System information

ManualsBrandsRed Hat ManualsComputer equipmentNETSCAPE ENTREPRISE SERVER 6.1 - 04-2002 ADMINISTRATOR

Collection Policy

Create a CSV with interface types and node types and some flags for defining the collection policy for devices.

ifType device_type role

default any any

ethernet switch any

atm any any

summarystats

I would actually like to rewrite the summary stats routine to return a "summarystats hash".

Interface Stats

For each interface which supports it collect in/out frames, broadcast, unicast. This would allow determination of average packet size on link and on network. It

would also allow thresholds to be set for percentage of broadcast traffic on network.

More Notes:

An excellent idea, this would be very good information, this is already

on the todo list but there are few "dependancies":

1. The poller should be enhanced first to be multi-threaded which will

enable improved performance for SNMP and increase the number of mibs

polled per second per node. I think this is the next major improvement

required. That along with making NMIS a daemon which go hand in hand.

2. I would also like to support packet stats, ie pkt in and out, as well

as unicast and non unicast. There are different error stats per

interface type.

3. Not all interfaces support full if-mib stats, ie frame relay

subinterfaces on Cisco devices only support ifInOctets and ifOutOctets,

all other stats are in the Frame Relay mibs. This is also true for

other interface types like ATM. So the solution is to keep the extended

stats in a different RRD for each interface, I would envisage that there

would be the current interface, extra bit (pkts, errors, etc), specific

(frame-relay, ATM, etherlike, etc), so one interface may have 1

interface RRD while another my have 3. This could be implemented in

phases too I suppose.

4. Yes putting error rates in the node health metric would be excellent.

5. Other metrics could be calculated for nodes and the network:

* like per interface/link average packet size;

* entire network average packet size;

* error rates per interface/network (this equates to the much heard

never seen "accuracy" metric);

* lots of others that we might think of later.

6. Thresholding on errors should be done (easy).

NMIS - Network Management Information System http://www.sins.com.au/nmis/

27 of 43 15/04/2002 3:59 PM