The third CAIDA/WIDE Workshop was held on April 22-23, 2004 (by invitation only) in Marina del Rey, CA. The Workshop covered three main areas: DNS, IPv6, and BGP, as well as miscellaneous research and technical topics of mutual interest for CAIDA and WIDE participants.
Place: Information Sciences Institute, Marina del Rey, CA
CAIDA/WIDE CollaborationThe objectives of informal joint CAIDA/WIDE workshops are:
- to discuss Internet measurement analysis activities going on in each organization;
- to explore opportunities for joint collaboration on projects;
- to continue data exchange.
April 22 (Thursday)
Review of 2003 activities
Jun Murai (WIDE),
- Internet traffic developments in Japan: rapid increase of stream-based traffic
- domestic 10GE network: triangle JAIST - Osaka- Tokyo. How to deal with lambda-networks from user's point of view? How to measure these networks if there is still no traffic?
bill: are there tools to measure trafic at multiple Gb rates?Kenjiro Cho (WIDE), WIDE measurement activities
kc: what's the goal for making such tools? what are the questions to answer?
nevil: hardware exists
kc: does it work?
bill: real time control of applications' performance is necessary. We need measurements to enable it.
- DNS, IPv6 topology, BGP simulation, netflow/sflow - subjects of this workshop
- 10G (lambda-networks vs. layer3), StarBED, distributed IX, satellite, AI3, UniDirectionalLinkRouting, mobile, InternetCar, Auto-ID - other topics
kc claffy (CAIDA), CAIDA report
- results in six main areas identified in CAIDA Progam Plan 2003-2005:
- macroscopic topology (most comprehensive in the world macroscopic topology data) & routing (new theoretical approaches to compact routing schemes)
- workload characterization (2005 goal: 24 hr packet trace from the core)
- DNS: this workshop (OARC proposal was submitted to NSF)
- performance (BWEST, intermediate RTT from skitter/scamper)
- IMDC - Trends (more important now than the amount of money NSF gave to it)
- security (automatic detection/protection is the main task)
- Jun Murai (WIDE), WIDE report
11:30-17:30 DNS measurements and modeling
Yuji Sekiya (WIDE),
Comparison of active and passive measurement results
- active measurements with dnsprobe (RTTs to root & ccTLD servers) vs. passive measurements with NeTraMet (2 locations since Oct03, at Keio & Tokyo Universities)
- consistent active&passive results at U. Tokyo, strange M-root RTTs at Keio (different active&passive results!)
- future plans: re-organize measurements, detect anycast servers, publish NeTraMet results
Nevil Brownlee (CAIDA),
Measurements and laboratory simulations of the upper DNS hierarchy
- PAM 2004 presentation
DNS projects at the U. of Auckland
- new statistical plots to visualize root servers' performance: X - time of day, Y - RTT distribution to a given root, color = # of queries
Duane Wessels (CAIDA), Data collection and
analysis for DNS-OARC
- components: passive collector, data files in XML, centralized system for archiving/analysis/display
- collected: IP protocol types, response codes, query types, etc.
- examples of graphs: no ICMP attacks on roots (entirely UDP traffic), mysterious spikes in rates of queries for root-server addresses, analysis by query types and clients' subnets
- software is run on two primary instances of F-root. It keeps a week of data & monthly summaries
k: CAIDA is commissioned by RSSAC to collect data on effects of anycast & dnssec on the global Internet stability
Akira Kato (WIDE), Measurements at M-root server
- M-root gets MOST of its queries from US (a factor of 5 more than from Japan!)
- other large contributors: Japan, Korea, China
bill: lots of people want to collect data from roots. what's the best way? (not everybody likes OARC)
Bradley Huffaker (CAIDA), Sources of
strange queries at F-root
- queries come sometimes from Europe and sometimes from the US, there is no clear pattern
- a small # of prefixes queried both instances of F-root. In principle, this should not happen
Matthew Luckie (WAND), IPv6 DNS
- easy to introduce errors into IPv6 DNS config files/records (typos)
- got 410 zones, 3% (12) had errors: not as many as expected, but rudimentary checks would help even more (i.e. Best Practices for IPv6 DNS config)
Francisco J. Martin (Oregon State University),
Toward rapid diagnosis and repair of DNS problems
- cognitive networking: self- configuring, diagnosing and tuning
- approach: build a model, identify 'normal' & 'abnormal' behavior, shift diagnostic discovery from endpoints (human users) to the network itself
- funded by DARPA, looking for more funding
Genevieve Bartlett (ISI),
Deploying DNSSEC at a root server*
* This talk was given on Friday, April 23.
- monitoring traffic at B-root
- preliminary analysis & statistics
General discussion: future directions of DNS research and collaboration
Bill Manning (ISI)
- IPv6 queries at root servers: about 20 per 10 minutes
"Bottom-up" DNS measurements
- is it feasible to study the behavior of "stub" resolvers?
- get the following data: version id of resolver, # of replies, timestamps, initial & subsequent queries, reply data
- need a tiny system that would fit into an embedded device/silicon
- Yuji Sekiya (WIDE), Comparison of active and passive measurement results
Matthew Luckie (WAND),
Active measurements of IPv6 topology: scamper project
- scamper PMTU: discover MTUs along the Path
- scamper analysis shows examples of suboptimal routing (US-US goes through Europe, Japan)
- discover alternative path and examine delays along each path
bill: what are possible path metrics? RTT is one, largest MTU may be another.
Kenjiro Cho (WIDE), Dual-stack world: a
view from .jp
- iWhy is the IPv6 performance often much worse than IPv4 one?
- pinged dual v4-v6 sites: 10% unreachable, ~60% reachable for both, 20% unreachable for IPv6, 10% unreachable for IPv4
- compared rtts: most sites - not bad, but there is a number of BAD sites
- next step: make measurements from Europre & US, improve visualization
kc: isn't it true that tunnels make routing bad?
bill: is info about tunnels in RTTs? but: in long or short ones?
General discussion: future directions of IPv6 research and collaboration
Kenjiro Cho (WIDE)
- IPv6 measurement metrics at IAJapan
k claffy (CAIDA)
- problem - we can't measure tunnels (layer 2)
- what should be the outcome of IPv6 research? a web page monitoring & comparing IPv6-IPv4 performance?
- should we encourage users to switch from tunnels to native?
Jun Murai (WIDE)
- measurement/analysis should be aimed to help improve the future design
- ping root & TLD servers from a number of locations, compare IPv6-IPv4 performance
- Matthew Luckie (WAND), Active measurements of IPv6 topology: scamper project
Bartek Wydrowski (Caltech),
FAST TCP - Internet Congestion Control*
* This talk was given on Thursday, April 22.
- tested to some extent (dummynet, PlanetLab, Internet2 backbone, ns-2), future - WAN in the lab
- algorithm yields steady throughput
- TCP benchmark - need for an independent verification, reproducibility, fair comparison. Who can do it?
Seiichi Yamamoto, Traffic monitoring
- The goal is to establish practical traffic monitoring & to deal with high-speed networks
- sflow simplifies the measurements by using packet sampling. ALthough it does not provide complete information, it reveals the trend (cf. RFC 3176)
- next steps: verify meauserement stability, find effective points for monitoring
david: we at UCSD developed an adaptive packet sampling technique
Colleen Shannon, Security data
collection at CAIDA
- trace collection involves:
- maintaining remote monitors
- transfering data to SDSC (huge files over slow links)
- sanitizing data (tradeoff between payload & IP anonymizing)
- providing access to external users
bill: how to distinguish between harmful worms & possible beneficial self-propagating code (i.e. patches & updates)?
- Bartek Wydrowski (Caltech), FAST TCP - Internet Congestion Control*
BGP and Routing
Kengo Nagahashi, BGP simulation
environment on Starbed
- STARBED is a large-scale PC cluster
scales: basic (14 hosts), middle (~100), large (~1000)
- comparison of convergence time for random topology & hierarchical topology (longer for random topology)
Dima Krioukov, Compact
- new theoretical approaches to Internet routing
Kenjiro Cho, Server
placement/selection for scale-free networks
- simulations to estimate the efficiency of different server selection algorithms on scale-free graphs
- future plans: animations
- Kengo Nagahashi, BGP simulation environment on Starbed
April 23 (Friday)
- statement of work for WIDE
- next workshop on August 6-7, 2004 in San Diego (after IETF meeting)