Performance Tracker project update

[Erik Cederstrand]

Kris Kennaway wrote:
> Robert Watson wrote:
> 
> One thing I am looking at is how to best create a library of world 
> tarballs that can be used to populate a nfsroot (or hybrid of periodic 
> tarballs + binary diffs to save space).  Then you could provide your 
> benchmark in a standardized format (start/end/cleanup scripts, etc) and 
> tell a machine "go and run this benchmark on every daily snapshot for 
> the last year and give me the numbers".

Actually, this just bit me. To make packages for installation, I do a 
standard installworld/kernel/distribution into a directory and a chroot 
install of the necessary ports. I create a .tgz and the end result is 
around 90MB (takes ca. 2 hours). Now I have 186 of those, and my file 
system is full (it's only 15GB, but even a 500GB disk will fill up in 
2-3 months with a fast build server).

I'd like a situation where I can very quickly set up a slave with a 
specific version of FreeBSD to run additional tests or provide shell 
access to a developer. This currently involves adding an entry to a 
queue, rebooting and waiting 2 minutes. Quick and easy, but the 
archiving strategy is obviously very inefficient.

I'm thinking of a couple of options:
1. Having one full install per month and archiving the rest as diffs
    against that by recursively bsdiff'ing every file in the tree (I
    could bsdiff a whole tarball, but bsdiff is very memory-intensive).
    Quick test: 25 mins.
2. Make a hash of all files and only store the binaries where the hash
    is different from the monthly tarball. Faster than 1., but less
    effective. Quick test: 5 mins.
3. Use some kind of VCS. My experience with Subversion and binary files
    is that it's very slow.
4. Throw hardware at the problem.

I'd say it should not take more than 10 mins to recreate an archived 
version. Any thoughts?

Thanks,
Erik