1 |
tomas posted <435D5011.6080506@××××××.lt>, excerpted below, on Mon, 24 |
2 |
Oct 2005 21:20:17 +0000: |
3 |
|
4 |
> What is the difference between gentoo-sources and: |
5 |
> * sys-kernel/ck-sources |
6 |
> Latest version available: 2.6.13_p8 |
7 |
> Latest version installed: [ Not Installed ] |
8 |
> Size of downloaded files: 37,599 kB |
9 |
> Homepage: http://members.optusnet.com.au/ckolivas/kernel/ |
10 |
> Description: Full sources for the Linux kernel with Con Kolivas' |
11 |
> high performance patchset and Gentoo's basic patchset. |
12 |
> License: GPL-2 |
13 |
> |
14 |
> and is there any way to make my gentoo work faster? for example encoding |
15 |
> dvd? |
16 |
|
17 |
The CK sources have a lot of work specifically designed to lower latency. |
18 |
Thus, they are popular with real-time and media production folks, where |
19 |
certain things need to occur within a specified time. |
20 |
|
21 |
There is a tradeoff, however, between latency and thruput. Generally |
22 |
speaking, one way to decrease latency is to break up large tasks into a |
23 |
bunch of smaller tasks, doing only one or a few of the smaller tasks at |
24 |
once, before checking to see if something else needs done. The problem is |
25 |
that each task switch requires some bit of overhead, some tiny bit of time |
26 |
just to do the task switch. Thus, it's possible to break tasks small |
27 |
enough that most of the time is spent changing tasks, rather than actually |
28 |
doing them. |
29 |
|
30 |
Analogy: Suppose you have two tasks to do in your eight hour work day. |
31 |
You can do each until it's done, then switch to the next, and it might |
32 |
take you seven hours, because you always know where in the task you are. |
33 |
You can take an hour lunch or knock off an hour early. However, if one of |
34 |
those tasks is something that you have to check every half hour, and spend |
35 |
five minutes working on it before going back to the other task you were |
36 |
working on, each time you go back to that other task, you have to remember |
37 |
where you were and find your place again, which might take you three to |
38 |
five minutes. Instead of getting done in 7 hours and having that hour |
39 |
free, you could now be working after quitting time, still trying to |
40 |
finish, because of all the time spent finding your place after each |
41 |
interruption. |
42 |
|
43 |
So... low latency generally means a more responsive system under heavy |
44 |
load... no freezing mouse syndrome... but the computer will probably take |
45 |
longer to finish what you give it to do... similar to a speed grade or two |
46 |
reduction. |
47 |
|
48 |
One way around the issue, the way I've taken, is a dual-CPU (or now, dual |
49 |
core) system. You can then tolerate normal latency settings, or even |
50 |
higher than normal latency settings, without affecting responsivenes, |
51 |
because even when both CPUs are heavily loaded, one or the other will get |
52 |
around to processing the latest mouse movement or sound processing in an |
53 |
average of half the time it would otherwise ordinarily take. For system |
54 |
responsiveness while encoding DVD or the like, a dual-CPU or dual-core |
55 |
system is really the way to go. I've been *MORE* than happy with my |
56 |
decision to go dual Opteron, shortly after the Opterons came out. Now, |
57 |
the same thing is cheaper, doing with single-CPU dual-core, therefore far |
58 |
cheaper and less complex motherboards (I paid over $400 for my dual CPU |
59 |
mobo, dual-core compatible single socket boards are now under $100), what |
60 |
I did with dual CPU a couple years ago. I don't expect I'll ever go back |
61 |
to single core/CPU, so it's a good thing they are going dual-core now. |
62 |
|
63 |
-- |
64 |
Duncan - List replies preferred. No HTML msgs. |
65 |
"Every nonfree program has a lord, a master -- |
66 |
and if you use the program, he is your master." Richard Stallman in |
67 |
http://www.linuxdevcenter.com/pub/a/linux/2004/12/22/rms_interview.html |
68 |
|
69 |
|
70 |
-- |
71 |
gentoo-amd64@g.o mailing list |