Network Working Group C. Partridge Request For Comment: 1021 BBN/NNSC G. Trewitt Stanford October 1987
An overview of the RFCs which comprise the High-Level Entity Management System is provided. This system is experimental, and is currently being tested in portions of the Internet. It is hoped that this work will help lead to a standard for IP internetwork management. Distribution of this memo is unlimited.
Until recently, a majority of critical components in IP networks, such as gateways, have come from a very small set of vendors. While each vendor had their own set of management protocols and mechanisms, the collection was small, and a knowledgeable system administrator could be expected to learn them all.
Now, however, the number of vendors has grown quite large, and the lack of an accepted standard for management of network components is causing severe management problems. Compounding this problem is the explosive growth of the connected IP networks known as the Internet. The combination of increased size and heterogeneity is making internetwork management extremely difficult. This memo discusses an effort to devise a standard protocol for all devices, which should help alleviate the management problem.
The RFCs that currently define the High-Level Entity Management System are this memo along with RFC-1022, 1024, and 1023. This list is expected to change and grow over time, and readers are strongly encouraged to check the RFC Index to find the most current versions.
Historically, the IP community has divided network management into two distinct types of activities: monitoring and control. Monitoring is the activity of extracting or collecting data from the network or a part of the network to observe its behavior. Control is the activity of taking actions to effect changes in the behavior of the network or a part of the network in real-time, typically in an attempt to improve the network's performance.
Note that the ability to control presupposes the ability to monitor. Changing the behavior of the network without being able to observe the effects of the changes is not useful. On the other hand, monitoring without control makes some sense. Simply understanding what is causing a network to misbehave can be useful.
Control is also a more difficult functionality to define. Control operations other than the most generic, are usually device-specific. The problem is not just a matter of providing a mechanism for control, but also defining a set of control operations which are generally applicable across a diverse set of devices. Permitting remote applications to exercise control over an entity also implies the need for a suite of safeguards to ensure that unauthorized applications cannot harm the network.
Because monitoring is the key first step, in this initial design of the system, the authors have concentrated more heavily on the problems of effective monitoring. Although the basic control mechanisms are defined, many components need for control, such as strong access control mechanisms, have not been fully defined.
The HEMS is made up of three parts: a query processor which can reside on any addressable entity, an event generator which also resides on entities, and applications which know how to send requests to the query processor and interpret the replies. The query processor and applications communicate using a message protocol which runs over a standard transport protocol.
The query processor is the key to the management system. It interprets all monitoring and control requests. For optimal network management, we would like to see query processors on most network entities.
To encourage the implementations of query processors, one of the
primary goals in designing the query processor was to make it as
small and simple as possible, consistent with management
Defining the management requirements was no small task, since the networking community has not yet reached a consensus about what kinds of monitoring information should be available from network entities, nor what control functions are required to properly manage those entities. The standards for HEMS were developed through discussions with several interest groups, and represent the authors' best effort
to distill the varying sets of needs.
The authors settled on a system which was extensible, robust and host-architecture independent, and as simple as possible, consistent with the other goals. Extensibility was essential because it is clear that management needs will continue to evolve, and a closed system which could not be changed would be obsolete almost as soon as it was defined. Unfortunately, extensibility is also the requirement least consistent with simplicity since the need to make the system extensible led the authors to use self-describing data formats and an interpreted query language.
A robust system is required if the system is to be useful for diagnosing network failures. If the monitoring system cannot survive at least moderate network failures, it is not useful.
The query processor is designed to be highly extensible. An application sends the query processor instructions about objects to be examined or changed. The query processor locates the objects in its host entity, and performs the requested operations. The objects are self-describing, using the binary-encoding scheme defined in ISO Standard ASN.1. Care has been taken to use a limited set of the ASN.1 coding set, so that query processor's handling of data can be optimized.
It is a key feature of HEMS that messages to the query processor contain multiple instructions. The authors felt that this would give much higher performance than a remote procedure system which limited an application to one operation per message.
The set of maintained objects is standardized across all entities. Every entity is required to manage a small set of objects. In addition, entities of a particular type (e.g., a gateway) may be required to manage a larger set of objects, which are optional on other entities. Entities are also permitted to make additional, entity-specific objects available to applications. A method for discovering the existence of additional objects is defined.
The combination of self-describing data, the ability to add to the standard data set, and a query language which can be easily enhanced appeared to offer the necessary extensibility.
On many network entities, particularly critical network components such as gateways, it is necessary to have a way for the devices to send unsolicited status messages to network management centers. In the IP community, these messages have historically been referred to
as "traps", but for compatibility with the ISO nomenclature, in the HEMS system they are called "events".
In the HEMS system, events are handled as slightly specialized replies to queries, and are sent to one or more management centers. Like all other HEMS messages, events are formatted in ASN.1 format.
Each event is given a well-known code, which is standardized across all entities. Provision is also made for entity specific event codes.
The HEMS expects that applications will be more intelligent than the query processor. Among other functions, the applications will have to be able to identify and parse entity-specific values which may be returned.
The details of applications are largely not discussed in the HEMS specifications because there is very little that needs to be standardized. Applications must send requests using the protocols discussed in the next section, but the interfaces the applications provide for displaying monitoring or control information are entirely application dependent.
Query processors and applications communicate using an application- specific monitoring protocol, the High-Level Entity Management Protocol (HEMP). This protocol provides the formatting rules for the queries and their replies.
HEMP runs over a standard transport protocol. There was a certain amount of debate in the community about what type of transport protocol was best suited for monitoring. The key issue was how reliable monitoring interactions needed to be.
The authors expect that three types of management activities will predominate: status monitoring, firefighting, and event reporting.
Status monitoring is envisioned as occasional retrieval of monitoring information, possibly in response to the receipt of event messages. In these situations, the network is expected to be in good working condition, and monitoring exchanges could probably comfortably work with an unreliable transport protocol. The chance of data loss is small, and probably not a serious problem since the data is unlikely to be so important that it must be reliably delivered. (However, it should be noted that some applications may prefer reliable delivery
because it is more convenient.)
Firefighting is a completely different situation. In this scenario, one or more sites are using management applications to try to locate and fix a network problem. Here we must assume that while the network functions (i.e., data can get through), it is not very healthy. We should assume that packets are being lost, that network routes will be non-optimal and that it is essential that the monitoring data (which is presumably diagnostic) get back to the application and that control requests are reliably delivered to the entity. In such circumstances, a reliable protocol is essential.
Events provide yet another bit of complexity. Events contain useful status information, but experience suggests that this information does not have to be delivered reliably. If the problem is serious enough, it will re-occur and the event will be sent again. Furthermore, events will often be sent to more than one management center, which would appear to preclude the use of connection- oriented, reliable protocols such as TCP for events.
The current decision has been to establish two possible transport options for HEMS. More experimental systems may use the Versatile Message Transaction Protocol (VMTP), an experimental IP transaction protocol. Near term production systems can use a combination of the Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP), as described in RFC-1022.
Several groups have expressed interest in being able to develop applications which can use both HEMS and the emerging ISO-defined Common Management Information Protocol (CMIP). It turns out that such a co-existence is feasible, and the authors have made an effort to accomodate it.
At the highest level, both CMIP and HEMS perform operations on objects stored in remote entities, and both systems use ASN.1 formatting to represent those objects. This makes it possible to develop a standard set of interface routines which can be used to access either system, even though underlying mechanics of the systems are quite different.