On 5/14/19 5:12 PM, Christopher Browne via talk wrote:

Sounds like it may be the wrong moment in time to be adopting BCache-related stuff ;-)

Has anyone been poking at this sort of stuff? A *loooong* time ago, I used to follow ReiserFS pretty closely, back when that project was a technical matter, as opposed to being (rather properly!) overshadowed by a murder investigation :-(

We use it for every customer's Openstack cloud for almost every openstack component. Nova, Ceph, Swift, Keystone, and MySQL are the major services, and each has a bcache underlay along with a hardware RAID controller. It has proven to be quite a reliable (and fast) setup, and even in the event of a power cut we haven't lost any data. For nova especially where users are interacting with VMs, it means a customer can have large amounts of storage allocated for a VM, and writes feel as though the whole system is running on an SSD. Likewise for Keystone & MySQL, where in busy clouds with clustered MySQL, tokens can take a long time to generate, fetch, and authenticate without the SSD backing. All around quite useful, just don't run it on the leading edge of GCC, Kernel, and OS releases!