Machine slow

[Steen, Glenn]

Julian Field wrote:
> On 18 May 2005, at 15:48, Steen, Glenn wrote:
> 
>> Billy A. Pumphrey wrote:
>> 
>>> Quick verification on load average if I may.
>>> 
>>> I do not understand how those numbers work.  They appear not to be
>>> percent usage. 
>>> 
>>> How do the load average numbers work?
>>> On my mailwatch, my usually is around .60-1.4
>>> 
>>> Billy Pumphrey
>>> IT Manager
>>> Wooden & McLaughlin
>>> 
>>> 
>> (snip)
>> 
>> The "load factors" are perhaps the most abused "performance
>> statistics" around... They're just the CPU run queue size (+ running
>> jobs) averages calculated for 1, 5 and 15 minutes. Nice to know and
>> a quick indicator, but nothing else.
>> 
>> For example: on a one CPU system a load of 2 might be less than
>> desirable, while on a system with 4 CPUs it shows two CPUs idling
>> away... So one needs weigh the system as a whole when determining if
>> a certain 
>> load is
>> fine or not.
> 
> It's not as simple as that. Jobs can be in the run queue if they are
> waiting for disk or network response. So if you have 10 processes all
> waiting to do DNS lookup, for example, then you will have a load
> greater than 10, but totally idle CPU(s).
> 
> This is why a busy MailScanner having a load of up to about 15 is
> nothing to worry about. It merely means there are 15 processes
> waiting for any of
>      (a) CPU time
>      (b) network response
>      (c) disk i/o.
> And that is a very simple view of it.
> 
> When the figure is over 1, it really doesn't tell you very much of
> any use at all. And all it tells you when it is less than 1 is that
> there is some time when your system is not doing anything.

It was just an example Julian. To be more precise the "unconditional 1"
is added per process in state D (non-interruptible wait state) which
_usually_ means some form of IO wait, but can be other things... As
you already know, of course:-).

Sole purpose of the example was to make Billy go buy not only your
MS book, but also the (in my view) excellent swordfish book too.
It's a friend to bring to the WC, trainride and nightstand....:-):-)
Your (as usual) excellent explanation should bring the point over.

Cheers
-- Glenn

------------------------ 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!