A Portal Based Approach to Viewing Aggregated Network Performance Data in Distributed Brokering Systems Gurhan Gunduz 1,2 , Shrideep Pallickara 2 and Geoffrey Fox 2 (ggunduz@syr.edu), Department of Electrical Engineering and Computer Science, Syracuse University. 1 (spallick,gcf@indiana.edu) Community Grid Labs, Indiana University. 2 1.0 Introduction Distributed messaging/brokering systems which provide a scalable infrastructure for several applications involve a set of nodes (called brokers) communicating with each other. Clients connected to any of these brokers can communicate with each other. Effective utilization of networks and responsiveness to changing network conditions are essential to mitigate network utilization problems associated with distributed messaging systems. This is predicated on four essential issues. First, there needs to be a monitoring of the communication links hosted at a broker node. This will involve monitoring the links to other brokers and of course to clients connected to the broker in question. Second, brokers should be able to able to respond to changing local and remote conditions. To respond to changing local conditions, a broker may deploy different transport protocols. For example, if there is a high concentration of clients from a given geographic location, a broker may choose to deploy multicast (if possible) for communications with these nodes. To enable a broker to respond to changing remote conditions there should be a scheme for aggregating network performance from individual broker nodes. These aggregator nodes thus snapshot the state of the network domain within which it aggregates performance. Third, it should be possible to view the aggregated network performance data and specify constraints to detect network thresholds under which remedial measures might need to be initiated. These measures could involve creation/purging of connections, migration of underlying transports for communications and forcing connected clients to connect to another broker. Finally, in a truly dynamic system, the detections and responses would be initiated dynamically. Such self healing systems can respond, in real-time, to changing local and remote network conditions. These systems are responsive to changing client concentrations and also optimize bandwidth utilizations associated with the interactions that they route. However, it should be noted that a truly dynamic system still would need performance monitoring at a broker node, require this broker to respond to changing local conditions and remote conditions based on performance accumulated at an aggregator node. In this paper, we present a scheme to view aggregated performance information inside a portal and specify constraints on the accumulated performance information to detect network conditions, based on specified thresholds on measured performance factors. This information would then be used to work around network bottlenecks and also be used to drive heuristics for generating optimal routes (and transport protocols deployed) to reach a set of computed destinations. The scheme that we have prototyped in our current work provides for detection of network conditions. The remedial measures that are initiated are currently based on static techniques. We suggest that this work is a precursor to a dynamically responsive system. This work builds upon our previous work [1] involving the design of a framework for aggregation network performance. The investigations that we report in this paper are in the context of NaradaBrokering [2-10] a distributed brokering system. The remainder of this paper is organized as follows. In section 2.0 we discuss related work in the areas relevant to this paper. In section 3.0 we present a brief overview of the NaradaBrokering system and our framework for aggregating network performance. In section 4.0 we discuss why we adopted a portal based approach to viewing aggregated network performance data and outline details pertaining to the incorporation of performance data within the portal. Section 5.0 presents results from our experiments. Finally we discuss the future work and conclusions derived from the work outlined in this paper. 2.0 Related Work The Network Weather System (NWS) [11,12] is one of the most well-known systems in network monitoring. NWS periodically monitors resources and dynamically forecasts the short-term performance. Sensors in NWS collect end- 1