| An old mail message on the subject, "Why don't we need a subnet mask?"
From: LEVERS::KRISHNA "Krishna Narayanaswamy TAY2-2/B4 227-3572" 10-FEB-1993 11:02:48.34
To: LEVERS::GALVIN
CC: ANIL,QUIVER::GALLAGHER,HISCOCK,KRISHNA
Subj: RE: Subnet mask issue...Can you help ?? THANKS!
Cheryl,
I have given below an explanation as to why we really don't need subnet
masks on our SNMP agents. I have taken the scenario in which the SNMP agent is
on one subnet and the NMS is on a different subnet and the two are connected
through routers. The only need for subnet mask arises when you need to talk to
NMS's which are not on your subnet (i.e you are separated from the NMS by
routers). Given an NMS's IP address you can use the subnet mask and figure out
if the NMS is on your own subnet or not. If the NMS is not on your own subnet
you should lookup your routing table to figure out which router you need to
send your message so that it reaches the NMS. But our IP satck design
circumvents this problem as I have explained below
+-------+ Subnet 1 +********+ Subnet 2 +#######+
| Agent |-----------| Router |-----------| NMS |
+-------+ +********+ +#######+
1) SNMP request when received by the Agent has the following details.
Dest DataLink Address = Agent's Ethernet Address
Src Datalink Address = Router's port 1 Ethernet address
Dest IP address = Agent's IP address
Src IP address = NMS's IP address
When the agent processes the request it updates its ARP cache to contain the
following:
IP Address Hardware Address Port #
---------- ---------------- ------
NMS's IP Address Router's port 1 ethernet address 1
So when it comes time to reply to the SNMP request, the ARP cache is looked up
and the message is sent out to the Router and it then forwards the reply to
the NMS.
The only sticky case comes when there is more than one router through which you
can reach the NMS and the router chosen when the request came in is not the
one which provides an optimal path for the reply. In this case we will send the
message to the router through which we received the request and it is the rtr's
responsibilty to notify us through an ICMP redirect message telling us
in future to use a different router to talk to the NMS, but the router is still
required to forward this message to the NMS though it is not the optimal path.
Once we receive the ICMP redirect message, we register it in our ARP cache and
use it in future to talk to the NMS, until the ARP entry ages out. So we really
don't need subnet masks to talk to NMS's which are not in our own subnet.
I have used HP Openview, NetLabs Netmanager in the UNH interoperability lab to
manage our DECbridges through SNMP and in this case the NMS's spoke through
routers to our bridges, so I know that it works.
As far as the other question :
2) Can you define a default router/gateway on the DECagent 90?
I know on the Bridge/Concentrator you can set it. Not sure about DECagent 90.
Krishna
|
| Default gateways are needed only for unsolicited traffic originated
by a hub900 module. The only example of such traffic is SNMP traps.
So you don't even need to configure a default gateway before starting
to use IP-based protocols including SNMP (responses), TFTP, pings.
(As an aside: SNMP traps are unacknowledged, so not guaranteed to
reach the manager.. as such, they are considered of limited use.
Polling is the norm for reliable monitoring in SNMP.)
Anil
|