1 |
On Thursday 07 March 2002 23:22, Jonathan Rippy wrote: |
2 |
> Sounds like it could really help out for those with slower |
3 |
> machines? |
4 |
> |
5 |
> Thought this might be interesting to everyone... |
6 |
|
7 |
hey indeed, thx for the hint! :) imho this looks like a really |
8 |
nice solution for most gentoo users. it's the classical approach |
9 |
of exchanging memory for performance (but in a new way :) |
10 |
|
11 |
due to this tools' capability to put the cache on a nfs drive |
12 |
it's suitable not only for fast/new but also for slow/old |
13 |
systems and especially for old notebooks (where buying another |
14 |
hd might not be "worth" the speedup or hd throughput is |
15 |
generally low anyway). for gentoo on old, space-limited systems |
16 |
and supposed there isn't any nfs available one might consider to |
17 |
delete seldom rebuilt packages from the cache to find a balance |
18 |
between "hd-space-waisting" and a faster remerge process. |
19 |
the real-world benefit of a compilercache for such systems |
20 |
probably depends a lot on how smart the cache-handling mechanism |
21 |
is in caching only often-used packages. thus it's probably worth |
22 |
to refrain from implementing a customized |
23 |
compilercache-like-tool but invest the time saved in developing |
24 |
a smart, customized cache-handling algorithm for the mentioned |
25 |
compilercache. |
26 |
|
27 |
things that might be interesting to discuss for cutting |
28 |
down compile-time or generally improving the (r)emerge process: |
29 |
|
30 |
- compilercache |
31 |
- cross-compilation option (compile on a fast system, rsync |
32 |
afterwards) |
33 |
- mosix (let a fast cpu compile, the slow machine only cares for |
34 |
the cached things etc.) |
35 |
- emerge --world rebuild (thus one could rsync everything to |
36 |
another box and rebuild the whole stuff there -> would still |
37 |
take a lot of time but be much more convenient than selecting |
38 |
and emerging the packages manually) |
39 |
- global (perhaps dynamic) option to set the priority/nice |
40 |
factor of gcc, wget/prozilla, portage and its children (would be |
41 |
suitable both for speeding it up and slowing it down -> gcc |
42 |
eating up all cpu power on a "live" server is not so cool) (-> |
43 |
or does there already exist such an option and i just have not |
44 |
noticed it?) |
45 |
- ?? |
46 |
|
47 |
and of course the combinations of these features |
48 |
|
49 |
regards |
50 |
|
51 |
dan |
52 |
|
53 |
-- |
54 |
...::: Daniel Mettler | http://www.numlock.ch :::.... |