🚀 go-pugleaf

In article <1992Dec11.232959.18819@kakwa.ucs.ualberta.ca>,
Bob Engley <Bob_Engley@quick.ucs.ualberta.ca> writes:

    >We would like to monitor our servers to get an idea
    >of the usage, to answer the inevitable questions:
    >Who is using then? Are they reaching capacity etc?

    >Is SNMP the way to do resource monitoring? Just UNIX for now!
    >Has anyone run across an SNMP MIB that defines
    >system resources - things like CPU usage, Disk utilization etc. 
    >How about user resource utilization?
    >Of course the next question is: What about an agent that
    >works with the "resource" MIB?

Let's go back to the basics: I recall the first RFCs about SNMP and its
predecesors. At that time, The Internet, with it's zillions of hosts, 
routes and IP interfaces was becoming unmanageable. The original motivation
for SNMP was network management, and specifically the management of TCP/IP
specific resources of a huge network. I'm not suggesting SNMP is forever 
confined to TCP/IP resources. Some manufacturers are defining MIBS for 
things which are already far from the original TCP/IP kind of resources. 

The problem comes when an administrator of a medium-sized Unix network
(20 hosts) is told that "SNMP is the standard for Unix management". Once 
the LAN is installed, hardware malfunctions are not so frequent. Much 
more often, the administrator is being faced with problems which are
not related to TCP/IP resources:
    - new users have to be added, others have to be removed.
        In a commercial Unix environment, there is typically some data
        base application, which uses its own UserIds.
    - disks or file systems run out of space.
    - some CPUs get overloaded, while others are idle. 

Perhaps I shouldn't explain here the problem domain of System Management.
We know it is more than pure Network Management. 

I recall I've seen a product named DDS-Patrol some time ago. It was doing
resource monitoring in a moderately sized Unix LAN. I think the basic 
concept was to periodically fork off things like 'ps' or 'df', although 
most of the data was obtained straight from /dev/kmem. Somehow, it was 
also polling SNMP resources, and it could be customized to execute any 
other resource-monitoring commands.

On the LAN, there were DDS-Patrol agents on each host to be managed. 
On a central monitor, a Motif window came up, showing disks, CPUs and 
other such 'things'. I remember drilling down to increasingly lower levels
of granularity, and I could define threshould values for parameters. When a 
disk got full, an icon turned red, and there was even a synthetized 
voice announcement. 

The people who showed me this is are at <mpicard@au.oracle.com>. If you are 
interested , I could look up their address or phone in US (it's not 
Oracle Corporation).

        Andrew