Network Configurations
The entire NCNI network consists of an OC48 ring connecting Duke, UNC—Chapel Hill, MCNC, NCSU Centennial, Qwest and NCSU Main Campus. Each of these locations have a gigabit connection to the OC48 ring.
Helios will use the part of the NCNI network that consists of an OC48 connection of two back-to-back linear systems connecting MCNC and NCSU Centennial and NCSU Centennial and Qwest/Interpath POP. This network utilizes OLS 40G multiplexor equipment donated by Lucent to NCNI. There are four wavelengths each of 2.5 Gbps. A pair of wavelengths is currently in production use for the NCNI backbone network and an additional pair of wavelengths is available for experimental use in the Helios project. Figure 1 is a schematic of this network.
Two distributed computing applications (described below) will be run on two quad CPU servers and 3 dual CPU servers. Four will be located at MCNC and one at the Centennial campus. The applications will be communicating over a pair of OC-48c PoS optical circuit on the OLS 40.
The fast tunable laser applications will be tested in a laboratory network.

Figure 1
Applications
Two applications will be investigated on the Helios network: S-Net, which is a distributed network traffic data collection and analysis application designed by Statistics Research at Bell Labs, and a real time meteorology application. Researchers from Bell Labs and MCNC will analyze the S-Net data collaboratively, sharing, comparing, and extending analyses built on the data collected at Murray Hill and on the NCNI network.
S-Net
S-Net begins with packet header collection on network wires and ends with data analysis on a cluster of linux PCs running S, a programming environment for organizing, visualizing, and analyzing data. Packets are captured on a PC with Berkeley Unix using the program tcpdump and time-stamped using the GPS (Global Positioning System) clock discipline, paying attention to packet drops. The compressed header files are moved to the cluster of linux PCs, which are linked by fast switches. Each PC has 1, 2 or 4 processors, 300 to 2000 megabytes of main memory, and a lot of disk space. An algorithm organizes the header information by TCP connection flow, and the flows are processed to create S flow objects that can be analyzed in S.
Flows and S flow objects are computed in parallel on the PC processors and are stored on the disks of all machines. S is run on high-end PCs with a lot of memory. Each analyst has a low-end PC with a lot of disk that stores the user's S directories. The analyst logs onto a high-end machine from the home machine to run S, mounting the home S directories as well as the directories across the cluster housing the S objects. In other words, each data analysis session is distributed across the application PC cluster shown in Figure 1.
Packet header collection at Bell Labs
Collection of packet headers crossing the Bell Labs research firewall has been underway at Murray Hill since November 1998. This extensive set of traffic measurements was used to develop a framework for analyzing the properties of interarrivals of http transactions (see Cleveland, Lin and Sun, 2000). The framework developed for analyzing http interarrivals data analysis will be the basis for analyzing the Helios data, extending the analysis from http to other protocols. This work is needed to understand the protocol mix and the traffic characteristics of each protocol. It will enable better network provisioning, for example.
Packet header collection in Helios
The Helios project will begin by collecting packet headers for traffic in both directions at the gateway to the UNC campus. The Computer Science Department at UNC will collect the packet data using a version of tcpdump that has been optimized by UNC. Fast Ethernet NICs will be used as packet capture tools. Testing in the network lab has shown that the GbE NIC in a P3 400 MHz CPU is capable of packet header capture at loads up to 600 Mbps. This performance is more than adequate for the Helios traffic at the present time. An initial test of collecting a peak hour’s worth of packet headers resulted in 100% packet header collection. Currently the GbE feeder is running at about 15% of its capacity.
Reference
Cleveland, W. S., Lin, D., and Sun, D. X. (2000). Network Simulation: Modeling the Nonstationarity and Long-Range Dependence of Client TCP Connection Start Times under HTTP.In Proc. ACM Sigmetrics (to appear).