next up previous
Next: 6 Conclusions and Outlook Up: Preventing Rather Repairing - Previous: 4.3.3 Congestion indication

5 Implementation Experiences

Figure 6: Test Environment

Figure 6 shows the test network at the Leibniz Computing Center in Munich and the connection to the HD012 projectgif. We used the network to drive performance tests in various configurations with switches from Netcommgif and Fore Systems. The performance, especially via the 34 Mbps link, was acceptable, when the end systems were able to shape the submitted traffic to a specified peak rate. This was not the case at the beginning of the tests and thus there were heavy throughput problems due to discarded cells.

The workstation denoted 'OpenView' in Figure 6 is used to implement the self-regulating system proposed in this work. The collection of traffic information is implemented with the SNMP-based platform HP OpenView. Several critical points arise here. Vendor specific MIBs have to be used, as standardized MIBs [10] are not yet available and it is not clear if they will support the detailed information requested in this work. To cope with the multi-vendor environment we implement a generic MIB and take necessary vendor specific details into consideration.

The information is collected using an event-driven technique. This means, whenever one of the interesting parameters changes a value, an event is generated to log the corresponding information. The delay detected when querying one switch, varies around one second and will certainly increase when the network gains complexity in components and traffic.

Figure 7: RTB production environment

Figure 7 presents the planned configuration for the RTB projectgif. The aim of the RTB project is to evaluate applications with high bandwidth demands in the wide area. ATM Switches and Routers from Cisco Systems are used to provide the infrastructure as depicted in Figure 7. The applications provide a suitable traffic mix to gain experience in the detection of regularity and the usage of action values.

The resulting delay for the collection of information will be evaluated. It is easily detectable that the appropriate environment to implement the suggested mechanism would be CMIP, where logging and threshold checking is done by the agents. The implementation in an SNMP environment can only reach the goal to a smaller extent. But solutions for SNMP have to be found, as this is the management protocol supported by most vendors. These solutions will be considered in further research.

next up previous
Next: 6 Conclusions and Outlook Up: Preventing Rather Repairing - Previous: 4.3.3 Congestion indication

Anja Schuhknecht
Wed May 31 09:56:07 MET DST 1995