lkml.org 
[lkml]   [2004]   [Nov]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [2.6.10-rc1 and prev] System unuseable while writing to disk
Nick Piggin wrote:
> Prakash K. Cheemplavam wrote:
>
>> 100 0
>> 0 0 0 774876 14532 136460 0 0 0 0 1180 436 1
>> 0 99 0
>> 1 1 0 571740 14732 333736 0 0 32 42628 1325 633 3
>> 43 52 2
>> 2 2 0 217812 15072 676648 0 0 0 45184 1523 1771 14
>> 78 0 8
>> 1 2 0 220436 15072 676648 0 0 0 49664 1582 1365 9
>> 6 0 85
>> 0 2 0 85460 15984 808012 0 0 164 73984 1487 1143 18
>> 32 0 49
>> 0 1 0 85716 15984 808140 0 0 0 29056 1493 1381 4
>> 6 0 90
>> 0 1 0 86100 15984 808140 0 0 0 40320 1420 1367 3
>> 2 0 95
>> 0 1 0 86100 15984 808140 0 0 0 44416 1386 693 0
>> 2 0 98
>> 0 1 0 86100 15984 808140 0 0 0 50304 1518 1413 5
>> 6 0 89
>> 0 1 0 86164 15984 808140 0 0 0 28544 1452 827 1
>> 2 0 97
>> 0 1 0 86292 15984 808140 0 0 0 35840 1386 809 1
>> 2 0 97
>> 0 1 0 86356 15984 808140 0 0 0 37376 1450 1373 3
>> 3 0 94
>> procs -----------memory---------- ---swap-- -----io---- --system--
>> ----cpu----
>> r b swpd free buff cache si so bi bo in cs us
>> sy id wa
>> 1 1 0 86484 15984 808140 0 0 0 42880 1752 2494 7
>> 6 0 87
>> 0 1 0 86484 15984 808140 0 0 0 31872 1608 2098 7
>> 4 0 89
>> 0 1 0 86484 15984 808140 0 0 0 45952 1530 795 1
>> 2 0 97
>> 0 1 0 86484 15984 808140 0 0 0 44288 1515 814 0
>> 3 0 97
>> 0 1 0 90580 15984 808140 0 0 0 29052 1471 1365 5
>> 5 0 90
>> 1 0 0 90148 16108 809356 0 0 976 2088 1383 1319 22
>> 11 0 67
>> 0 0 0 90148 16108 809360 0 0 4 16 1444 833 10
>> 5 84 1
>> 0 0 0 90148 16108 809360 0 0 0 0 1256 473 0
>> 0 100 0
>> 0 0 0 90148 16108 809360 0 0 0 0 1379 611 3
>> 1 96 0
>> 0 0 0 90148 16108 809360 0 0 0 0 1526 2439 16
>> 5 79 0
>> 0 0 0 90148 16164 809360 0 0 0 120 1563 1824 16
>> 4 80 0
>>
>>
>> Beginning and end the system is idle and then I start writing down the
>> file. During that time thunderbird gets unusable.
>>
>
> It looks like it may be a CPU scheduler issue: Nothing needs to
> be read, no swap activity, enough free memory.

I see this behaviour with default scheduler in vanilla and with staircase...


> However, it also does look like there is a lot of CPU power free,
> which makes the cpu scheduler less likely... Maybe interrupts are
> using a lot of CPU time or latency, and that is being hidden from
> CPU time accounting.
>
> I think Ingo did notice quite large latencies in disk interrupts,
> although whether that would be enough to make thunderbird "unusable",
> I'm not sure.

I tried preempt (inkl preempt "big kernel lock" in ck) which doesn't
avoid above. (I don't rember whether it was better, though.)

bye,

Prakash
[unhandled content-type:application/pgp-signature]
\
 
 \ /
  Last update: 2005-03-22 14:07    [from the cache]
©2003-2011 Jasper Spaans