1 |
On Wed, Mar 27, 2013 at 7:50 PM, <gottlieb@×××.edu> wrote: |
2 |
> On Tue, Mar 26 2013, Neil Bothwick wrote: |
3 |
> |
4 |
>> On Tue, 26 Mar 2013 20:43:25 +0530, Nilesh Govindrajan wrote: |
5 |
>> |
6 |
>>> It's better to limit the number of jobs to 2*CPUs (or cores) with a |
7 |
>>> load control like --load-average=N where N is number of CPUs. |
8 |
> |
9 |
> I have two i7-3520Ms. Each has hyperthreading so "counts" as 2. |
10 |
> In particular /proc/cpuinfo describes 4 "cpus". |
11 |
> |
12 |
> Nilesh, Does that mean you recommend |
13 |
> |
14 |
> EMERGE_DEFAULT_OPTS="--jobs --load-average=4" |
15 |
> MAKEOPTS="--jobs=8 --load-average=4" |
16 |
> |
17 |
> In particular I am not sure if your recommendation for load-average |
18 |
> applied to EMERGE_DEFAULT_OPTS as well or if you were just discussing |
19 |
> MAKEOPTS. |
20 |
> |
21 |
> thanks, |
22 |
> allan |
23 |
> |
24 |
|
25 |
It was for MAKEOPTS. If you have a really powerful processor, consider |
26 |
splitting it for emerge options and make options. Would be faster. |
27 |
|
28 |
@Walter, I'm also on a dual core machine, and as per my observation |
29 |
over long emerges, load doesn't cross 2.2. |
30 |
I have also observed that if it is limited to 2, system seems to be |
31 |
under utilized, because make checks the 1 minute average instead of 15 |
32 |
minute average (well, it doesn't make sense otherwise). |
33 |
|
34 |
This could apply to bigger processors too, so if you want full |
35 |
utilization, slightly extrapolate the load average limit. Say by |
36 |
20-30%. |
37 |
|
38 |
-- |
39 |
Nilesh Govindrajan |
40 |
http://nileshgr.com |