[CLUE-Tech] top and system load

[Adam Bultman]

The "Load Average" is kind of a thumbnail idea of what load is.  A system
running say, dnetc will have a load of 1, I think . At least, mine did.
It's a kludge of the amount of processes vs. the amount of high CPU time
processes, etc.  The mail server at my college would start freaking out at
around 8-11 during heavy usage (chapel break). Regardless:  I had two
servers (unused at the moment)  with loads of 80 (seriously) because they
both at processes that were spawning their own processes that were
crashed, and taking all of the CPU.  It was like this for I don't know how
long.  Weeks. I had to kill -9 many, many processes to get it back down.
The systems still responded in a timely manner, which is odd.  So
therefore, it's not perfect. My system running dnetc with a load of 1
responds crappier than when I'm compiling, and it's higher than that.

-- 
Adam Bultman
adam at glaven.org
[ http://www.glaven.org ]


On 10 May 2002, Ed Hill wrote:

> On Fri, 2002-05-10 at 20:27, Jason Friedman wrote:
> > I've printed below the output from the top command:
> >
> >    2:25am  up 27 days, 13:12,  7 users,  load average: 2.00, 2.00, 2.00
> > 93 processes: 85 sleeping, 7 running, 0 zombie, 1 stopped
> > CPU states: 42.8% user, 57.1% system,  0.0% nice,  0.0% idle
> > Mem:   383960K av,  380748K used,    3212K free,    1756K shrd,
> > 15532K buff
> > Swap:  265032K av,      20K used,  265012K free
> > 208416K cached
> >
> > I'm not familiar with the meaning of these numbers.  Is my system being
> > barely touched, overwhelmed, or something in the middle?
>
>
> Deciphering top/ps output should be a FAQ item for all LUGs...  ;-)
>
> Basically, heres what it says:
>
>   - Your CPU is 100% busy with 1/2 the load due to user
>     processes and 1/2 due to kernel-space code.  You should
>     look to see if you have a "hung" process or something else
>     going on.  Take a look at which program(s) are using up
>     the CPU cycles and take it from there.
>
>   - Your memory is nearly 100% used but this is very *NORMAL*
>     for almost *ANY* load (even very light loads) since the
>     Linux virtual memory subsystem (VM) will aggressively use
>     memory as a cache for disk access.  A better indicator of
>     a memory-congested system is to see the VM digging deeply
>     into the swap space.  Which it is not doing in your case.
>     But note that even if your system is using a lot of swap
>     space then it might just be a lot of sleeping processes.
>     Again, look into the per-process memory usage as reported
>     by top and ps.
>
> hth,
> Ed
>
> ps - I took a look at the Process-Monitor-HOWTO.html at linuxdoc and was
> disappointed by how little info it provides.  I think someone should
> flesh it out a bit more...
>
>
>