I started a bulk build of the same pkgsrc-current code at the same time on two systems: avalon.dragonflybsd.org and df.v12.su.
Avalon has a Xeon X3220 2.4G CPU, 3G RAM, 150M SATA drives. df.v12.su has a Core 2 Duo 2.4G CPU, 3G RAM, 300M SATA drives. Both have comparable bandwidth, as far as I can tell. Both are on Hammer. Avalon is running a cpu-bound process that is tying up core 1 almost continuously, causing occasional and noticable pauses at the terminal. When starting the build, df.v12.su finished scanning and started building packages first, and continued to lead. I figured this was because of that long-running process on avalon. Checking in now, df.v12.su has 6257/8849 packages built, and avalon has 7218/8849. The only differences are that snapshots are still on (though infrequent) on df.12.su and avalon is running that one program eating up CPU. So, either the snapshots are affecting the run a lot more than that CPU-eating program, or there's some other factor I missed. It's not a very scientific test, but it makes a case that we may want to test snapshots to see if that makes a significant performance difference. Assuming my theory is right, that is.
