Reported by Marshall T. Rose, Dover Beach Consulting, Inc.
Minutes of the SMNP Agent Extensibility Working Group (AGENTX)
The AGENTX Working Group met for two sessions at the 34th IETF.
Session 1
The first session was held on Wednesday, December 6th. There were no
changes to the agenda published on November 27.
The Chair began with a presentation (also available in postscript)
which:
o reviewed the Working Group's charter;
o described, in high-level terms, the multiple component problem; and,
o reviewed previous IETF publications in this area.
There was considerable discussion with respect to the high-level
description of the multiple component problem. The resulting comments
were incorporated into the presentation.
The Chair then called for presentations by interested parties.
Jim West made a presentation describing his experiences with packet-
based multiplexing. He noted that the use of port 161 is driving the
need for sub-agents and that independent functional entities are best
implemented using packet-based multiplexing. Other points included:
o the "real" snmp agent is started on a non-standard port;
o the proxy-agent listens on port 161, and forwards requests to proxy-
targets based on community name; and,
o if the community is unknown, the request is forwarded to the original
agent.
Limitations to this approach are:
o there is no "standard" API for component developers; and,
o moving the "real" snmp agent to another port is problematic in some
environments, e.g., if the agent reads a configuration file to
determine its port number, then other processes on that system (e.g.
management applications) will malfunction if the configuration file
is altered; but, if the agent has the port number hard-coded, then its
object code must be patched directly.
Bert Wijnen made a presentation describing his experiences with
variable-based multiplexing. He noted that too many interfaces for
component developers is the driving force, and that in developing his
approach, DPI, he sought to:
o hide complexity from the component developer;
o split the core of the agent code from the instrumentation;
o require no modifications of management applications; and,
o provide a "drop-in" solution for component developers.
His approach is targeted for developers of non-network related
components and is documented as RFC 1592.
With only a few minutes left, the Chair asked for other presentations
to be made in the second session. In the interim, there was some general
discussion on whether management applications could support multiple
community strings when communicating with the same agent.
SESSION 2
The second session was held on Friday, December 8.
The Chair indicated that he was stepping down at the end of this
meeting, and that Bob Natale would assume the position of Chair.
Jeff Case made a presentation describing his experiences with variable-
based multiplexing. Jeff noted that the need for implementation
standard MIB modules is the driving force, and that in developing his
approach, EMANATE, he sought to:
o achieve source code portability and binary interoperability;
o achieve compliance with both the SMI and the protocol (e.g., support
multi-phase operators); and,
o achieve flexibility by utilizing the most efficient communications
facilities available on the target platform.
His approach supports environments in which there is interaction
between multiple MIB modules which may be cooperatively realized
by multiple implementation units. In addition, his approach is a
programmatic rather than a protocol approach, and hence is not openly
published.
Randy Presuhn made a presentation describing ongoing work in the area
of variable-based multiplexing. This work focused on issues of both
correctness ("as if simultaneous") and performance (optimizing multi-
phase operations). In addition, he suggested that the key issues were
to develop:
o a set of service requirements;
o a service definition (i.e., an abstract api); and,
o a specification for the protocol used for intra-agent communication.
Brian O'Keefe made a presentation on behalf of himself, Andy
Bierman, Baria Greene, and Dave Perkins, in which they suggested
that the Working Group distinguish between:
o logical entities, which implement functional components, and contain
a system group; and,
o implementation units, which are realized by sub-agents.
For example, a managed system might contain four logical entities:
itself, a router, and two bridges. Further, the two bridge entities might
be realized by three implementational units, one of which is shared
between the two logical entities. Although there needn't be any logical
structure imposed on how implementation units comprise functional
components, an important question is whether there is a one-to-one or a
one-to-many relationship between an operational scope and a logical
entity. It was noted that this dichotomy helps to explain the
difficulties in implementing interMIB relationships.
Prior to leaving, the new Chair presented a few thoughts on the
AGENTX Working Group. He stated that he would ask the Area
Director to appoint a facilitator for the Working Group, to oversee both
content discussion and process. He also felt that the Working Group
should:
o focus on the solution space;
o base approved solutions on a developed and published specification;
o take a "bottom-up" approach towards a solution; and,
o leverage SNMP whenever possible.
Dan Romanscieu made a presentation describing his experiences with
variable-based multiplexing. His approach sought to:
o minimize the number of managed elements visible to management
applications;
o provide an alternate multiplexing agent in case of failure; and,
o require only minimal configuration.
He expressed concern over the difficulty of mapping MIB modules onto
implementational units. And, in view of his experiences, he felt that
the Working Group's goals were ambitious.
Finally, there was group discussion on the issues presented.
One thread dealt with the issue as to the merits of the information
hiding approach implicit with variable-based multiplexing, versus
the information publication approach implicit with packet-based
multiplexing. In particular, Dave Bridgham and Randy Presuhn
presented conflicting arguments as to these merits. A voluntary action
has been assigned asking each participant to post concise summaries of
their positions.
A second thread dealt with the issues as to the difference between the
information selection inherent with packet-based multiplexing versus
the information integration inherent with variable-based
multiplexing. For example, a service provider might provision a
customer-specific subset of the available management information
based on community string.
Further, a third thread dealt with the goal of the Working Group, viz.
to allow multiple, independent contributors to implement a single
system. In addition, three functional requirements were identified:
o applicable system types (outside the scope of the Working
Group);
o consistency with overall framework; and,
o end-to-end performance.
ACTION ITEMS
The Chair assigned the following voluntary actions:
Aleksey Romanov: Please provide the Working Group with a URL for
the proposal on agent extensibility that you posted some time ago (his
response).
Bert Wijnen, Jeff Case, and Dan Romanscieu: Please send ASCII,
Postscript, or HTML versions of your presentation to the Chair so that
your presentation may be included in the ftp archive.
Dave Bridgham and Randy Presuhn: Please post concise summaries of
your position with respect to the efficacy of the proxy approach to the
mailing list.
Andy Bierman, Editor for the entity mib Working Group: Please
consider the relationship between packet-based multiplexing and the
logical entity table, especially how proxy relates to chaining or
referral of requests to logical entities.
