Overview

This charm performs best-practice performance tuning of bcache devices on SD/NVMe based installations.

Cache Devices

Disable performance based congestion behaviour - the SSD or NVMe will in all likelyhood always be faster than the underlying spindle:

congested_read_threshold_us: 0
congested_write_threshold_us: 0

Bcache Devices

Disable writethrough of sequential data writes:

sequential_cutoff: 0

Increase the amount of dirty data the cache device will hold before starting to persist to the backing device:

writeback_percent: 10

Usage

To use this charm, simple deploy and relation to principle services:

juju deploy bcache-tuning
juju add-relation bcache-tuning ceph-osd
juju add-relation bcache-tuning nova-compute

This charm presents no configuration options.