[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Savannah-hackers-public] Re: locate on sv.gnu.org?
From: |
Karl Berry |
Subject: |
Re: [Savannah-hackers-public] Re: locate on sv.gnu.org? |
Date: |
Tue, 9 Jan 2007 09:06:15 -0600 |
'info nice' says that's an advisory to the scheduler, apparently
without clear meaning.
My experience is that it helps a little, but not as much as one would
like. Certainly can't hurt, though.
... assuming the load is a meaningful value to determine the server's
activity, though.
My experience is that it is more meaningful than often said, but
certainly not infallible. Of course there are lots of other system load
gathering programs, like sa.
I never really understood how that works.
My understanding is that it's the average length of the run queue, ie,
the number of processes that would like to run but don't have the cpu.
Do you wish to look for such pattern in:
/vservers/savannah/var/www/uptimes.txt
Well, just because it was easy, I tried to look for times when the load
average was less than one:
awk '$10 < 1 {print substr($1,0,2)}' \
/vservers/savannah/var/www/uptimes.txt |sort | uniq -c | sort -nr
The results (below) seem quite surprising, that the middle of the EST
afternoon is the quietest time. But some greps indicate similar
numbers, so I guess it's right.
So maybe set it up to run with nice -19 once a week at Sunday 13:30? I
doubt that will impact anything too severely. We can always revisit
later, too.
BTW, without dates, uptimes.txt seems like it can only be used for such
massive stats, not for tracking actual problems when the load was high.
Does it have another purpose? Could do a date>> to the file once a day.
Thanks,
Karl
17231 13
17063 14
16591 15
16577 12
16356 18
15024 17
13095 16
12981 23
12082 11
12008 03
11941 19
11812 05
11793 21
11283 22
11131 04
10595 00
10589 02
9948 01
9105 20
7250 10
6561 09
5940 06
2848 08
844 07