I should say that there's also the possibility of writing a block
device plugin which is highly tuned in some way to the Koji use case.
We've already done discarding flushes and showed that you get all the
benefit of a RAM disk just by doing that (even when backed by a disk).
Is there anything else specific to Koji builds:
- Prepopulating the filesystem in the block device?
- Making a block device which is highly tuned to the filesystem it
will contain (eg. keeping metadata in faster storage)?
- Sharing / deduplicating?
I collected traces of the access pattern of existing Koji builds, but
they are huge and I'm not sure how best to analyze them.
Rich.
--
Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
Read my programming and virtualization blog:
http://rwmj.wordpress.com
virt-df lists disk usage of guests without needing to install any
software inside the virtual machine. Supports Linux and Windows.
http://people.redhat.com/~rjones/virt-df/