|
|
Visualization ISMA 9904 survey replies |
Co-editor of an upcoming book: External Memory Algorithms, DIMACS-AMS special volume (J. Abello, J. Vitter, Eds)
We plan to have some of our tools available with a software infrastructure for our power wall;
Prashant Rajvaidya
Graduate Research Assistant
University of California, Santa Barbara
We have worked on several projects that look at monitoring multicast group membership and traffic statistics. To date, our group has done little formal work in visualization but that is the purpose of attending this workshop.
sdr session monitoring effort:
http://imj.ucsb.edu/sdr-monitor/
multicast router traffic collection:
http://imj.ucsb.edu/mantra/
I'll give a demo as necessary.
I work with Bill Cheswick of Bell Labs on gathering and visualizing networks, especially the Internet, as seen from one host.
Tamara Munzner's tool is helpful, although displays only a smaller portion of the network at a time
Our tool (ches and I) can display fairly good pictures, but must be done off-line, and still not real satisfying
The rest of the tools I've seen do fairly well on smaller graphs, but not so well on large graphs (CAIDA'a otter, for example).
Bell Labs:
http://www.cs.bell-labs.com/~ches/map/index.html
http://www.cs.cmu.edu/~hburch/map/ (well, not really it)
CAIDA:
http://www.caida.org/tools/visualization/otter/
http://www.caida.org/tools/visualization/mapnet/
Cybergeography:
http://www.geog.ucl.ac.uk/casa/martin/atlas/topology.html
Tamara's stuff was really designed for web sites:
http://www.sgi.com/software/sitemgr.html
others:
http://www.jevans.com/pubnetmap.html
traffic data:
http://www.mids.org/
http://www.mids.org/weather
http://computer.ncsa.uiuc.edu/~vwelch/net_perf_tools.htm
many others
With Hal Burch I have been systematically collecting internet reachability data from a test host to targets on most announced networks. This database probably locates most important routers on the Internet, and maps the topology with regard to our outgoing packets.
A portion of the Internet is scanned daily, with occasional full scans. Each day's database has been archived, and may become a useful historical resource. The archive includes includes naming information for each node, when available.
We have used a spring-simulation algorithm to attempt to lay out this data in a pleasing and useful graph. Our recent graphs plot a minimal distance tree, which makes most paths traceable. Though this does not necessarily plot the actual path, it does show connectivity information.
The mapping technology is quite useful on intranets, and we are working on tools to map and monitor intranets.
We also plan to commercialize the Internet layout as a poster.
Involved in the development of applications for AARNET (Australian Academic Research Network) to monitor/record and summarise the traffic flows and consumption of the member institutions as well as the performance of the circuts of the providers. Designed to use NNstat and CISCO Netflow data sources.
Visualisation of BGP routing tables using knowledge modeling tools.
we're trying to make them more relevant to ISPs but really need feedback from those using them in the field to do so. (Otter, Skitter, cflowd, Coral)
will give overview of caida viz tools on main agenda
(2) but very difficult to make them relevant to espeically large/backbone ISPs without feedback on use in field.
(3) also difficulty scaling up to number of nodes ISPs need to visualize/simulate (tens of thousands)
Actively researching the geography of the Internet, the Web and Cyberspace for past three years in a project rather grandly titled "Cyber-Geography Research" and results are presented on the cybergeography web site.
My background is in human geography and geographical information systems (GIS).
I am involved in analysing and mapping the geographic patterns of ownership of Internet addresses (IP and domain names) in the UK and Japan with colleagues Naru Shiode and Matt Zook.
I am beginning a project to analyse and map the content and hyperlink structure of the WWW in the UK, again with Naru Shiode. I am also investigating the social geography of visual MUDs. I am interested in whether multi-user virtual worlds could be used as environments for data visualisation and exploration.
With geography / urban planning colleagues from UK and US, we are beginning a small project called "Internet Census" (http://www.netcensus.org/) to work on means to measure the Internet in ways meaningful useful for geographic analysis.
Lastly, I am very interested in the visual style and metaphors others employ to map the Internet and other parts of Cyberspace. I attempt to catalogue these into an "Atlas of Cyberspaces" and I am writing a book (with a colleague Rob Kitchin) on this.
Weaknesses include constraints of planar, map storage and representation inherent in most systems, are well as costs and proprietary nature of many GIS.
Also, as someone from outside the US, I would argue for more attempts to measure and map the Internet to extend beyond the North America and try to show more of the global nature of the Net.
I've been involved in the creation of several innovative and novel tools to visualize network data.
I'll try to demo these tools using my laptop.
I don't claim to have any good answers.
Due to the growth in Internet usage and adoption of Java in networking applications, we have seen more need for high performance visualization. As a result, we have developed features in our new products that address these needs.
In the past year, we have seen market demand for Java rise significantly and responded to that by introducing Java versions of our successful software components, Graph Layout Toolkit and Graph Editor Toolkit. The latter is not just a Java port of an existing C++ product, but a complete redevelopment with many new features, such as unlimited undo/redo capabilities and a client-server architecture to support thin client solutions. One area that we worked hard on was making sure Java could meet performance requirements for large network/Internet scale applications. Our 3.0 release is capable of displaying and laying out graphs with thousands of nodes.
In the next few months, we plan to introduce a new version of the Graph Layout Toolkit, featuring
However, I'd be happy to demonstrate the brand-new Graph Editor Toolkit for Java 3.0, which is a 100% Pure Java solution that provides customers with a thin-client fully programmable graph editor. In addition, it is capable of communicating with an industry-standard Graph Layout Toolkit server component, which can be either locally installed or a remote server, and which provides automatic layout capabilities.
NicheWork: Developed by Graham Wills of Bell Labs/Visual Insight.
The future challenges include among many scalability, real time, and applications. First, Real networks such as an Intranet are usually very large and complicated, and users could be required to simultaneously visualize and control many kinds of heterogeneous information. We need to find the best level of detail mechanism to display the network information. Second, to deal with large amount of real time information possibly collected through Internet, simple animation is usually not enough. We need to be able to present the visual interface to assist users aggregating and comparing network attributes. Finally, the network applications, especially for IP networks, are quickly expanding. This has posed many interesting problems to network researchers. For example, how to collect information from MIB and put probes in a network to automatically detect its configuration and/or predict network faults? Network visualization could potentially play a very important role in these applications. However, we need to first address problems like distributing visualization, security, and bandwidth. More importantly, we need to invent more effective visual metaphors and interactive controls for these applications.
Working on the VINT project with the ns and nam, the LBL/UCB/VINT network simulator and animator.
Currently nam is primarily used to visualize network simulation output, but it has been used to visualize network traces. We are working to make playback/conversion of real network traces easier.
Over time we've come up with some simple metrics for tracking the overall performance of our network. When they change, the question is why and we find ourselves looking at nxn loss or delay matrices. Visual ways to "diff" loss or latency for a small to medium network (tens to hundreds of nodes) at different points in time or against "expected" values would be very useful.
Beyond absolute traffic kinds of measures, we'd like to be able to determine "normal" behavior of bits of the network over various time scales and display values relative to that.
We'd like to make tools that are useful in practice to network operators and application developers on current networks, and look nice, rather than creating a general framework on which all future efforts should build, etc.
Besides creating desktop graphical tools, we'd like to integrate the intended vis system into the Virtual Director, a collaborative framework for navigating through & recording scenes in VR (e.g. in the CAVE). Stuart Levy could demonstrate Virtual Director on an SGI workstation (say an O2). This wouldn't be a network-vis demo though.
Characterizing "normal" behavior is something I'd like to learn about, too.
ftp://ftp.ncsa.uiuc.edu/cosmic/larry/mpegs/NSFNET.mpg
Robert Patterson's animation of the growth of the NSFNet (9MB).
http://www.geom.umn.edu/~slevy/mtraces/
From Stuart Levy, a little study of MBone packet loss.
we're trying to make them more relevant to ISPs but really need feedback from those using them in the field to do so. (Otter, Skitter, cflowd, Coral)
will give overview of caida viz tools on main agenda
Skitter data results in a spanning tree structure originating at the polling host and extending into the infrastructure toward the destination hosts in the polling set. We then aggregate data into a centralized database for correlation and depiction as a top-down, macroscopic view of a cross-section of the Internet from at least a small set of sources. Juxtaposition of such data sets has been a remarkably unattended area given its potential utility in a variety of areas. Analysis of real-world trends in routing behavior across the Internet has direct implications for next generation networking hardware, software and operational policies. Observations of macro-level traffic patterns provide insights into:
tracking related performance effects in real-time using skitter's RTT variance data to indicate regions of the infrastructure experiencing abnormal delay
as a quality check for backbone engineers to compare the expected with actual results of configuration/policy changes
identifying critical paths in the infrastructure, i.e., routers or exchange points that might be sources of significant network vulnerability.
performance testing of fielded Internet hardware, e.g., initial skitter measurements identified statistically significant problems on certain routers using network route cache technology.
Skitter also offers promise in potential correlation to BGP data to allow engineers to discern who is announcing what to whom over specific paths. (We have software to hold BGP RIB attributes (AS path, origin, next hop, MEDs, localpref, aggregators, community) for each path, in addition to RTT values (microsecond granularity) and dropped packet counters). We will eventually integrate Skitter output data with a comprehensive database of physical topologies (e.g., prototypes are CAIDA's java-based topology mapping tools for ISP backbones, the Mbone and caching hierarchy topologies). These data can also help pinpoint routing instabilities and other anomalies, and track their secondary, downstream effects, e.g. on round trip times, availability, packet loss across specific paths. A repository of these data/analyses will significantly enhance our predictive capabilities on the Internet, and holds promise for insights into the infrastructure as a whole.
H3/H3Viewer, a 3D hyperbolic graph layout and drawing system that scales to graphs with around 100,000 edges.
Planet Multicast a 3D geographic visualization of the 1994 MBone.
----------
current:
----------
Constellation, a 2D visualization system for linguistic queries of the MindNet semantic network. (no URL yet)
----------
future:
----------
Checkerboard, a visualization front end for extremely large graphs which have been chopped into bite-sized pieces by James Abello's hierarchical decomposition algorithm. The targeted platform is a large high-resolution (4000x1500 pixel) display system.
task-appropriateness: many viz tools try to handle the generic case, but a custom-designed tool that exploits domain knowledge might be more effective
I'm not any kind of expert on Internet statistics and metrics, but I've been working in software and network visualization (applied algorithms and tool hacking) for more than 10 years. In January 1998 my colleagues and I started a new department (Information Visualization) in Network Services Research. Most of us participate in the AT&T Infolab, an interdisciplinary program for exploring and understanding data, often at extreme scale, about telecomm networks, services and their underlying information systems.
In the long term our goal is to solve the scale problem in graph (network) visualization. We are looking into external memory graph algorithms to help with the data management part of this problem and to organize graphs for visualization.
I would mention that our group has several mature tools for graph visualization:
graphviz (gviz) is good for static layout of graphs having up to say several hundred nodes (hierarchical layout and force-directed placement). Works with Unix/X, Java, win32, HTML (GIF/ismap).
dynagraph is a demo-able, almost useful prototype for dynamic hierarchical layouts. We have front ends in TCL/tk and win32/OLE. (We tried using Pad++ hoping to support compound graphs, but it had too many problems.) I'm hoping to use this for experiments in incrementally browsing small pieces of very large graphs.
SWIFT-3d is an interactive viewer for large scale network data displayed with geographic maps and standard statistical controls. It can handle drill-down on extremely large scale data, e.g. a day's worth of long distance calls. We're now adapting it to the AT&T Frame Relay network and from there we hope to go to IP networks and web traffic. I have a video though it's a year old, and the work has progressed beyond that point. I can also bring videos of the graph layout work, but I think it's a little far afield for the IP data analysis + metrics who are probably more interested in what the tools can do, than in applied comp geom.
Lead engineer on design and implementation of IP and ATM routing models (BGP, EIGRP, IGRP, OSPF, IS-IS, RIP, PNNI) in NetMaker MainStation 3.0. Project lead on automatic model generation from Cisco configuration files.
In other words, visualization is both of collected data (current or historic) as well as data generated from a model of the network (futures, what-if scenarios) that includes devices, links, traffic routes, utilizations, evaluations of service level criteria, etc. Because of the complexity involved, we strongly associate visualization with the ability to filter information and perform visualization transformation operations on the filtered data.
Our recent activities at Make Systems have been to complete NetMaker MainStation 3.0, which shipped on March 30. A major focus of this release is the ability to model and analyze the performance of networks with multiple technologies (IP, ATM, and bridging) constructed from equipment from multiple vendors, carrying traffic with diverse stochastic characteristics and service requirements. Another focus has been simplifying model creation. MainStation automatically constructs accurate topology and configuration models from Cisco configuration files or from ASCII data exported from network management tools. MainStation also automates creation of traffic models from sniffers/probes, NETFLOW data, and baseline data. A key aspect of this system is the ability to model the achievement or failure of application service level criteria (delay, jitter, loss) computed on an end-to-end, transaction oriented basis.
Future plans include
Other MainStation tools facilitate network "visualization" in a broader sense. Object Editor and Report Engine display detailed information on the configuration of network elements. Planner models the operation of a network to help you visualize the network's performance under what-if scenarios. The Report Engine displays performance results with text and graphics. The Analyzer tool automates failure analyses. The Interpreter and Baseliner tool facilitate the visualization of measured traffic statistics. Interpreter and Baseliner offer extensive sampling, filtering, and aggregation features. The Accountant tool analyzes network cost.
Equally important, in our view, is the question of which variables drive the visualizations. One interest of ours is the extent to which key variables might shed some light on the comparative market power of various Internet actors and the ways in which traffic flows (loads) may map to payments. What can the graphic display of available -- or possible-- Internet metrics tell us about the viability of different Internet economic models (peering, access and interconnection prices).
In a more consumer-oriented vein, we also have an ongoing interest in user-friendly graphics which can help someone see how their terminal, server, ISP, backbone fits into the larger Net. Here, perspective may be more important than utility (navigation). Ditto for information searches, which is why some of the tools being developed at Alexa and Invisible Worlds seem promising to us.
For those interested in sizing the total international bandwidth available by country see www.telegeography.com/Publications/cmap99.html
I specify and project-manage the relatively simple visualization tools that we write in-house here, and I try to keep abreast of the commercial tools available in the same space, as well, obviously, as those k's folks turn out.
| for more information: info @ caida.org | last update: |
| this page: | |