| 1 |
On Sun, Dec 28, 2008 at 2:07 AM, Mick <michaelkintzios@×××××.com> wrote: |
| 2 |
> On Saturday 27 December 2008, Nikos Chantziaras wrote: |
| 3 |
>> >> Another reason I |
| 4 |
>> >> didn't put Gentoo on the server is because everyone would start spamming |
| 5 |
>> >> the forums about lag when I emerge -u world while they're getting frags |
| 6 |
>> >> in Counter-Strike :P |
| 7 |
>> > |
| 8 |
>> > set PORTAGE_NICENESS=19 in /etc/make.conf |
| 9 |
>> |
| 10 |
>> I'll just hay "ah, ah, ah" at that one :P OK, I'll also say that it |
| 11 |
>> doesn't work. Everything lags even with 19. |
| 12 |
> |
| 13 |
> Is that measurable? |
| 14 |
Niceness effects are most easily noticeable for CPU bottlenecks, |
| 15 |
because context switching CPUs is relatively painless. Server loads |
| 16 |
are probably more of Memory / IO bottlenecks, and context switching |
| 17 |
between that involves disk swaps and disk prereads. If it is anything |
| 18 |
more than trivial prereads / swaps, then even a 19-niced app can cause |
| 19 |
noticeable bursts of slowdown on a modern system. Niceness isn't |
| 20 |
magic, and compiling, which could easily max out all 3 resources (CPU, |
| 21 |
memory, disk access, 4 if you add network), probably gets the least |
| 22 |
bump from niceness. |
| 23 |
|
| 24 |
Although that isn't to say that PORTAGE_NICENESS doesn't take effect. |
| 25 |
Desktop machines with a single user will probably have speed bursts |
| 26 |
small enough that it wouldn't matter. But I'd think twice if there was |
| 27 |
something intensive supposed to be done on the server. |
Archives