| 1 |
On Sun, Oct 15, 2017 at 8:51 PM, Walter Dnes <waltdnes@××××××××.org> wrote: |
| 2 |
> On Sat, Oct 14, 2017 at 06:25:35AM -0500, Dale wrote |
| 3 |
> |
| 4 |
>> Some of my make.conf entries. You may not need all of these so edit out |
| 5 |
>> what you don't want or change values if you need to. I have a four core |
| 6 |
>> CPU. |
| 7 |
>> |
| 8 |
>> FEATURES="-usersync -userpriv -usersandbox buildpkg sandbox parallel-fetch" |
| 9 |
>> |
| 10 |
>> MAKEOPTS="-j5" |
| 11 |
> |
| 12 |
> There is some controversy over setting MAKEOPTS=${number of threads} |
| 13 |
> possibly being better than MAKEOPTS=${number of threads} + 1 |
| 14 |
> https://blogs.gentoo.org/ago/2013/01/14/makeopts-jcore-1-is-not-the-best-optimization/ |
| 15 |
> |
| 16 |
|
| 17 |
I received the same results when I tested it myself. On modern servers |
| 18 |
especially, RAM access speed seems to be matched to core processing |
| 19 |
capability very well. |
| 20 |
|
| 21 |
Strangely, adding to the build thread count doesn't seem to help when |
| 22 |
disk IO is the bottleneck. It is conceivable that it could but in |
| 23 |
practice the location of build data and files seems to be disperse |
| 24 |
enough that there are no great access optimizations, and each read |
| 25 |
blocks individually. |
| 26 |
|
| 27 |
Cheers, |
| 28 |
R0b0t1 |
Archives