1 |
Hi! |
2 |
|
3 |
On Wed, 29 Aug 2012, Duncan wrote: |
4 |
> So in practice, just what are the sorts of times, relative to stand-alone- |
5 |
> build udev, we're talking about? In all this discussion, what, hundreds |
6 |
> of posts by now?, I've not seen ANYONE actually ask, let alone answer, |
7 |
> THAT. But it would seem to be a rather important question... |
8 |
|
9 |
As a first crude datapoint, I compared the build times |
10 |
(configure+make) of udev-171-r6 and -188 on our dev Alpha. This |
11 |
is a machine that's on the speedier side of off-mainstream |
12 |
architecures, but as a datapoint, it should be enough. |
13 |
|
14 |
Tests were run just after ebuild [foo] prepare, i.e. patching was |
15 |
_not_ measured. Since the machine has enough RAM, everything was |
16 |
in the page cache. |
17 |
|
18 |
ver (1) (2) (1+2) |
19 |
udev-171-r6: 15s / 71s / 86s; 1m26s |
20 |
udev-188 : 28s / 636s / 664s; 11m04s |
21 |
|
22 |
1: "./configure" time |
23 |
2: "make" time |
24 |
|
25 |
Other observations: |
26 |
- The machine in question did not have dbus, or libcap, which |
27 |
some would argue need to be factored into "build cost". |
28 |
- I expected more difference for the configure phase, since it |
29 |
is what I usually perceive as the slow part of many builds |
30 |
when archtesting. Probably some packages suffer more from this |
31 |
than others. Also, configure does not run in parallel, make |
32 |
does. |
33 |
- Test suite run times were not checked. Though it looks like |
34 |
the -188 ebuild builds the test suite binaries regardless. |
35 |
- This was run as make -j1 even though the machine has 4 CPUs. |
36 |
One reason was to make the compile time longer for better |
37 |
measurement granularity, the other was that most slower |
38 |
machines (as far as we are concerned) are single-cpu. |
39 |
|
40 |
HTH, |
41 |
Tobias |