1 |
I posted this to the gentoo-user list by mistake last night (it was |
2 |
after midnight and ...) - wondered why I didnt get too many |
3 |
flames/replies! |
4 |
|
5 |
Apologies to those who get two copies, but it was originally meant to go |
6 |
with this thread to stop the mis-information: |
7 |
|
8 |
______posted to gentoo-user_______________________ |
9 |
|
10 |
I'll stick my hand up and say I was the person who installed gentoo for |
11 |
this test. For those who made the previous posts (mostly crap, and who |
12 |
dont seem to have read the article very well - though it could have been |
13 |
more informative), perhaps a few facts may help: |
14 |
|
15 |
1. was fully bootstrapped and compiled as stage 1/2/3 on the machine - |
16 |
not a binary install |
17 |
2. gentoo-sources 2.4.20 was used - Mandrake came with a newer kernel |
18 |
than gentoo's reccomended one (still does), debian was a dogs breakfast |
19 |
because stable is so old. We actually tried to put the gentoo kernel on |
20 |
mandrake/debian when tracking down the ide cable prob, but got too hard |
21 |
- not the way some posts tried to imply) |
22 |
3. optimisations were EXACTLY as recommended by both the make.conf |
23 |
entries, which were supported by the cflags from the forum for this cpu: |
24 |
a 2G celery (P4 based core) I am not sure now, but I believe I ran |
25 |
prelink as well (to match mandrake) - need to find and check the notes. |
26 |
4. Gnumerics problems have been identified and come down to the |
27 |
particular version - is fixed in the upcoming stable release even before |
28 |
this was found, but the project was unaware that what they believed was |
29 |
a slightly slower mod in this version, could be so bad on particular |
30 |
data sets - i.e., 30 odd mins in 1.0.13, but is less that 30s on 1.0.19 |
31 |
on my laptop |
32 |
|
33 |
There seems to be quite a few myths about this test and people upset |
34 |
that months were not spent tuning gentoo and every effort made to |
35 |
cripple the competition! (one person even suggested the faulty ide cable |
36 |
should have been left in the debian box, as that was the way it was |
37 |
delivered!) Read the article, and if you need extra information to |
38 |
reproduce it, email me or or the author (Indy). It is reproducable - if |
39 |
you can obtain the same hardware - I would be very interested if someone |
40 |
has this and the time to really go into the why these results occurred |
41 |
in more detail than I had the chance to. |
42 |
|
43 |
and why was this the result? Daniel Robbins suggested on this list that |
44 |
gentoo-sources may be the problem, but tests on another machine (we had |
45 |
the trial machines for only a couple of days, all of which time was used |
46 |
to build gentoo right up until I ctrl-c'd the OO build so we could do |
47 |
the tests before handing the hardware back) showed that turning off |
48 |
pre-empt and low-latency had zero effect, but changing to an open-mosix |
49 |
kernel 2.4.20 was ~10% slower (no thread export). It seemed to come |
50 |
down to the fact we used -O3 instead of -O2 (think spider might have |
51 |
suggested this ?)- in effect over-optimised, and we didnt have a chance |
52 |
to correct. From my perspective, most of the "he should have used ... |
53 |
may actually have made performance even worse! And besides the time |
54 |
issue, these were supposedly the safe, reccomended flags so we went with |
55 |
them. Please note that even Mandrake made only a slight gain on debian, |
56 |
so 386.586/686 does not make a lot of difference in real world tasks |
57 |
(the original aim of the tests) - the tests did tasks that particular |
58 |
people used linux for in their day-to-day work - no special tests, so no |
59 |
special bias. Yes, I could choose tests that make gentoo shine, or |
60 |
debian, or windowsXP. But I dont do those tests every day, whilst that |
61 |
spreadsheet was/is used as part of my normal work. And its the same |
62 |
with the other tests. |
63 |
|
64 |
So how many gentoo systems out there have every possible optimisation in |
65 |
the book, and are actually running slower than ideal? This is a real |
66 |
problem, and I will be interested in how the cflags projects around |
67 |
handle this, as most seem to aim at setting the maximum possible flags: |
68 |
not actually tune the system for the ones that work best/most stably. A |
69 |
live benchmark test might be more appropriate. |
70 |
|
71 |
Most posts on irc and lists have settled down to "he doesnt know what |
72 |
he's doing" (I do), or the tests were unfair to gentoo (they werent, but |
73 |
then the same criteria were met by all 3 systems, but with some question |
74 |
marks over debian because of its mix - some packages had to be compiled |
75 |
locally, not binary) - but the thrust of the article was not that gentoo |
76 |
was a dud, but that this was the result within the criteria and time we |
77 |
were given, not what we expected, so we need to find out why. Also note |
78 |
that this was not intentionally a debian/mandrake/gentto distro test. |
79 |
|
80 |
We may be getting a P4 hyperthreaded system to play with, but under |
81 |
different rules, where I get to do a bit of tuning first. I have |
82 |
already built the core system on another machine using gcc-3.2.3, |
83 |
"-march=pentium4 -O3 -pipe -fomit-frame-pointer" I note that the |
84 |
pentium4 warning still appears in make.conf, though I believe it no |
85 |
longer applies to this gcc. |
86 |
|
87 |
A while ago I emailed this list and asked for information on tests and |
88 |
settings for HT P4's, without a reply. So again, has anyone done any |
89 |
tests on a HT P4 and is willing to support the flags they chose as being |
90 |
"the best"? In particular, does -ffast-math give a measurable gain? |
91 |
Most of my machines have been built as scientific stations, so accuracy |
92 |
is more important than ultimate speed, so this is one I have never |
93 |
tested. I am not interested in the -O9 -max-everything kiddies who have |
94 |
been so vocal, but reasoned choices. |
95 |
|
96 |
If you want to flame, go ahead - but support your statements! |
97 |
|
98 |
:) |
99 |
|
100 |
BillK |
101 |
|
102 |
|
103 |
|
104 |
On Thu, 2003-08-14 at 05:15, FRLinux wrote: |
105 |
> On Wed, 2003-08-13 at 15:20, Philippe Lafoucrière wrote: |
106 |
> > totally agree ! Btw, gnumeric speed is related to version apparently, |
107 |
> > and they didn't use the official gentoo (patched) kernel ("The same |
108 |
> > 2.4.21 source was copied to all machines"). This sux !! |
109 |
> |
110 |
> Well i don't think this sucks, keeping at least consistency on the |
111 |
> kernel between the 3 distributions is a good idea. Now, that being said, |
112 |
> they seriously fsck'd up the rest of the test and optimisations. |
113 |
> |
114 |
> I don't use specific gentoo kernels on my boxes, this is a personal |
115 |
> choice made about a year ago and actually since 2.6, i don't use 2.4 |
116 |
> anymore on personal machines (my laptop and my workstation). |
117 |
> |
118 |
> Steph |
119 |
-- |
120 |
William Kenworthy <billk@×××××××××.au> |
121 |
|
122 |
|
123 |
-- |
124 |
gentoo-dev@g.o mailing list |