1 |
On Thursday 27 January 2011 23:05:22 Paul Hartman wrote: |
2 |
> On Thu, Jan 27, 2011 at 3:46 PM, J. Roeleveld <joost@××××××××.org> wrote: |
3 |
> > On Thursday 27 January 2011 21:25:02 Paul Hartman wrote: |
4 |
> >> On Thu, Jan 27, 2011 at 2:15 PM, Nikos Chantziaras <realnc@×××××.de> |
5 |
wrote: |
6 |
> >> > On 01/27/2011 09:41 PM, Dale wrote: |
7 |
> >> >> YoYo Siska wrote: |
8 |
> >> >>> Yes. |
9 |
> >> >>> It might not be perfect, but mostly it works pretty well. |
10 |
> >> >>> Once make started 10 or so process, which ate all my ram, because I |
11 |
> >> >>> forgot to reenable swap, when I was playing with something before |
12 |
> >> >>> that |
13 |
> >> >>> |
14 |
> >> >>> :) |
15 |
> >> >>> |
16 |
> >> >>> yoyo |
17 |
> >> >> |
18 |
> >> >> I noticed the same thing with mine. It used a LOT of ram. I have 4Gbs |
19 |
> >> >> and it was up to about 3Gbs at one point and using some swap as well. |
20 |
> >> >> I'm hoping to max out to 16Gbs as soon as I can. May upgrade to a 6 |
21 |
> >> >> core CPU too. |
22 |
> >> >> |
23 |
> >> >> I wonder how much faster it would be if the work directory is put on |
24 |
> >> >> tmpfs? With 16Gbs, that should work even for OOo. |
25 |
> >> > |
26 |
> >> > Btw, if you're using more instances than the amount of CPUs, the |
27 |
> >> > result will be slow-down. |
28 |
> >> > |
29 |
> >> > With the default kernel scheduler, best if amount of CPUs + 1. (On a |
30 |
> >> > 4-core, that's -j5). |
31 |
> >> |
32 |
> >> Once, when building my kernel, I accidentally forgot to specify the |
33 |
> >> number of makes and ran "make -j all". That was a really bad idea, the |
34 |
> >> system became totally unresponsive for quite a long time, much longer |
35 |
> >> than normal kernel build time, but it did eventually finish! |
36 |
> > |
37 |
> > I have found that multi-core systems with sufficient memory can handle |
38 |
> > "-j" (no value) a lot better then sindle-core systems. I do on occasion |
39 |
> > do it with the kernel and can still continue using the system. (For |
40 |
> > comparison, my desktop is a 4-core AMD64 with 8GB memory) |
41 |
> |
42 |
> Strange, in my case it was an i7 920 (4 cores, hyperthreaded, appears |
43 |
> as 8 CPUs to Linux) with 12GB of RAM. Maybe if I prefixed it |
44 |
> with"nice" it would not have brought my computer to its knees... or |
45 |
> maybe related to the schedulers and other kernel voodoo that I don't |
46 |
> understand. I might try it again someday :) |
47 |
|
48 |
That is strange, unless your harddrive is really underperforming? |
49 |
Or do you have all the options in the kernel selected? |
50 |
|
51 |
Btw, HyperThreading doesn't work too well when you have a lot of identical |
52 |
tasks. In that case, you might end up with lesser performance as there are no |
53 |
"usable unused" parts in your cores, but the CPU-schedules (the hardware one |
54 |
for HT) is looking for things to fill those last few bits with. |
55 |
|
56 |
-- |
57 |
Joost |