Subversion Compilation Benchmarks

Interesting. I didn't quite believe that daily reboots would help, but this is pretty good evidence. I'll ask that people give #s in the future by 1. quitting all apps, and 2. another set after rebooting.

Incidentally, did you reboot your iMac Core Duo before running the compile that gave you 1:57?

No. I didn't bother because I hadn't been using it very hard. (The G5 is still my main Mac.) But, for consistency, I just ran the compile after rebooting the iMac, and it took 1:56.

The version of gcc also seems to matter. The above numbers are for gcc 3.3 on the G5 and gcc 4 on the iMac. When I changed to gcc 4 on the G5 and rebooted it took 3:05.

PowerBook G4 (1.5 GHz, 1.25 GB) using gcc 4 after a restart: 8:54

I thought it was weird when you mentioned how long Python took to startup after a few days of uptime because I'd never seen that. I compared compiles on my Dual 1.8 G5, 2GB RAM.

mark@eagle:/usr/local/src/subversion-1.3.0$ uptime
22:15 up 13 days, 9:40, 2 users, load averages: 0.39 2.67 2.21

Downloaded subversion, opened Terminal and did the tar-configure-make dance:

mark@eagle:/usr/local/src/subversion-1.3.0$ time make -j8
.
.
real 3m29.461s
user 3m45.418s
sys 2m31.063s

That's with all my apps open, and my wife logged in via fast user switching. gcc 4.0.1.

I immediately ran the make a second time and shaved 4 seconds off.

The 4 second difference doesn't seem like much, but is a bit curious. On a Linux box I did the same test and got:

real 1m16.499s
user 1m53.587s
sys 0m34.572s

That was the second run. First run was about 0.6s slower. Not nearly as much of a difference between runs. The Linux box has been up for 122 days and is running several instances of Tomcat and Zope, so it's anything but "fresh".

Coming to Mac from the UNIX world, I have to admit that I normally leave my G5 running all the time, only rebooting to apply security updates.

Here are results from the Mac Pro.

MacBook Pro (2.33 GHz, 3 GB) using gcc 4 after a restart: 1:44