lmbench/lat-mem-rd
Memory read latency benchmark sweeping working-set size and stride across the memory hierarchy.
Benchmark Profile
Type: Micro
Kernel subsystem: Memory management, cache/TLB behavior
Coverage: Directly measured
Topology: Single role: executer
Overview
lat-mem-rd measures sequential and random read latency across the memory hierarchy (L1/L2/L3 caches and main memory) for a range of working-set sizes and access strides. It helps identify cache capacity boundaries and memory controller behaviour differences across SUT configurations.
Fastpath Parameters
Name |
Description |
|---|---|
parallel |
Number of parallel lat-mem-rd processes to run (-p flag). Mandatory. |
repeat |
Number of measurement repetitions per data point (-r flag). Mandatory. |
work-set-size-start |
Sweep start value for working-set size in MB. Mandatory. |
work-set-size-end |
Sweep end value for working-set size in MB. Optional; defaults to 256. |
work-set-size-multiply |
Multiplicative sweep step for working-set size. Optional; defaults to 4. |
work-set-strides |
Access stride in bytes. Can be a single value or comma-separated list. Mandatory. |
work-set-random |
Enable random access (-t flag). One of: enabled, disabled, both. Mandatory. |
Result Classes
Result class names encode the stride and working-set size, for example ‘Stride: 128 Size: 4 MB’. The improvement direction is lower-is-better (latency).
Fastpath Metadata
Field |
Value |
|---|---|
Suite |
lmbench |
Name |
lat-mem-rd |
Type |
memory |
Container Image |
registry.gitlab.arm.com/tooling/fastpath/containers/lmbench:v1.0 |
Documentation Path |
documentation/user-guide/benchmarkcatalog/lmbench-lat-mem-rd.rst |
Documentation URL |
https://fastpath.docs.arm.com/user-guide/benchmarkcatalog/lmbench-lat-mem-rd.html |