2.2.2-pre2: it's back (cache vs. data) & kpiod

Regis Duchesne (samkaski@cs.helsinki.fi)
Mon, 15 Feb 1999 11:55:05 +0200 (EET)

2.2.2-pre2 seems to be again aggressively (not as bad as < 2.2.0-final
though) throwing out data pages in preference of cache pages. This can
be produced by running bonnie. Neither 2.2.1 nor 2.2.0-ac2 behaved this
badly.

Now, for the kpiod matter. I haven't dug any deeper into this issue but
I guess it is because of kpiod since this didn't happen before when I
was running 2.2.1. When running bonnie just after the "Rewriting
intelligently phase" when bonnie begins the "reading with getc()..." my
machine gets totally inresponsive for 30 seconds or so. I observed this
by watching at the "vmstat 1" output in a second xterm, it didn't get
updated for a while (infact I couldn't do anything, not even switch
windows in X). When it did, vmstat gave the appended output. I have the
complete "vmstat 1" log of the run if somebody is interested.

Seems like kpiod isn't doing it's work in a continuous manner and when
it notices the amount of uncompleted requests filling up rapidly it is
already too late. Then the rest of the time is spent to clear up that
queue. Just a guess, I haven't looked how kpiod really works.

procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
..
0 5 0 14760 1264 39908 2136 32 0 69 7000 514 68 3 29 69
0 5 0 14760 1304 39864 2128 0 0 3 3500 454 38 2 16 82
0 5 0 14760 1104 40044 2152 4 0 48 6500 503 51 3 30 68
0 6 1 14760 1444 39612 2264 0 0 115 6801 482 79 2 31 67
0 7 0 14760 1572 39484 2268 0 0 1 699 520 21 1 3 96
0 7 0 14760 1224 39664 2416 0 0 152 7000 522 55 1 31 68
0 8 0 14760 648 40200 2480 32 0 144 7000 485 71 3 31 66
1 7 0 14760 920 39820 2580 16 0 95 6000 497 53 3 27 70
0 8 0 14760 1632 39124 2580 0 0 1 1500 601 28 1 10 89
0 7 0 14760 1460 39140 2736 88 0 105 7500 328 81 2 32 66
0 7 0 14760 1588 39012 2736 0 0 0 500 358 7 2 2 96
0 7 0 14760 1200 39328 2816 4 0 115 6500 440 109 3 31 66
0 7 0 14760 972 39508 2868 0 0 132 7000 507 75 4 31 65
1 7 0 14756 1060 39816 2428 0 0 2 3000 482 31 3 14 83
0 8 0 14756 1604 39328 2364 0 0 1 3000 325 22 2 14 84
0 8 0 14756 1528 39168 2604 132 0 196 6500 481 58 1 30 70
0 7 0 14756 720 39052 3532 292 0 743 0 385 91 2 4 94
1 5 0 14752 1500 36892 4912 300 0 1150 500 235 138 4 5 91
1 0 0 14748 732 35784 6844 124 0 1888 500 211 82 58 9 33
1 0 0 14748 664 33744 9048 0 0 2213 1000 205 36 74 6 21
..

--
Samuli Kaski, samkaski@cs.helsinki.fi
Department of Computer Science, University of Helsinki, Finland.






-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/