On 11/05/14 06:11, Ian Albany wrote:
IO. Has anyone ever played with preloading the cache into the VFS
cache ie. RAM as opposed to disk. I have loads of ram and when
building say the linux kernel, I can start the build an in the
background read the ccache from one disk and the headers and c files
using a simple script much in the same way as the preload mechanism
works for booting some linux systems. But I think that there is also
some history based on previous usage, so the question is, has anyone
implemented a second level cache preload?
The problem with this idea is that ccache is only ever executed when you
run a compile job, and at that time it is in the critical path of your
build so it must do that absolute minimum to get the job done.
However, you could prime the caches yourself, perhaps.
A simple grep, or md5sum, or something, over the cache directory will
load all the files into memory.
If you have a large cache directory and only want to load some of it
into RAM then perhaps a "find .ccache -anewer somefile" to find any
cache files newer than a file you created previously (unfortunately find
does not appear to provide a way to do that numerically). A source file
you modified just before a recent full build would probably be an
appropriate reference point.
Hope that helps
Andrew
P.S. for maximum build performance, I would recommend mounting your
volumes with "noatime", which defeats the above suggestion. It's a problem.
_______________________________________________
ccache mailing list
ccache@lists.samba.org
https://lists.samba.org/mailman/listinfo/ccache
