Monitor by Anvil Designs

[John C. Tompkins]

In article <1042 at anvil.oz>, mob at anvil.oz (Michael O'Brien) writes:

> In August 1989,  Anvil Designs Pty. Ltd. released Monitor a Unix performance
> and tuning software package.  This package is not to be confused with a 
> piece of "netware" by the same name that was released onto the net about the
> same time.  That monitor package was like a visual ps command and its focus
> was to visually give per-process feedback.

Hopefully it did that, but it also displays the sysinfo data on overall
system performance which is about all that is available on XENIX-286. There
is no "vmstat" information or disk access times and so on available. There
is data on CPU usage, swapping, free memory, system calls, CPU hogs, disk
IO ...

> MONITOR by Anvil Designs Pty. Ltd. is quite different in approach.  MONITOR
> gives complete performance data about the whole system.  All data is 
> presented graphically with bar graphs and charts.  MONITOR goes far beyond
> the Unix sar or ps command.  Major subsystems within unix are examined.  
> Disk fragmentation, disk activity, and per filesystem accesses can be 
> compared.

I don't quite agree with "quite different approach" but it does sound nifty.
I supply all the data in bar graphs too! I think the disk fragmentation data 
might not be appropriate for continuous, realtime monitoring. It sounds
something like a prettied up fsanalyze.

> Cpu utilization for processes and within the Operating system can be 
> isolated.  The percentage of cpu time  spent servicing floating point
> emulation, performing interrupts, swapping, in process control and in 
> many other kernel subsystems is displayed.  Indeed, these subsystems
> can be altered by the end user to profile inside Unix as required.  This

So is this something beyond the UNIX/386 PROFILER (ADM) commands? The cpu time
spent "servicing floating point emulation" would be a neat trick. I've been
trying to figure that one myself, haven't got it yet. Swapping and
interrupts are easy, although it appears that XENIX doesn't use the sysinfo
supplied serial IO interrupt counters for anything. Are these interrupts
something beyond this?

> Memory usage for each process is displayed, and the working set of memory
> resident in memory and being heavily accessed is also available.  This 
> allows exact computation of how much memory is used by an application, and
> how many more users can be added to the system.

Displaying the memory usage ain't so tough! :-). I don't want to be picky
but obviously there is more to deciding if more users may be added than
just how much memory is left (which this implies to me) Does this include
the virtual memory activity as well?

> MONITOR can operate in interactive mode or in logging mode where all data is
> stored in a log file.  This log file can be replayed later to view the
> performance data.  The replay speed can be controlled to give a fast forward
> effect.

OK so I don't do that, I wrote my monitor just to satisfy my needs. If I want
logging and long term averaging I just use sadc/sar instead.

> MONITOR uses a device driver to gather the statistics necessary. 

Oh yeah?, an Anvil supplied driver, the /dev/prf driver for PROFILER (ADM) or
what? A driver just for this purpose would be nice, solve some problems and
perhaps increase the monitoring efficiency (a little). But all in all I dont 
know that its worth the effort. After all /dev/mem and /dev/kmem are drivers
too!

> MONITOR costs $350 US and comes complete with a 200 page manual.  This manual
> comprehensively deals with all unix tuning and performance issues and is
> also a collection of much of the unix tuning folklore.
> 
> Michael O'Brien
> Director Development
> Anvil Designs Pty. Ltd.

How about network services and streams monitoring as well? I still think 
monitoring the print queues (an improved lpstat) and serial ports (siomon is 
handy but needs to be expanded and integrated into a larger package) would be 
handy. Does the Anvil monitor do this?

What bothers me is why should we have to pay money for a system administration
tool that should come as a normal part of the operation system? Why does
SCO put the sysinfo data in XENIX-286 and then not tell anyone about it much
less have utilities available to access it? This is how the nonsense that
UNIX is hard to use got started and continues getting passed around. UNIX is
not hard to use, the OS vendors just don't go to the trouble of making it
easy. It seems to be the old "well this is the way its always been" attitude.
I give away my monitor utility just to show it doesn't have to be that way.

Am I the only one who feels this way? I better not be if UNIX is to compete
with OS/2 on the desktop or VMS on minis. Dollars/mip and development
environmets are great but I want even more! :-)

John C. Tompkins
uokmax!occrsh!jct!jct