SNMP

The Simple Network Management Protocol (SNMP) [3] allows for the management of systems in a TCP/IP network within a coherent framework. In the SNMP world, network management consists of network management stations, called managers, communicating with the various systems in the network (hosts, routers, terminal servers etc.), called network elements. SNMP based management consists of three parts: 1) a Management Information Base (MIB) [18] that defines the various variables (both standardized and vendor-specific) that network elements maintain that can be queried and set by the manager, 2) a set of common structures and an identification schema, called the Structure of Management Information (SMI) [28], that is used to reference the variables in the MIB, and 3) the protocol with which managers and elements communicate, i.e., SNMP.

The system works as follows: The network managers periodically send queries to the elements to get the state of the various elements. Elements send traps to managers when certain events happen. The manager may analyse the information available to it via results of queries to build a picture of the health of the network and present this information to the human network manager in a variety of ways. Plugins that extend a managers functionality in a vendor-specific manner are available to handle vendor specific MIBs. An example of a commonly used manager is HP's OpenView [11].

The problem of using SNMP is in some ways similar to the problem of defining appropriate checks for our continuous monitoring system. The various system variables that are checked by continuous monitoring equations correspond to MIB variables. The auto-diagnosis checking logic corresponds to logic in the network manager plugin handling the vendor-specific MIBs. Thus issues that arise in defining the checks that a continuous monitoring system should execute also apply to the design of SNMP logic.

SNMP is different from our system in two main ways: First, SNMP does not really have a parallel for our active tests. A manager can manipulate a network element in some limited fashion, e.g., by using setting appropriate MIB variables. However, this is not nearly as general or as powerful as what can be done by an active test executing in the concerned system itself.

Second, the fact that SNMP depends on the network connectivity to be present between the network elements and the manager limits the types of problems that can be effectively auto-diagnosed by using SNMP. In particular, problems effecting network connectivity may not be easily diagnosed by SNMP.

In some ways the use of SNMP complements our approach. A system of auto-diagnosis using the techniques that we described earlier may be responsible for the ``local'' health of a system and its interactions with other networking entities that it communicates with. An SNMP based network management infrastructure may provide overall information about the health of a network using information gained by communication with network elements and their auto-diagnosis subsystems.