Date: Mon, 7 Jul 1997 22:19:16 -0400 (EDT) From: Thomas David Rivers <ponds!rivers@dg-rtp.dg.com> To: ponds!freefall.cdrom.com!freebsd-hackers, ponds!lakes.water.net!rivers, ponds!sdf.com!tom Subject: Re: Why news expiration is sooo slowww with 2.2.x. Message-ID: <199707080219.WAA03025@lakes.water.net>
next in thread | raw e-mail | index | archive | help
> > I should have mentioned, this is a 386-33 (with an Intel 387) and
> > only 8megs of RAM. Prior to 2.2.1, it took only a few hours to expire;
> > now it takes many days.... Examining swapinfo during this process
> > indicates swap is mostly free, so I don't believe my constrained memory
> > is the problem (I'm not even swapping...) Certainly my CPU is
> > slow, but it wasn't any faster prior to version 2.2.1.. :-)
>
> Yes, but is your machine doing for these two to three days?
Yes - it seems to be...
> Is the CPU
> load high?
According to top - the CPU seems to be about 25-30% idle. Also,
interestingly enough; doing an rm on as many files as I can before I run
into the sh's command line limit takes about 20 minutes itself. top
indicates the machine is still about 20-30% idle, with ~26% of the
time taken up doing the 'rm'. The rm process seems to be in biowait
a good portion of that time; which tends to indicate a disk bottle neck...
e.g. - here's the first few lines of top - shortly after one of the
"big" rm commands got started...
last pid: 25001; load averages: 0.57, 0.61, 0.70 22:08:23
32 processes: 1 running, 30 sleeping, 1 zombie
CPU states: 1.2% user, 0.0% nice, 25.2% system, 35.1% interrupt, 38.5% idle
Mem: 3208K Active, 36K Inact, 2020K Wired, 1096K Cache, 515K Buf, 456K Free
Swap: 150M Total, 5340K Used, 145M Free, 3% Inuse
PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND
25000 news -6 0 300K 88K biowai 0:14 24.16% 22.96% rm
25001 root 38 0 308K 680K RUN 0:07 12.83% 11.48% top
23095 root 18 0 448K 236K pause 0:03 0.00% 0.00% csh
> Is the disk load high (iostat)? To what disk?
Only have the one disk - the news directories are on a different
partition; but that shouldn't affect the number of I/Os... Here's
the result of iostat -c 3 (this was done during one the the 'rm' jobs,
I had to kill expire and begin rebuilding the history file):
# iostat -c 3
tty wd0 cpu
tin tout sps tps msps us ni sy in id
972 93 684 43 17.3 7 0 29 28 37
0 37 922 57 10.5 1 0 29 33 38
0 35 773 47 13.2 1 0 26 32 41
> Do you do
> delayed article removal?
I run expire once a night; my explist does delay some things
(hold on to them for a week) but the default for the bulk of news
is to expire after two days...
>
> I can't tell the difference between 2.2 and 2.2.1 in terms of speed.
Hmm... right now I can't recall if I ever had 2.2 running on the news
server; so my experience may be relative to 2.1.7.1.
I'd have to guess that things seem to be waiting in biowait() for
quite some time... could it be changes in the wd driver (this is an IDE
disk)?
- Dave Rivers -
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199707080219.WAA03025>