
Kubernetes accepts memory values in bytes or with these suffixes:
- Plain number: bytes (e.g.,
1000000= 1 million bytes) - E, P, T, G, M, K: Powers of 1000 (decimal)
- Ei, Pi, Ti, Gi, Mi, Ki: Powers of 1024 (binary)
Binary vs Decimal
This is where it gets interesting. Computer memory naturally works in powers of 1024 (binary), while marketing and SI units often use powers of 1000 (decimal).
Binary (base-2):
1 Ki= 1024 bytes1 Mi= 1024 Ki = 1,048,576 bytes1 Gi= 1024 Mi = 1,073,741,824 bytes
Decimal (base-10):
1 K= 1000 bytes1 M= 1000 K = 1,000,000 bytes1 G= 1000 M = 1,000,000,000 bytes
Practical Example
If you want to request 128 megabytes of memory, you could write:
134217728(bytes—hard to read!)128Mi(binary—preferred, matches how RAM actually works)128M(decimal—slightly less than 128Mi)
Best practice: Use binary units (Ki, Mi, Gi) since they align with how memory is physically allocated.
Key Takeaway: Kubernetes measures memory in bytes with convenient suffixes like Mi (mebibytes, 1024²) and Gi (gibibytes, 1024³)—always prefer binary units for accuracy when requesting memory resources.