High CPU system usage

[Glenn Steen]

    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "US-ASCII" character set.  ]
    [ Some characters may be displayed incorrectly. ]

On 24/08/05, Ugo Bellavance <ugob at camo-route.com> wrote:
> Glenn Steen wrote:
> > On 24/08/05, Ugo Bellavance <ugob at camo-route.com> wrote:
> >
> >>Glenn Steen wrote:
> >>
> >>>On 23/08/05, Ugo Bellavance <ugob at camo-route.com> wrote:
> >>>(snip)
> 
> <snip>
> 
> >
> > Not really, although this should do... I was thinking of
> >
> > 00:01:01          DEV       tps  rd_sec/s  wr_sec/s
> > 00:11:01       dev3-0      3.24      1.93    112.65
> 
> Here it is:
> 
(snippety-snip)
> 11:00:01       dev8-0    232.24     97.07   2778.52
> 
> # mount
> /dev/sda5 on / type ext3 (rw)
> none on /proc type proc (rw)
> none on /dev/pts type devpts (rw,gid=5,mode=620)
> usbdevfs on /proc/bus/usb type usbdevfs (rw)
> /dev/sda1 on /boot type ext3 (rw)
> none on /dev/shm type tmpfs (rw)
> /dev/sda3 on /tmp type ext3 (rw)
> /dev/sda6 on /var type ext3 (rw)
> none on /var/spool/MailScanner/incoming type tmpfs (rw)

Perfect. And this concurs with what we already saw in the "old style"
listing... That is: a bit of write activity, but nothing one should
(need to) write home about:-). Of course, you might be having "hidden
peaks" that are "killing you in the short term", so when you see one
next time, try capture some samples with a shorter interval (I'd
perhaps use iostat for that).

> > ....
> > which perhaps works better in 1.6 anyway, so...
> > You seem to have a "write peak" at ~ 1.36 MiB/s, which shouldn't be
> > partcularily close to your theoretical max, ergo not likely to be
> > close to your real max (whatever that is:-).
> 
> I get this from hdparm, while the server has a load of 2.5 (not idle)
> 
> /dev/sda:
>  Timing buffer-cache reads:   3348 MB in  2.00 seconds = 1674.00 MB/sec
>  Timing buffered disk reads:  168 MB in  3.01 seconds =  55.81 MB/sec
> 
Nice.
Assuming abysmal write performance of ~ 5MiB/s you'd still not be
pushing the limits.

> >>I can't really think of a way to determine this 'measurable max'.
> >>HDparm?  Analysis of a sar report while testing hard drive?
> >
> >
> > More or less so, yes. Also try using iostat.
> > Perhaps add in Bonnie there too
> > (http://www.garloff.de/kurt/linux/bonnie/)... and perhaps eye through
> > http://www.netlib.org/performance/papers/bonnie/ ... disk and fs
> > performance measuring isn't easy;-).
> 
> I am affraid of installing bonnie, as it screws Mondorescue, which I use
> for backups :(.

? Install ??
Just grab the source, compile it, run a test through, delete it ...
Why would Mondorescue take issue with that?
Oh well, just don't use it then, we're fairly sure it's perhaps,
probably, not that likely disk IO anyway:-).


> > <thinking_out_load>
(snip)
> > What fs do you use? Perhaps an issue with that...
> 
> ext3, might be worthwhile trying without journalling.  We have an UPS
> and the server is controlled by Apcupsd.

Perhaps. And perhaps without atime updating, if you don't do that already.

> > Or perhaps some really crummy NIC driver and a spell of network
> > congestion driving the CPU nuts? It has been known to happen (mostly
> > back in the dark ages, but still:-).
> 
> Stock driver in RHEL, e1000 driver, Dell poweredge server, intel pro
> 1000 NIC.

Should be OK... Then again, I've only got Broadcoms, so I wouldn't
know for sure:-).

> > Oh well, 'nuff rambling
> > </thinking_out_load>
> 
> Thanks a lot,
> 
> --
> Ugo

Glad to help.

-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at jiscmail.ac.uk with the words:
'leave mailscanner' in the body of the email.
Before posting, read the Wiki (http://wiki.mailscanner.info/) and
the archives (http://www.jiscmail.ac.uk/lists/mailscanner.html).

Support MailScanner development - buy the book off the website!