hpr-knowledge-base/hpr_transcripts/hpr1012.txt

Episode: 1012
Title: HPR1012: LiTS 009: w command and linux load averages
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr1012/hpr1012.mp3
Transcribed: 2025-10-17 17:22:39

---

Welcome to Linux in the Shell episode 9, W command.
My name is Dan Washkel.
I'll be your host, and before I begin, I'd like to.
Thanks.
Thank you for your time, and I'd like to remind you that if you have not already visited
the website, please do so at your earliest convenience for the full right up of the W command
and the corresponding example video.
Alright, the W command.
What the W command does is essentially displays information about the users that are currently
logged on to the machine and their processes.
It displays it in two separate sections.
The first section is the header information, and that tells general system information.
The current time, it also shows the up time, how many users are logged in, and then the
CPU load average.
Now, I'm going to come back to the CPU load average because that's very important, and
it's pretty much the reason why I chose the W command for this episode.
So in conjunction with the after the header information, you have the body of the W command,
which is a list of all the users that are logged into the system, and some corresponding
information.
Now, each line on the body of the W command lists the user where they're logged in from
their TTY, whether they're on a TTY 1, 2, or 3, or on a PTS.
There's also the time that they are logged in at, and how long they have been idle, if
they are idle, in minutes or hours and seconds.
Then there's a JCPU, which means the time that is used by all processes attached to the
TTY that are logged into, and then there's PCPU, which is just a time used by the current
processes that are executed by that user at that moment, and then a line that says what
process that the user is currently running, or what is running at the moment, or the
last process.
Now the W command takes a few options, there's a dash H or dash, dash, no, dash header, which
omits printing the header, then there's the dash, ass, or the dash, dash, short, which
uses the short format, and it does not print the log in time, the JCPU or the PCPU times,
and then there's the dash F or the dash, dash, from, which toggles printing the from remote
host name field.
Now by default, chances are your version of W is compiled without showing the from, which
is where they're logged in from, and if you choose the dash F option when you execute
it, it will put in an extra column between TTY and log in, which will say from where.
So if they're logged into the local host, it will show, chances are show nothing, or if
they're in an accession and they have like a terminal window open, it will show their accession
and the display that they're logged in from, so more than likely it'll be 0.0 that's
coming from.
If it's an remote session, it would show from the host that they're logged in from either
the host name or the IP address.
Now there's the old style dash, or dash dash, old dash style output, which prints blank
spaces for idle times less than one minute.
And then finally, you can choose to display just specific information on a single user
by typing W and the user name, and it will show only that user that's logged in and
the corresponding information.
So to step back a second and just understand a little difference between JCPU and PCPU,
what JCPU offers is the time used by all the processes attached to the TTY.
So any process that's running currently by that TTY might not be the current process
that is being displayed, but any processes that are running for that TTY will be, it
will display the CPU time for that JCPU session.
So for instance, I have a process running called XINIT on my X session that I started X
with, and that's showing that it's been idle for 17 hours and 21 minutes.
Total JCPU, the CPU time that's been utilized by it is 7 minutes and 46 seconds, so that's
the total of all the running processes of CPU time.
And then PCPU, which is the time used by the current process named in the what field.
So XINIT, like right now, is pretty much lying dormant, so it's not using any PCPU cycles
at the moment.
So that's what that basically displays.
So that's the difference between these two, but the big thing I wanted to talk about
was load average.
And load average is not only shown by the W command, but it's also shown by a very helpful
command called top.
And I wanted to cover load average in W because it gives me a little bit of time to do that,
whereas trying to cover it in top, there's a whole lot of stuff to cover in top.
It just makes it a little more convenient.
So a little foreshadowing, expect a show about top in the very near future, if not maybe
the next show, haven't decided yet.
But what load average is, you will see three values up there.
It's the CPU load average of the past minute, the past five minutes, and the past 15 minutes.
And that's not a percentage of CPU usage time.
It's load average over a minute.
Now, if you look, if you think about CPU usage, and you can say, well, CPU has one state.
It's either being used or not, so it's either zero or a hundred percent usage.
But what load average means over the period of one minute, five minutes, and fifteen minutes,
it's how much the processor's being used and how many processes are waiting for the CPU
over that past minute so that it to be able to be processed by the CPU.
If you read a lot about this, the common analogy is a highway or a traffic pattern, where
the CPU is a highway and an on ramp is the processes waiting to be processed by the CPU.
So if your on ramp is empty, that means the CPU is handling the processes as they come
in without any overhead.
So it's running just fine, the system is not under any significant load.
But if all the processes that come in are in the highway on the highway lanes, there's
nobody waiting, but the highways filled up with all your processes, that means the CPU
is running at full bore.
And that's not necessarily a bad thing, so to speak.
It just means that your CPU is perfectly suited to handle all the processes that are coming
in and, you know, relatively no more.
Nobody's waiting, but there's no, there's, it's running full bore.
Okay, so what you would see in that case under load average is a one for that process.
So if the CPU is running load averages of one, that means that CPU is handling all the
processes, but, and there are no processes waiting, but it's, it's constantly running.
Then anything above one means that processes are still starting to back up on that on ramp.
So the highway is full, and the CPU is processing all this stuff, and there's a backup on
the own ramp, so they're starting to get a little traffic backup.
And that's when your load average is going to start to increase beyond one.
To look at it a different way, you can look at it like in a percentage that if, if you're,
let's say you see your load average is 0.50, that means your system, your CPU is capable
of handling twice the amount of load that you currently have on it.
So you're only using, utilizing about half the CPU processing power.
Now if it is at one, that means you're utilizing 100% of the CPU processing power.
And if the load average goes to say two, that means you're using twice as much processing
power.
You'd need twice as much processing power to handle the load.
So you really need to think about your system, and now those are rough estimates, and
it's not exactly a one to one ratio, but that is kind of a way to look at it.
So what you have to bear in mind, not just these values, but the values over the periods
of time from one to five and 15 minutes.
Because just looking at one value is not indicative of the common load on the system.
And be aware of what you're doing on the system.
If you're running your system and not doing anything on it, or just doing normal activities,
and you pop over there to look at your load averages, and you see numbers between 50 and
1.2, you might want to start observing what's going on.
Whereas if you're under like 0.70, that's a pretty good indication that your system's
not being taxed at all.
Now if you're normally under 0.70, and you start compiling a program and you're playing
a game, and for a couple of, you know, for the minute load average, you spike up to maybe
1.5 or two, but then it drops back down.
You don't notice that same level of spike in the five and 15 minutes.
You're not necessarily taxing your system.
It just means that when you're, you know, that application started up, it put a load on
your system, and things went back to normal, and it was able to handle things on a regular
basis that you were throwing at it.
But be aware of those numbers.
If you're wanting to observe your system performance and determine whether what you're
doing with your system, you have at least the processing power to perform whatever it
is that you need to do.
Now it's not just a one-to-one ratio here of percentages.
Well, be in mind that if you are on a multi-processor or on a dual-course or quad-core multi-course
system, the load average is not just one and above is a problem.
It's essentially a value of 1 per processor.
So if you're on a dual-course system like this laptop I have right here is, and this
is currently showing me a load average of 0.93 for the past minute, 0.87 for the past
five minutes and 0.80 for the past 15 minutes.
Now I had said you probably want to be careful if your load average goes above 0.7 and
start observing what's going on.
So this is a dual-course system.
So there's essentially two processors in here.
And one, two processors means you got to double those load average values.
So a value of two in a dual-course processor system means that both processors are being
utilized at a hundred percent.
Instead of it being one, it's a value of two.
So you would want to be alarmed at what's going on in your system when you've reached
roughly something like 1.7 instead of 0.7.
In a quad-core processing system, you have four.
So a value of two or three is not necessarily a significant load on the system.
But when you get to value of four, that means all four cores are being utilized to a hundred
percent of their capacity.
And it really can't take on any more load without there being processes having to wait.
So again, load averages scale up with the number of CPUs or cores that you have on the
system.
Adjust accordingly.
So as an example, at work, we were load testing a new application and whether it was the
system that we had for it was adequate and whether there were any runaway processes, memory
problems or anything going on with this application, when we fired it up, it skyrocketed showing
a load average between 5 and 13.
Now this was on a 8-processor system.
So between 5 and 7 is not necessarily bad, but you might want to keep an eye on that.
If the normal load is between 5 and 7, if you would get an excess load on that system,
you probably tax the system and that's what happened when it went to 13 and above.
That means that that system, with load that we were putting on that, was exceeding the
full load of the system by almost twice as much.
So it was one, almost one and a half times beyond the load capacity system and as it grew,
if it would have, when it reached 16, that means twice the load that the CPU can process
was being put on the system.
So that's load average and kind of in a nutshell what it means.
If you're still a little confused about that, I have some great, I won't say the right
up I did on it, it's great, but there's some really great resources in the bibliography
on the website that really break this down and even into the equations and stuff like
that from the Linux journal side of things, it was really good article.
So that's it right there.
So just remember, W command shows some snapshot of the system in the header and then it shows
all the people that are logged in into the system and what processes they're running
and how much they're utilizing the CPU at that time.
Load average is really what I want you to take away from this because when we cover top,
that's going to be an important metric.
My name is Dan, thank you very much.
This is Linux in a shell.
Again, I want to thank Hacker Public Radio and I hope to see you at the next episode.
Have a good one.
Thank you for listening to Hacker Public Radio, HPR is sponsored by Kero.net, so head
on over to caro.net for all your hosting needs.