ABM benchmark and experiment definitions used in the Usage Metering project to generate cpu and memory usage statistics.
The benchmarks
directory contains the benchmark configuration files for hg38
and chm13
using the 1x, 10x, and 30x datasets. The benchmark configuration files can be generated using the benchmark.j2
template and the render_template.py
program (link needed):
python3 render_template.py --template benchmark.j2 size=10x genome=chm13
The experiments
directory contains the definitions for the experments defined in the benchmarking plan.
- Tool scalability
- Intel vs. AMD CPUs
- Peak mem usage for tool and scale, given the same input & ref
- Provider
- GCP instance class (n2 vs e2)
Run the hg38
workflow with the Human DNA 10x dataset on the the AWS clusters (m6a and m6i) and the GCP cluster (n2-standard) with the following CPU and memory configurations
CPU | Memory |
---|---|
4 | 16 |
8 | 32 |
16 | 64 |
32 | 128 |
Impact reference genome size has on tool:
- memory consumption and
- runtime
Run the hg38
and chm13
(chicken and locust coming soon) 1x datasets on an AWS m6i 16x64 cluster.
Impact input size has on tool:
- runtime,
- memory consumption
- size of output
Run the hg38
workflow on the 1x, 10x, and 30x datasets on and AWS m6i 16x64 cluster.
Impact memory-to-CPU ratio has on:
- tool runtime and
- memory consumption
Run the hg38
workflow on the 10x dataset on AWS m6i
, r6i
, and c6i
cluster types with 4x8, 4x16, and 4x32 CPU and memory configurations.