home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zephyr.ens.tek.com!shaman!thistle!stevet
- From: stevet@thistle.wv.tek.com (Steve Tarr;685-3528;63-356;;)
- Newsgroups: comp.protocols.snmp
- Subject: tcpConnEntry instance naming
- Message-ID: <1495@shaman.wv.tek.com>
- Date: 27 Jul 92 22:54:58 GMT
- Sender: nobody@shaman.wv.tek.com
- Reply-To: stevet@pogo.wv.tek.com
- Organization: Tektronix, Inc., GPID
- Lines: 44
-
- I'm trying to make sense of the last few paragraphs of Section 5.3.2
- "Problems with Naming" in _The Simple Book_:
-
- "Of course, it might be possible that there is no way to
- uniquely identify a particular entry in the table. ..."
-
- "A special "hack" can be used in this case. If multiple
- rows in a table have the same key, the agent marks one of
- these rows as the primary. The other rows with the same key
- are ... assigned an implementation-dependent, small, unique
- non-negative integer. To identify an instance of a secondary
- row, this integer is appended onto the instance identifier of
- the primary row."
-
- My (embedded) protocol stack maintains a separate connection endpoint for
- each possible TCP connection to a server. New connection requests are
- handled by finding the first entry in the table with the correct local
- address/port in state LISTEN, copying the remote address/port information
- from the packet into the entry, and setting the entry state to SYN-RECEIVED.
- What this means for SNMP is that there are often a number of connection
- endpoints with identical address/port information, i.e. local address
- 0.0.0.0, local port 8000, remote address 0.0.0.0, and remote port 0,
- all in state LISTEN. I'd like to make all these endpoints visible for
- monitoring, since the number of "available" endpoints is an important
- resource limitation.
-
- Assuming three such entries, the text would seem to imply that the
- corresponding instances of tcpConnState would be named:
-
- tcpConnState.0.0.0.0.8000.0.0.0.0.0
- tcpConnState.0.0.0.0.8000.0.0.0.0.0.1
- tcpConnState.0.0.0.0.8000.0.0.0.0.0.2
-
- This is admittedly somewhat awkward to implement, since the resulting
- instance identifiers aren't all the same length, but I could live with
- that. The nagging problem is that I can't find any support for this
- "hack" in the SNMP or MIB RFCs or in any of the public implementations
- I've looked at.
-
- Is my interpretation correct?
-
- Steve Tarr
- Tektronix, Inc. GPID
- stevet@pogo.wv.tek.com
-