1 |
On Thursday 27 January 2011 22:06:30 YoYo Siska wrote: |
2 |
> On Thu, Jan 27, 2011 at 08:18:34PM +0100, J. Roeleveld wrote: |
3 |
> > On Thursday 27 January 2011 19:56:23 Allan Gottlieb wrote: |
4 |
> > > On Thu, Jan 27 2011, Neil Bothwick wrote: |
5 |
> > > > On Thu, 27 Jan 2011 17:09:27 +0200, Nikos Chantziaras wrote: |
6 |
> > > >> > So on a 20 package world update, only 19 are faster while the 20th |
7 |
> > > >> > runs at the same speed? Where's the loss there? Even if the last |
8 |
> > > >> > were slower, it would be worth it. |
9 |
> > > >> |
10 |
> > > >> Given the amount of time unpack/configure/install of most packages |
11 |
> > > >> needs (very short), my observation is that it would not be worth it. |
12 |
> > > > |
13 |
> > > > Even if that were true, how much time would you have to save to |
14 |
> > > > justify adding -j 2 to EMERGE_DEFAULTS in make.conf? |
15 |
> > > > |
16 |
> > > > But it's not true, large packages spend a lot of time on these phases |
17 |
> > > > of the install. |
18 |
> > > |
19 |
> > > OK I'm convinced since I know that those phases do take noticeable |
20 |
> > > time. |
21 |
> > > |
22 |
> > > I have a "4" processor i7 model 620 (2 cores, doubled for |
23 |
> > > hyperthreading) and have set MAKEOPTS="-j5". |
24 |
> > > If I add -jobs=2 to EMERGE_DEFAULT_OPTS, should I lower |
25 |
> > > MAKEOPTS to 3 (to 4)? |
26 |
> > > |
27 |
> > > thanks, |
28 |
> > > allan |
29 |
> > |
30 |
> > You could, as if you leave it at -j5, you can end up with 2 * 5 = 10 |
31 |
> > processed, eg: similar as if running with MAKEOPTS="-j10" |
32 |
> > |
33 |
> > I think the option that YoYo came with is a good compromise: |
34 |
> > # MAKEOPTS="-j -l5" emerge -j --load-average=5 |
35 |
> > |
36 |
> > Next time I am doing a big upgrade, I'm going to test that to see how it |
37 |
> > behaves. |
38 |
> |
39 |
> I was just building the whole system for my notebook in a chroot on my |
40 |
> desktop machine ( I use FEATURES=buildpkg to build binary packages in |
41 |
> the chroot on a fast desktop machine and then upgrade the notebook with |
42 |
> the binary packages) and I used exactly that (-j -l5 for bot make and |
43 |
> emerge). Can't say if it really is better or not ;) but most of the time |
44 |
> all four cores were busy, though sometimes I saw even 6 or 7 gcc-s |
45 |
> simultaneously in top ;) |
46 |
> emerge was running 3 to 4 jobs most of the time, sometimes dropping to 1 |
47 |
> and once I saw it emerging about 10 parallel packages ;) (mostly small |
48 |
> things, which I guess were doing a lot of |
49 |
> unpacking/configuring/installing but almost nothing of compiling ;) |
50 |
> |
51 |
> |
52 |
> btw, just now I got this error from dev-lang/v8: |
53 |
> |
54 |
> SCons error: option -j: invalid integer value: '-l4' |
55 |
> |
56 |
> seems scons honors MAKEOPTS, but doesn't understand the "loadaverage" |
57 |
> version (-j -l4) |
58 |
|
59 |
You might be able to avoid this by using the "long" version in the MAKEOPTS |
60 |
for "-l"? |
61 |
|
62 |
Eg: |
63 |
MAKEOPTS="-j --load-average=5" emerge -j --load-average=5 |
64 |
|
65 |
-- |
66 |
Joost |