[ previous ] [ next ] [ threads ]
 From:  "Rick Ruggiero" <rick at scorpion dot co dot th>
 To:  "'M0n0wall'" <m0n0wall at lists dot m0n0 dot ch>
 Subject:  RE: [m0n0wall] about the load averages: 2.11, 0.07, 0.05
 Date:  Thu, 8 Apr 2004 12:24:26 +0700
> -----Original Message-----
> From: Adam Nellemann [mailto:adam at nellemann dot nu] 
> Sent: Thursday, 8 April 2004 11:45 AM
> To: rick at scorpion dot co dot th; m0n0wall at lists dot m0n0 dot ch
> Subject: Re: [m0n0wall] about the load averages: 2.11, 0.07, 0.05
> Hi Rick,
> Rick Ruggiero wrote:
> > It means, the following:
> > 
> > load averages: 2.11% (Avg. process queue last minute), 0.07% (Avg.
> > process queue last 5 minutes), 0.05% (avg process queue last 15 
> > minutes).
> > 
> > I don't have the m0n0wall running properly yet, but generally
> > uptime is the same on all UNIX platforms.
> I think it was previously established that the three numbers 
> corrospond to the average percentage for the last 1, 5 and 15 
> minutes respectivly, but I do find it somewhat illuminating 
> that these numbers are related to the process queue.
> However, there still remain some things to be explained:
> - How to understand the term "process queue" precisely (this 
> might be simple stuff for a *NIX guru, whereas I, not knowing 
> much about *NIX, can only guess at the exact meaning).

The load average numbers give the number of jobs in the run queue
(actively being processed)

> - What the number (percentage) indicates, the average 
> wait-time in the queue, the average queue length or something 
> else entirely?

No, unfortunately you cannot use this number to predict CPU,IO latency,
etc.  Each process has a discrete impact many other system metrics which
does affect overall system performance (and the values being displayed).

> - What is this percentage relative to? (That is, what is 
> 100.00%, and can it become higher than that?)

Yes, it is relative to 100%; and no, it does not display greater than
100% (well from my experience at least).

> - And finally, how this "average process queue" percentage 
> relates to such things as CPU load, latency and general 
> system performance?

This is all dependant on the processes currently in the run queue.  Some
processes may be blocked on I/O which often has a severe impact on this
value.  However, generally speaking (and I am opening myself to abuse
here) any time your process average for 1-5 minutes is over 6, you will
find that process spawning is dramatically affected.

I do not have a magic answer on this, or a formula where you can start
doing transaction prediction modelling.

> Unless one has some idea of the answers to the above 
> questions, the numbers have little meaning and will be of 
> even less use. If, on the other hand, you know how these 
> numbers are arrived at, you could use them, among many 
> things, to see if it is time to start looking for faster 
> hardware. Also, we might perhaps finally get some 
> authoritative statements about how much bandwidth m0n0wall 
> can handle on particular platforms (such as the Soekris boards).

I think if you look at my statement above, you may have a glimpse as to
my approach to this question;  If my CPU is regularly demonstrating a
process queue of 6 or greater, I would look seriously into a processor
upgrade, however if my load averages were "load averages: 2.11, 0.07,
0.05", I'd be very happy.

I hope I was a little clearer with this response.