Date: Tue, 18 Jan 2000 11:40:29 -0500 (EST) From: spork <spork@super-g.com> To: freebsd-hackers@freebsd.org Cc: noc@inch.com Subject: high load, nothing happening? (LONG) Message-ID: <Pine.BSF.4.00.10001181137380.10422-100000@super-g.inch.com>
next in thread | raw e-mail | index | archive | help
Hi,
I'm trying my luck over here at -hackers, I already posted to -questions
and -isp without any resolution. I'm including my original post and below
that a summary of some responses and my answers... The bulk of the
recent answers seem to say "you should see a load of 3.0 on a machine with
500 processes", but I don't quite agree as I have other similar machines
that are in production that don't even approach a load of 1.0 with 3-400
processes. I can only assume something is wrong somewhere... You all are
my last hope...
[begin orginal post]
We just built a large webserver machine (PII-450, 896MB RAM, 30-odd G of
Mylex RAID, 3.3-R) that constantly runs a load of from 1 to 3, even though
it's not doing anything (still sitting as a staging server). The initial
startup is also very slow; after about 40 of the servers start there's
about a 15 second pause, then another 40, pause, etc...
This box is running about 170 virtual hosts (and a full class C
of addresses aliased to fxp0) under Apache 1.3.9, with each vhost running
as it's own user and starting 3 servers at startup, so there are a large
*number* of processes, but no swapping with about half a gig of RAM left
free.
I have maxusers at 512, NMBCLUSTERS at 4096, and the following sysctl
adjustments:
kern.maxproc: 8212
kern.maxfiles: 100000 kern.maxfilesperproc: 16424
kern.maxprocperuid: 8211 kern.ipc.somaxconn: 512
This is all gathered from various "tuning for a big webserver" posts from
the various FBSD lists.
systat, vmstat, iostat all look normal, and I've not seen any curious
entries in the logs.
So that's the info, my questions are "why the load", and "is that OK"?
Something seems wrong here, but I'm at a loss.
Any ideas where to start looking?
[followup #1]
> What does top(1) report?
last pid: 23684; load averages: 3.74, 1.96, 1.46 up 7+21:10:15 10:35:38
449 processes: 1 running, 448 sleeping
CPU states: 0.0% user, 0.0% nice, 2.7% system, 0.0% interrupt, 97.3% idle
Mem: 62M Active, 355M Inact, 45M Wired, 8350K Buf, 418M Free
Swap: 784M Total, 784M Free
PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND
23684 root 30 0 1976K 944K RUN 0:00 3.08% 0.29% top
904 root 2 -12 1036K 720K select 0:31 0.00% 0.00% xntpd
4163 root 2 0 1468K 1096K select 0:13 0.00% 0.00%
httpd-apache_1
3399 root 2 0 1468K 1096K select 0:13 0.00% 0.00%
httpd-apache_1
[followup #2]
> that value for NMBCLUSTERS is going to be lower than what maxusers at
> 512sets it to, try 16384 or leave it up to maxusers.
[followup #3]
> Hum....that could certainly contribute to load. Have you checked vmstat
^^^^ (he's referring to the number of processes)
> to see what the system calls are like (frequency that is).
Nothing's blocked, and the other numbers look very similar to much smaller
boxes doing nothing:
procs memory page disks faults cpu
r b w avm fre flt re pi po fr sr da0 fd0 pa0 in sy cs us sy id
0 0 0 106760426976 2 0 0 0 2 0 0 0 0 230 474 155 0 0 99
0 0 0 106760426976 4 0 0 0 0 0 0 0 0 233 408 136 0 2 98
0 0 0 106760426976 3 0 0 0 0 0 0 0 0 235 408 136 0 2 98
[followup #4]
> what ????
> you are asking why high load ???
> don;t you see you have 500 processes on your box ??
> it's normal to have 3 of load average if you got 500 processes!
Here's a snippet from a shell/web server that is doing actual work. It
has less memory, a slower processor and a number of interactive users.
The load however rarely climbs above 1.0 unless a process goes runaway:
last pid: 25042; load averages: 0.38, 0.35, 0.63 13:26:43
301 processes: 1 running, 300 sleeping
CPU states: 0.4% user, 0.0% nice, 0.8% system, 0.8% interrupt, 98.1% idle
Mem: 119M Active, 44M Inact, 36M Wired, 34M Cache, 6027K Buf, 17M Free
Swap: 640M Total, 37M Used, 603M Free, 6% Inuse
PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND
25040 root 28 0 844K 1120K RUN 0:00 1.89% 0.34% top
24823 freddy 2 0 4180K 2964K select 0:00 0.23% 0.23% pine4.21
24919 byman 3 0 796K 1040K ttyin 0:00 0.04% 0.04% tcsh
24537 inch_hom 2 0 640K 872K sbwait 0:00 0.04% 0.04%httpd-1.3.3-us
So I'd kind of assume I wouldn't see a radical difference between a
machine with 500 idle processes and one that's running 300 and is in
active use...
So if anyone even has a similarly configured box, I'd love to hear from
you. I feel something is wrong here, but I can't find it...
Thanks,
Charles
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.00.10001181137380.10422-100000>