Interpreting HPUX "top" command
Vicente E. Zamora
I want to know whether my system (HP9000-807s) is overloaded or not and I
found the top command, i get a lot of information, I just don't know what
to make out of some figures.
>> System: deimos Thu Jun 20 19:37:19 1996
>> Load averages: 1.78, 1.39, 1.23
>> 82 processes: 80 sleeping, 2 waiting
>> Cpu states:
>> CPU LOAD USER NICE SYS IDLE UNK5 UNK6 INTR SSYS
>> 1 1.78 33.3% 0.0% 66.7% 0.0% 0.0% 0.0% 0.0% 0.0%
System, OK,
Load averages, I get it.
Number of process, I can understand that.
But CPU LOAD=1.78... is it percentage?
what about USER, NICE, SYS, etc,etc.
Can anybody tell me about those?
Where can I get some benchmarking software for this machine, I want to
justify to my boss that we need to upgrade this system, I have 200+
e-mail users in it and a Web Server running (http://www.uaslp.mx/).
--
***************************************************************
* Vicente E. Zamora zam...@deimos.tc.uaslp.mx *
* Depto. de Telecomunicaciones +52 (48) 11-8321 *
* Div. de Informatica, U A S L P +52 (48) 26-1309 *
* San Luis Potosi, SLP, MEXICO http://www.uaslp.mx/~zamora/ *
********************************************************************
Stephen J. Weihman
In article <31CA0C...@deimos.tc.uaslp.mx>,
"Vicente E. Zamora" <zam...@deimos.tc.uaslp.mx> writes:
>I want to know whether my system (HP9000-807s) is overloaded or not and I
>found the top command, i get a lot of information, I just don't know what
>to make out of some figures.
>
>>> System: deimos Thu Jun 20 19:37:19 1996
>>> Load averages: 1.78, 1.39, 1.23
>>> 82 processes: 80 sleeping, 2 waiting
>>> Cpu states:
>>> CPU LOAD USER NICE SYS IDLE UNK5 UNK6 INTR SSYS
>>> 1 1.78 33.3% 0.0% 66.7% 0.0% 0.0% 0.0% 0.0% 0.0%
>
>System, OK,
>Load averages, I get it.
>Number of process, I can understand that.
>But CPU LOAD=1.78... is it percentage?
>what about USER, NICE, SYS, etc,etc.
>
>Can anybody tell me about those?
A "man top" will give you all of the info... Look in the "System Data"
section.
--
How do I set my laser printer to "stun?"
--
-------------------------------------------------------
-- Stephen J. Weihman -*- GTE Data Services --
-------------------------------------------------------
-- The opinions expressed herein are entirely my --
-- own, and are not necessarily those of GTE. --
-------------------------------------------------------
Chris Riney
Not what you want to hear, but RTFM top(1).
On a friendlier note,
Load averages are a guestimate of how many processes are 'Runnable' (at any
given second) over the past 1, 5 and 15 minutes.
At the time this snapshot was taken you had 82 processes defined, 80 of which
where waiting for some event to happen, and 2 where actually doing something.
The columns user, nice, syst, and idle are percentages of the CPU's usage,
where:
user normal users/applications
nice user/applications that have been given a lower prioriity
sys Kernel interaction, such as I/O or other system calls
idle when the CPU isn't doing any thing at all!
From what you are showing the system is spending a heck of a lot of time
doing SYSTEM related operations (usually disk I/O, but could be a few other
things). Other tools that come with the system that would help are iostat(1),
vmstat(1), and sar(1). If you are in the market for puchasing, then hPUX has
GlancePlus!
Right now, I would suggest finding out what process is HOGGING the CPU, and
address that.
Vicente E. Zamora (zam...@deimos.tc.uaslp.mx) wrote:
: I want to know whether my system (HP9000-807s) is overloaded or not and I
: found the top command, i get a lot of information, I just don't know what
: to make out of some figures.
: >> System: deimos Thu Jun 20 19:37:19 1996
: >> Load averages: 1.78, 1.39, 1.23
: >> 82 processes: 80 sleeping, 2 waiting
: >> Cpu states:
: >> CPU LOAD USER NICE SYS IDLE UNK5 UNK6 INTR SSYS
: >> 1 1.78 33.3% 0.0% 66.7% 0.0% 0.0% 0.0% 0.0% 0.0%
: System, OK,
: Load averages, I get it.
: Number of process, I can understand that.
: But CPU LOAD=1.78... is it percentage?
: what about USER, NICE, SYS, etc,etc.
: Can anybody tell me about those?
: Where can I get some benchmarking software for this machine, I want to
: justify to my boss that we need to upgrade this system, I have 200+
: e-mail users in it and a Web Server running (http://www.uaslp.mx/).
: --
: ***************************************************************
: * Vicente E. Zamora zam...@deimos.tc.uaslp.mx *
: * Depto. de Telecomunicaciones +52 (48) 11-8321 *
: * Div. de Informatica, U A S L P +52 (48) 26-1309 *
: * San Luis Potosi, SLP, MEXICO http://www.uaslp.mx/~zamora/ *
: ********************************************************************
--
Chris Riney E-mail: cri...@tandy.com
Technical Services PostM...@tandy.com
Tandy Information Services Phone: (817) 878-0308
Vicente E. Zamora
Stephen J. Weihman wrote:
>
> A "man top" will give you all of the info... Look in the "System Data"
> section.
>
> --
> How do I set my laser printer to "stun?"
> --
> -------------------------------------------------------
> -- Stephen J. Weihman -*- GTE Data Services --
> -------------------------------------------------------
System Data
+ System name and current time.
+ Load averages in the last one, five, and fifteen
minutes.
+ Number of existing processes and the number of
processes in each state (sleeping, waiting,
running, starting, zombie, and stopped).
---> + Percentage of time spent in each of the processor
---> states (user, nice, system, idle, interrupt and
---> swapper) per processor on the system.
This is the output of "man top" and it says nothing that solves my
doubts.
Can anybody help me with the meanings of these?
What is nice? system? user? interrupt? swapper?
Vicente E. Zamora
Aaron R. Kulkis
news:31CA0C...@deimos.tc.uaslp.mx
zam...@deimos.tc.uaslp.mx (Vicente E. Zamora) wrote:
>I want to know whether my system (HP9000-807s) is overloaded or not and I
>found the top command, i get a lot of information, I just don't know what
>to make out of some figures.
>
>>> System: deimos Thu Jun 20 19:37:19 1996
>>> Load averages: 1.78, 1.39, 1.23
>>> 82 processes: 80 sleeping, 2 waiting
>>> Cpu states:
>>> CPU LOAD USER NICE SYS IDLE UNK5 UNK6 INTR SSYS
>>> 1 1.78 33.3% 0.0% 66.7% 0.0% 0.0% 0.0% 0.0% 0.0%
>
>System, OK,
>Load averages, I get it.
>Number of process, I can understand that.
>But CPU LOAD=1.78... is it percentage?
>what about USER, NICE, SYS, etc,etc.
>
>Can anybody tell me about those?
>
(typically you see three, as abouve: a 10 second average,
a 1 minute average, and a 5 min average)
On some machines, a load average of 2 is very noticable, on other
machines, you don't notice a slowdown until the loadaverage hits 30-60, so
interpretation of the load average is stat is very CPU dependant.
Also, Load average = Total Load / CPUs, so to find the total load of
the system, compute load average * CPUs.
the 2 lines following CPU States are statistics on a per/cpu basis
(this may be different with different vendors, but on HP, you get
individual stats...some give you individual stats, plus collective
stats on the last line for multi-cpu machies
Reading across
CPU Which cpu is the following set of stats for
LOAD = instantenous context-switches per second on this cpu
USER = % of time this CPU spent executing User code in the last time slice
SYS = % of time this CPU spent executing kernal codee in the last time slice
IDLE = 100% - USER - SYS
NICE = % of time was spent executing jobs that have been nice(1)'ed
>Where can I get some benchmarking software for this machine, I want to
>justify to my boss that we need to upgrade this system, I have 200+
>e-mail users in it and a Web Server running (http://www.uaslp.mx/).
benchmarch schmenchmarks. Benchmarks are for marketing new equipement,
not evaluating equipment that is already in place. The best benchmark
for installed equipment is this: "Does it perform the required tasks
in a timely manner?"
If you answer yes, then there is no need to upgrade.
If the answer is no, then find the bottle neck. HP has some sort of
diagnostic--I can't remember what it is. Call your dealer, he'll know,
because HP puts a 30-day license on it's performance meter to help you
do initial tuning (and hoping that you will buy a license for long
term use).
Anyways, HP's kernal-tuning tool will tell you where the bottlenecks are.
You might just need more memory.
--
Aaron R. Kulkis aku...@wae.gmpt.gmeds.com
Unix Systems Administrator
EDS/GM Powertrain
GM Warren Technical Center
Warren, Michigan, USA
---------------------------------------------------------------------
Have you smoke-tested your (pick one:cpu/hard-drive/power-supply) today?
---------------------------------------------------------------------
There are 2 Kinds of planes: Fighters, and targets.
There are 2 Kinds of boats: Submarines, and targets.
There are 2 Kinds of 4x4's: Jeeps and SOB's (Some other brand)
There are 2 Kinds of O/S's: Unix, and brain-farts.
---------------------------------------------------------------------
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GE(GAT)d+(-) s:- a30 C+++$:() UHI++++$>BL P!>+++ L++ !E- W+(-)$ N++ o K?
w--- !0 !M-(--) V-- PS+,-- PE++() Y+>++ PGP>++ !t !5 !X !R tv-->---
b++1/2 DI+++ D G e*>++++ h>--- r* y?
------END GEEK CODE BLOCK------
Aaron R. Kulkis
Jason K. Smith
You may want to try the Glance Plus product that HP markets. I
believe there is a free trial, and its output is quite a bit more
useful than top. I've found it to be an invaluable tool for
monitoring the performance of systems.
Jason Smith
===============================
Jason K. Smith
j...@hubcap.clemson.edu
===============================
Check out my homepage at http://www.eng.clemson.edu/~jks
Quentin Fennessy
In article <4qn6jh$d...@maverick.tad.eds.com>, Aaron R. Kulkis <akulkis> wrote:
[much stuff snipped]
>Load averages: context switches per second
> (typically you see three, as abouve: a 10 second average,
> a 1 minute average, and a 5 min average)
>On some machines, a load average of 2 is very noticable, on other
>machines, you don't notice a slowdown until the loadaverage hits 30-60, so
>interpretation of the load average is stat is very CPU dependant.
>Also, Load average = Total Load / CPUs, so to find the total load of
>the system, compute load average * CPUs.
Aaron-
My understanding of load average is the average length of the
run queue over the last 1, 5 and 15 minutes. The `length of the run queue'
is another term for `number of processes ready to run' or waiting
on the CPU.
For instance if I execute this command:
find / -print > /dev/null
it will probably not increase the load average much. Rather than waiting
on the CPU it will wait on the disk io (or network io). But if I execute
a program that generates pi to the nth digit it will probably be CPU bound
and this increase the load average.
The number of context switches is an interesting figure but if your
system is only doing 2 context switches per second it is not working very
hard!
Check out `System Performance Tuning' for a good reference on
the topic.
--
Quentin Fennessy AMD, Austin Texas
Rob Musquetier
Hi there,
Indeed, Glance Plus is quite handy but only usefull for about 3 month
after this trail periode it expires. You have to buy it if you like it.
Another fine product is PerfView. You can't try this application out first
but is it must have for every system administrator !
Greeting,
Rob Musquetier.
On dinsdag 25 juni 1996, Jason K. Smith wrote...
Stephen Hearn
: news:31CA0C...@deimos.tc.uaslp.mx
: zam...@deimos.tc.uaslp.mx (Vicente E. Zamora) wrote:
: >I want to know whether my system (HP9000-807s) is overloaded or not and I
: >found the top command, i get a lot of information, I just don't know what
: >to make out of some figures.
: >
: >>> System: deimos Thu Jun 20 19:37:19 1996
: >>> Load averages: 1.78, 1.39, 1.23
: >>> 82 processes: 80 sleeping, 2 waiting
: >>> Cpu states:
: >>> CPU LOAD USER NICE SYS IDLE UNK5 UNK6 INTR SSYS
: >>> 1 1.78 33.3% 0.0% 66.7% 0.0% 0.0% 0.0% 0.0% 0.0%
: >
: >System, OK,
: >Load averages, I get it.
: >Number of process, I can understand that.
: >But CPU LOAD=1.78... is it percentage?
: >what about USER, NICE, SYS, etc,etc.
: >
: >Can anybody tell me about those?
: Load averages: context switches per second
: (typically you see three, as abouve: a 10 second average,
: a 1 minute average, and a 5 min average)
I always thought load average was the average number of runnable processes
waiting to use the CPU within a given time frame???
steve
Doug Grumann
> I always thought load average was the average number of runnable processes
> waiting to use the CPU within a given time frame???
This is correct.. the load average can be impacted by processes waiting
for the CPU but also processes "short waited" on Disk I/O, so it is not
always indicative of a cpu bottleneck.
All the HP performance tools are available for 60-day trials, including
PerfView. Note that PerfView is an analysis tool that displays data
provided by the MeasureWare logfiles. Glance is a diagnostic tool.
HP GlancePlus, HP MeasureWare, and HP PerfView offer a full range of
resource and performance management capabilities.
To request FREE o Send an electronic mail (e-mail) message to:
60-day trial kit, perf...@mayfield.hp.com (Include your Tel. No.)
or for additional
product information o FAX your questions to the HP Hotline for
Performance Tools at FAX No: 1-415-691-5466 (U.S.)
o Call the HP Hotline for Performance Tools:
1-800-237-3990 - within the U.S. and Canada
1-415-691-3911 - outside the U.S. and Canada
Please call between 8:00 AM - 4:00 PM Pacific Time
Stephen Baynes
Stephen Hearn (she...@minotaur.labyrinth.net.au) wrote:
: Aaron R. Kulkis (akulkis) wrote:
: : Load averages: context switches per second
: : (typically you see three, as abouve: a 10 second average,
: : a 1 minute average, and a 5 min average)
: I always thought load average was the average number of runnable processes
: waiting to use the CPU within a given time frame???
It is. From the HPUX man page for uptime:
..., and the average
number of jobs in the run queue over the last 1, 5 and 15 minutes.
--
Stephen Baynes bay...@ukpsshp1.serigate.philips.nl
Philips Semiconductors Ltd
Southampton My views are my own.
United Kingdom
Are you using ISO8859-1? Do you see © as copyright, ÷ as division and ½ as 1/2?
Sam Nelson
In article <4ra9sa$s...@rosenews.rose.hp.com>, do...@hpptc80.rose.hp.com (Doug Grumann) writes:
> All the HP performance tools are available for 60-day trials, including
> PerfView. Note that PerfView is an analysis tool that displays data
> provided by the MeasureWare logfiles. Glance is a diagnostic tool.
>
> HP GlancePlus, HP MeasureWare, and HP PerfView offer a full range of
> resource and performance management capabilities.
>
CPU for fifteen CPUs? Almost GBP16K. Hell, if I had that sort of money for
systems-admin software I'd go out and buy a nice new fast workstation
regularly and solve my performance problems that way. If it was
network-licenced, I'd buy a couple of licences so's I could run it twice
simultaneously, but I'm told by my supplier that it isn't. So I use `dmon',
`yamm' or `top' for the moment.
--
Sam. Scunthorpe! (Insert bandwidth-wasting disclaimer here)
If all you have is AltaVista, everything looks like a web-search...