Network Analysis Infrastructure (NAI)The goal of the NLANR Measurement and Network Analysis Group (NLANR/MNA) is to characterize the behavior of high performance connection (HPC) networks. Network measurements are essential for assessing performance issues, identifying and locating problems (malfunctions, bottlenecks, inefficiencies, incompatibilities, etc.) in ultrafast research networks and in high-speed international links. To meet our research goals, we have created, and continue to develop, a Network Analysis Infrastructure (NAI). The NAI includes not only a growing collection of measurement data and multiple analyses, but also tools and methods, numerous avenues for sharing information, and many collaborations (domestic and international) with other researchers both within and outside of the HPC network measurement community. The NLANR/MNA NAI is the largest project of its kind (of which we are aware) that makes all data publicly available (as both raw data and Web-based performance graphs and visualizations), as well as all analyses, tools, and methods, for use by other network researchers, systems administrators, engineers, and students. Two projects form the core of our research efforts: Passive Measurement and Analysis (PMA) Project
The goal of the PMA project is to deliver new insights into the operation, behavior, and health of the Internet, for the benefit of network users and operations. Passive header trace data provides the means to study workload profiles for a number of strategically located measurement points in high speed environments. We currently take measurements from OC3 through OC48 speeds. In addition, we have recently taken OC192mon/10GigE traces and are working towards an OC192c instrumentation on the Abilene backbone. Since 1995, the National Laboratory for Applied Network Research (NLANR) has been collecting IP packet header traces to support research in understanding the systemic nature of the Internet. The placement strategy for the original set of monitors was to provide good coverage of the overall network, so that each monitor captures a unique portion of the overall network data. The measurement strategy was to capture samples eight times a day for a defined length of time. Initially, this sampling interval was set to 2 minutes, today it is 90 seconds. The reason for the change was to limit the amount of data being captured per day. We now collect between 1.5 and 3.2 Gigabytes of compressed data each day. Both the placement and measurement strategies have been expanded with strategic new deployments, including sites which have longer samples captured just twice a day, special traces (see next paragraph), and new deployments in locations which complement existing monitors and allow for multi-hop instrumentations (several monitors along a communications path). An example of the later is a set of three OC48MONs instrumented at the Abilene GigaPop (Indianapolis), which measured traffic also seen by four to five other PMA monitors, at different locations. In addition to our PMA Daily Traces Archive (~ 2GB per day), we have many special traces available, including a variety of OC48 trace data and long, contiguous traces. These include Abilene-I, the first publicly available OC48c backbone trace data, Auckland-IV, a 45 days continuous trace, and Bell Labs-I, a one week contiguous Internet access IP header trace - to name a few. Leipzig-I, an illustrated continuous 5-day GPS-synchronized IP header trace, Auckland-VIII, NCAR-I, a one hour trace from our new NCAR Gigabit tap (at the National Center for Atmospheric Research, Boulder), have just been published; other special traces will be published as they are captured. Current PMA research efforts are focused on the development of real-time passive measurement and analysis and new metrics. We are pursuing a number of different approaches to this effort and making good progress. Related and complementary projects:
Active Measurement Project (AMP)
The Active Measurement Project (AMP) has a joint research/engineering focus with site-to-site active measurements and analyses conducted between campuses connected by high performance networks. The data collected by AMP is proving to be a valuable resource for network analysis to study the network and derive performance models for various aspects of Internet traffic. The AMP monitors measure the "heartbeat" of the high performance connection (HPC) networks, dissecting the networks in over 15,000 ways every minute. Because the performance of AMP monitors is well known, and coverage is so extensive, the monitoring mesh allows engineers to quickly identify the location, extent, and duration of network events. This gives the engineers a head start on the road to a solution. The long term history kept by AMP also allows engineers, designers, and researchers to see the impact of increasing traffic, and improved network design and capacity on the long term behavior of the network. The first AMP monitor deployment took place in December 1998. Now approximately 150 AMP monitors are deployed and take site-to-site measurements on high-speed research networks throughout the United States, as well as at strategic sites in other countries. The current AMP architecture has two meshes, the HPC mesh (approximately 140 sites) and an international mesh (which includes all international sites and a small subset of the HPC mesh). Each uses a full mesh, with each monitor testing to all the others in the mesh. Across all sites, round trip time (RTT), packet loss, topology, and throughput (user/event driven) are currently measured. Recent expansion and development of AMP has included monitor deployments at GigaPop and backbone sites and development of a GigE interface for AMP. AMP has led the way to a new approach to network measurement (dense coverage with simple monitors). Consequently, there are a growing number of organizations that realize they would benefit from a performance measurement system like AMP. The current goals for AMP include moving it into new domains, such as, other national meshes, deeper into campuses, more international sites, and developing a meta measurement system. To facilitate these needs of the community, the major exercise of reimplementing the original AMP software into a package easily installed on a variety of hardware and software bases is underway. Significant progress has been made. The AMP package will be made freely available; however, we do not intend the various users of AMP to be totally independent from one another. We will encourage and promote a meta-measurement system, formed from select cross project measurements, thus creating measurement linkages between projects. We will also manage the host to host testing (locating AMP software only monitors on an HPC machine). Related and complementary projects:
|
|
|||||||
| ||||||||
|
Top last modified: 08 Oct 2004 Comments, questions are welcome: Feedback 
| ||||||||
