Propaganda

1
infrastructure and new methods and systems
devoted to measurement, mock-up and and analysis
of present and future network traffic, topology
and logical structure, to bridge the gap in
theory, protocols and understanding to what the
Internet can be in 2025.
2
EVERGROW
ever-growing global scale-free networks, their
provisioning, repair and unique functions
The Project
message-passing as a control paradigm
expand the recent success of complex systems
from combinatorial optimization to a general
scheme for control and design
infrastructure nodes, fast internet links,
dedicated software for distributed admin and
access control
network traffic measurement across Europe in
real-time, analysis, unique charactistics
fully distributed infrastructure for the future
internet for data provisioning, error correction
beyond collaborative peer-to-peer solutions to
distributed computing in an unfriendly world
3
EVERGROW PARTNERS

• SICS, Stockholm, Sweden
• Abdus Salam ICTP, Trieste, Italy
• Aston Univ., UK
• CETIC, Belgium
• Collegium Budapest, Hungary
• ENS, Paris, France
• EPFL Lausanne, Switzerland
• Ericsson AB
• France Telecom
• Hebrew Univ, Jerusalem, Israel
• ISI Torino, italy
• IBM (Hawthorne, Belgium)
• KTH, Stockholm, Sweden

• Kopenhagen Univ, DK
• Uni-magdeburg, Gemany
• Sheer Networks, Tel Aviv, Israel
• TU Crete
• TAU, Israel
• TeliaSonera, Stockholm, Sweden
• UC Louvain, Belgium
• Univ. Navarra, Spain
• Oxford Univ, UK
• Univ. Paris-Sud, France
• INFN, Rome, Italy
• Univ. Juan Carlos, Spain
• CLAES, Cairo, Egypt

4
EVERGROW PARTNERSMeasurement and Network
Modelling

• Collegium Budapest, Hungary
• Ericsson AB
• France Telecom
• Hebrew Univ, Jerusalem, Israel
• Kopenhagen University

• Sheer Networks, Tel Aviv, Israel
• TAU, Israel
• TeliaSonera, Stockholm, Sweden
• Univ. Navarra, Spain
• Univ. Juan Carlos, Spain
• Weizmann Inst, Israel

5
EVERGROW PARTNERSDistributed Services and
Management

• SICS, Stockholm, Sweden
• Aston Univ., UK
• CETIC, Belgium
• EPFL Lausanne, Switzerland
• Hebrew Univ, Jerusalem, Israel
• IBM (Hawthorne, Belgium)
• KTH, Stockholm, Sweden

• Sheer Networks, Tel Aviv, Israel
• TU Crete
• UC Louvain, Belgium
• CLAES, Cairo, Egypt

6
EVERGROW PARTNERSStochastic and Mechanism-based
Controls

• SICS, Stockholm, Sweden
• Abdus Salam ICTP, Trieste, Italy
• Aston Univ., UK
• ENS, Paris, France
• Hebrew Univ, Jerusalem, Israel
• ISI Torino, Italy
• IBM (Hawthorne, Belgium)
• KTH, Stockholm, Sweden

• Uni-magdeburg, Gemany
• Oxford Univ, UK
• Univ. Paris-Sud, France
• INFN, Rome, Italy

7
Other Projects in FET

• STREPS (3-6 partners, FP5 and 6)
• BISON
• Distributed algorithms based on biological
  metaphors
• Wireless adhoc networks
• COSIN
• Web and other network topologies
• Integrated Projects (12 25 partners, FP6)
• DELIS
• EC-Agents
• PACE
• Not in FET, but clearly related
• IRIS (MIT-ICSI-NYU-Rice) and PlanetLab
  (Intel,Princeton)

8
EVERGROW Scientific Questions

• DIMES_at_home attempts to break through the roughly
  100-site barrier in distributed measurement,
  create a screen-saver client of value to
  individuals.
• What services can we provide to make it
  attractive?
• What new active measurements are possible with
  100-nsec time stamping precision and moderate
  distribution?
• Move beyond capacity and bottleneck on single
  path to interactions
• What new distributed services can organize our
  thinking on secure, usable peer to peer
  computing, and can they be architected and tuned
  to the real properties of the Internet?
• What are the actual (not worst-case)
  characteristics of stochastic algorithms for
  managing distributed systems? Does an expected
  average complexity even exist?
• New applications of method design and
  message-passing to IT.

9
EVERGROW Schedule and Deliverables

• Budgets and work plan get revised yearly, tied to
  review
• First year deliverables
• Measurement
• DIMES server and clients function, first dark
  matter observations
• Dynamic measurement systems in place in telcos,
  partner sites
• Infrastructure
• Working server clusters, integrated into a grid
• Virtual Observatory content and standards
• Peer to Peer
• Plan for new integrated effort and common
  architecture
• Services, and properties proposed
• Stat Mech and Message Passing
• Game Theory activities start in year 2
• Workshops and summer schools (many are join with
  other IPs)
• Santorini on Game Theory in October 2004
• Les Houches on Message Passing in 2006

10
Characterizing Performance of Distributed
Algorithms

• Classic problem resolution of constraints
  (logistics, design verification, etc)
• Several intensely studied simplified versions,
  such as K-SAT
• For K-SAT, two approaches used in computational
  practice
• Depth-first search, with or without backtracking
  to permit exact results
• DPLL exact, but limited to 100s of variables in
  hard 3-SAT
• decimation schemes opportunistic may find
  SAT, cannot prove UNSAT
• survey propagation, belief propagation and
  hybrids used to direct these
• Random walk-based local improvement heuristics
• RWalkSAT (Papadimitriou, recent work by
  Ben-Sasson et al.)
• GSAT/WSAT family of algorithms (Kautz and Selman)
• Ill show some surprising results on how these
  work.

11
Local Improvement methods for 3-SAT

• Note depth-first search requires synchronization,
  hard to distribute
• Decimation and local improvement require
  monitoring, easy to distribute and run
  asynchronously.
• The RWalkSAT ideas
• Start with random assignment, satisfying (1
  2-K) of the clauses
• Improve by flipping
• A random spin? (NO)
• A random spin if delta-E lt 0 or as in Simulated
  Anealing? (NO)
• Go to a random unsat clause and flip one of its
  spins at random (YES!)
• The WSAT idea
• In a randomly chosen unsat clause flip the
  optimal spin
• Standard practice, quite successful, is a 50-50
  mixture of these.

12
Performance of RWalkSAT on 3-SAT
Expect linear time when WSAT finds ground state
directly, exponential time otherwise. This
change occurs at alpha 2.6. (Barthel, Hartmann,
and Weigt, PRE, 2003)
13
Results with WSAT in standard form
14
WSAT has a nearly log-normal distribution
15
Behavior of WSAT in the two critical regimes

• Normal regime (Easy-SAT) 0 lt alpha lt 3.92
• Full RSB 3.92 lt alpha lt 4.15
• 1-RSB, for which good analytical results are
  available
• 4.15 lt alpha to slightly above critical point
• Analysis using finite-size scaling shows that
  behavior is different in each of the two RSB
  regions, with changes at about 3.9 and 4.15, as
  predicted. Work in progress

16
(No Transcript)