| 1 |
lee <lee@××××××××.de> wrote: |
| 2 |
|
| 3 |
> <wabenbau@×××××.com> writes: |
| 4 |
> |
| 5 |
> > lee <lee@××××××××.de> wrote: |
| 6 |
> > |
| 7 |
> >> Hi, |
| 8 |
> >> |
| 9 |
> >> what's taking so long when emerging packages despite distcc is |
| 10 |
> >> used? |
| 11 |
> >> |
| 12 |
> >> I have disallowed compiling on the local machine (which is the one |
| 13 |
> >> emerge is running on) through distcc settings because the local |
| 14 |
> >> machine is relatively slow. Yet I can see some gcc processes |
| 15 |
> >> running on the local machine, and emerging goes painfully slow. |
| 16 |
> >> Using distcc doesn't seem to make it any faster, though disabling |
| 17 |
> >> local compiling seems to help a bit. |
| 18 |
> >> |
| 19 |
> >> Some compilations are being run on the remote machine, so distcc |
| 20 |
> >> does work. The log file on the remote machine shows compilation |
| 21 |
> >> times of a few milliseconds up to about 1.5 seconds at most. The |
| 22 |
> >> distcc server would be finished with the emerging within maybe 15 |
| 23 |
> >> minutes, and the client takes several hours already. |
| 24 |
> >> |
| 25 |
> >> Is there something going wrong? Is there a way to speed things up |
| 26 |
> >> as much as I would expect from using distcc? |
| 27 |
> > |
| 28 |
> > You can try pump mode. Preprocessing is then done on the remote |
| 29 |
> > server. Depending on your hardware, this could be faster. |
| 30 |
> > |
| 31 |
> > But read carefully the manpages of pump and distcc before you use |
| 32 |
> > it. There are some restrictions you should be aware of. |
| 33 |
> |
| 34 |
> I followed the instructions on the wiki which suggest to have |
| 35 |
> 'FEATURES="distcc distcc-pump"' in make.conf and give instructions how |
| 36 |
> to set the CPUs. |
| 37 |
|
| 38 |
I didn't read the wiki. It's some years ago that I used distcc and IIRC |
| 39 |
there was no wiki available for it. |
| 40 |
|
| 41 |
> > You can also try to optimize the number of concurrent compile |
| 42 |
> > processes (-j). Watching the load counts of your client and |
| 43 |
> > server(s) will help you to find out the best value. |
| 44 |
> |
| 45 |
> Using -j doesn't really help. The server is pretty much idling --- or |
| 46 |
> you could say waiting for stuff to compile --- while the client |
| 47 |
> progresses awfully slowly and isn't overloaded with compilation |
| 48 |
> processes. If the server would get more load, emerging could be much |
| 49 |
> much faster. |
| 50 |
> |
| 51 |
> Can it be that the client is simply too slow compared to the server to |
| 52 |
> give it any significant load? (The client isn't exactly slow; it's |
| 53 |
> slow compared to the server.) |
| 54 |
|
| 55 |
I used a pentium 4 laptop as client and two phenom2 quadcore pc as |
| 56 |
server. I don't remember the settings that I used but I think it |
| 57 |
was something about -j10 or so. |
| 58 |
|
| 59 |
When I compiled large programs, the load count of the servers was |
| 60 |
high most of the time and they were very busy with compiling. Only |
| 61 |
at linking time they were waiting for new data. |
| 62 |
Compilation time was much lower than without distcc. |
| 63 |
|
| 64 |
However when I compiled small programs, the benefit of distcc was |
| 65 |
very small or even null. Also compilation time of OpenOffice was |
| 66 |
very long, because of the -j1 setting in the ebuild. |
| 67 |
|
| 68 |
I don't know the reason of your problem. Maybe you should try it |
| 69 |
without pump mode to see if this makes a difference. |
| 70 |
|
| 71 |
Have you used distccmon to see what happens while compiling? IIRC |
| 72 |
it shows you exactly what's going on at each host (preprocessing, |
| 73 |
compiling, waiting). Maybe this will bring some light into the |
| 74 |
whole thing. |
| 75 |
|
| 76 |
And as Dale already said, network speed is very important. |
| 77 |
|
| 78 |
-- |
| 79 |
Regards |
| 80 |
wabe |
Archives