1 |
Hi there, I have big performance issues with the gentoo kernels, both |
2 |
2.4.17-r5 and 2.4.19-r1. |
3 |
My system is a SMP machine (dual-athlon) with only 256MB of ddr ram. |
4 |
(so.. I swap a lot, 1gb swap partition) |
5 |
The system is pretty up to date. and spread around two disks(both with a |
6 |
similar performance of 15-18MB/s), 1 for the /home and swap partition, |
7 |
another for the "/". The filesystem is reiserfs, Window Manager: kde3 |
8 |
|
9 |
The problem: |
10 |
While running heavy aplications: |
11 |
kde3, robocode(java), netbeans(java IDE), mozilla, and others like, |
12 |
konqueror, xmms, xchat, gkrellm.. and konsole. |
13 |
please note that java is sun's jre-1.4.0 (the faster one around, |
14 |
acondingly to my benchmarks) |
15 |
|
16 |
Something like 140 processes. |
17 |
Some use plenty of ram: 59M(java),31M(mozilla-bin),28M(X), 19M(java) |
18 |
|
19 |
while, kernel compiling, running championchips in robocode (very |
20 |
intensive java aplication, with a lots of processes and ram (robots |
21 |
fighting each other in a arena)), web browsing with konqueror, checking |
22 |
email with mozilla-mail |
23 |
and the other apps are normal apps, xmms is very good to check system |
24 |
latency because its playing music(so we notice when things start to "hand") |
25 |
|
26 |
Basically, the system was so unresponsive that I had to close everything |
27 |
but robocode, wich crashed at the end. |
28 |
|
29 |
Now, i have my own kernel, 2.4.18 with o(1)-scheduler-k3, preempt-patch, |
30 |
and nothing else. |
31 |
I'm reproducing the load and the system is quite fine, obviously this |
32 |
isn't super-responsive, but its usable. |
33 |
|
34 |
here's a sample of vmstat: |
35 |
--(m3thos#:~/robocode/robots)-- vmstat 5 |
36 |
procs memory swap io |
37 |
system cpu |
38 |
r b w swpd free buff cache si so bi bo in cs us |
39 |
sy id |
40 |
7 1 0 314508 66424 10164 28856 30 25 75 77 150 1416 |
41 |
32 5 63 |
42 |
25 0 0 314508 18304 12384 30044 9 0 194 815 334 7721 70 |
43 |
11 19 |
44 |
6 0 0 314508 53764 13616 30836 0 0 70 530 298 3074 |
45 |
92 7 1 |
46 |
3 5 1 314508 61260 14624 32360 1294 0 1495 360 378 3939 |
47 |
86 8 6 |
48 |
8 1 0 314508 33216 14700 35216 2290 0 2807 69 518 3004 |
49 |
91 7 1 |
50 |
16 4 0 314508 19588 15816 37464 310 0 600 503 336 4516 80 |
51 |
10 10 |
52 |
17 2 0 314516 9172 8696 30876 105 25 198 851 330 2903 |
53 |
91 9 0 |
54 |
11 0 0 314516 10644 9696 31540 86 0 175 434 315 2849 |
55 |
94 6 0 |
56 |
10 1 1 316252 30228 8336 29592 729 374 874 766 357 3552 |
57 |
93 7 0 |
58 |
8 0 1 316252 30968 8404 30432 526 0 640 141 339 6229 |
59 |
92 8 0 |
60 |
8 0 0 316252 52048 9320 31956 1966 0 2053 415 398 6080 |
61 |
83 8 8 |
62 |
|
63 |
here's a sample of free: |
64 |
--(m3thos#:~/robocode/robots)-- free |
65 |
total used free shared buffers cached |
66 |
Mem: 255660 237400 18260 0 6680 24328 |
67 |
-/+ buffers/cache: 206392 49268 |
68 |
Swap: 1225996 318812 907184 |
69 |
heres the uptime: |
70 |
--(tigre)-(pts1)-(22:04/09-Apr-02)-- |
71 |
--(m3thos#:~/robocode/robots)-- uptime |
72 |
10:08pm up 4:04, 1 user, load average: 17.72, 15.03, 10.99 |
73 |
|
74 |
|
75 |
and I'm attaching the config file that it's being used |
76 |
|
77 |
I'v noticed first in 2.4.17-r5 that responsive was flacky, and after |
78 |
upgrading to 2.4.19 saw that it was still pretty bad under high loads, |
79 |
but after my own kernel the responsiveness increased a lot! |
80 |
|
81 |
I'm sending this mail to this ML because I think that this conserns the |
82 |
kernel mantainers, and all developers since a bat base kernel in terms |
83 |
of responsiveness is not a good thing to such a "performance oriented" |
84 |
distribution. |
85 |
p.s.: obviously that I don't normally have this kind of high loads, but |
86 |
.. some times I do, and not only for testing(has this time), but also on |
87 |
current use(has was the time I experimented big responsiveness issues |
88 |
with 2.4.19-r1). |
89 |
|
90 |
|
91 |
Thank you all! |
92 |
Great Distro! |
93 |
|
94 |
Miguel Sousa Filipe |
95 |
handle: m3thos |
96 |
More human than human |
97 |
(PORTUGAL) |