CS 34701: Large-Scale Networked Systems

1
CS 34701 Large-Scale Networked Systems

• Professor Ian Foster
• TA Adriana Iamnitchi
• http//dsl.cs.uchicago.edu/Courses/cs347-2002/

2
CS 34701 Course Goals

• Primary
• Gain deep understanding of fundamental issues
  that effect design of large-scale networked
  systems
• Map primary contemporary research themes
• Gain experience in network research
• Secondary
• By studying a set of outstanding papers, build
  knowledge of how to present research
• Learn how to read papers evaluate ideas

3
How the Class Works

• Research papers
• Prior to each class, we all read and evaluate two
  research papers
• During each class, we discuss those papers
• Project
• One-page project description by 2nd week
• Five-page project summary by 5th week
• 10-20 final paper by 9th week
• Project presentations 9th and 10th weeks.

4
Paper Review Discussion

• Everyone reads two papers per class and submits
  an evaluation (see below)
• We discuss (not present) papers in class
• A team of 2-3 leads each discussion
• The leading team submits discussion plan before
  class, submits master critique and summarizes
  discussion at the beginning of following class
• Look over schedule between now Friday, when we
  will allocate discussants

5
Evaluations

• You must submit evaluations of papers
• Email them by 6pm the day before
• Answer a set of standard questions
• State the main contribution of the paper
• Critique the main contribution
• What are the three strongest and/or most
  interesting ideas in the paper?
• Three most striking weaknesses in the paper?
• Three questions to ask the authors?
• Detail an interesting extension to the work not
  mentioned in the future work section.
• Optional comments on the paper that youd like to
  see discussed in class.

6
What Ill Assume You Know

• Basic Internet architecture
• IP, TCP, DNS, HTTP
• Basic principles of distributed computing
• Asynchrony (cannot distinguish between
  communication failures and latency)
• Partial global state knowledge (cannot know
  everything correctly)
• Failures happen. In very large systems, even rare
  failures happen often
• If there are things that dont make sense, ask!

7
Large-Scale Networked Systems

• Internet-connected networks with a large number
  of components, spanning multiple DNS domains
  (usually WAN)
• Designed to solve specific problems
• Content distribution
• Cycle sharing
• File sharing
• Sensor data fusion
• Distributed data analysis

8
Example Gnutella

• Peer-to-peer file sharing system
• File sharing goal is to enable publication and
  access to files
• P2P no central servers all clients also act as
  servers and are equivalent (more or less)
• Issues
• Scaling to very large numbers of nodes
• Properties bootstrapping, reliability, cost,
  anonymity, security, freeloading,

9
Gnutella Protocol Overview

• P2P file sharing application on top of an overlay
  network
• Nodes maintain open TCP connections.
• Messages are broadcasted (flooded) or
  back-propagated.
• Protocol

Broadcast (Flooding) Back-propagated Node to node
Membership PING PONG
Query QUERY QUERY HIT
File download GET, PUSH
10
Gnutella search mechanism

• Steps
• Node 2 initiates search for file A

7
1
4
2
6
3
5
11
Gnutella search mechanism

• Steps
• Node 2 initiates search for file A
• Sends message to all neighbors

7
1
4
2
6
3
5
12
Gnutella search mechanism

• Steps
• Node 2 initiates search for file A
• Sends message to all neighbors
• Neighbors forward message

7
1
4
2
6
3
5
13
Gnutella search mechanism

• Steps
• Node 2 initiates search for A
• Sends message to all neighbors
• Neighbors forward message
• Nodes that have file A initiate a reply message

7
1
4
2
6
3
5
14
Gnutella search mechanism

• Steps
• Node 2 initiates search for A
• Sends message to all neighbors
• Neighbors forward message
• Nodes that have file A initiate a reply message
• Query reply message is back-propagated

7
1
4
2
6
3
5
15
Gnutella search mechanism

• Steps
• Node 2 initiates search for A
• Sends message to all neighbors
• Neighbors forward message
• Nodes that have file A initiate a reply message
• Query reply message is back-propagated
• Node 2 gets replies

7
1
4
2
6
3
5
16
Gnutella search mechanism

• Steps
• Node 2 initiates search for A
• Sends message to all neighbors
• Neighbors forward message
• Nodes that have file A initiate a reply message
• Query reply message is back-propagated
• Node 2 gets replies
• File download

download A
7
1
4
2
6
3
5
17
Tools for network exploration

• Eavesdropper - modified node inserted into the
  network to log traffic.
• Crawler - connects to all active nodes and uses
  the membership protocol to discover graph
  topology.
• Parallel crawling.
• Graph analysis tools
• high-volume offline
• computations.

18
Network growth

• High user interest
• Users tolerate high latency, low quality results.
• Better resources
• DSL and cable modem nodes grew from 24 to 41
  over 6 months.

• Open architecture / open-source environment
• Competing implementations,
• Lower overhead network traffic, improved
• resource utilization, better structure,
• Recently, two-level structure.

19
Growth invariants

1. Graph connectivity 3.4 links per node on
   average.
2. Path length distribution node-to-node distance
   maintains similar distributions.

• Avg. node-to-node distance grew 25 while the
  network grew 50 times over 6 months.
• Random graph theory predicts about 75 increase.

20
Is Gnutella a power-law network?
Power-law networks the number of nodes N with
exactly L links is proportional to L-k N L-k

• Examples
• The Internet,
• In/out links to/from
• HTML pages,
• Citations network,
• US power grid,
• Social networks.

November 2000
Implication High tolerance to random node
failure but low reliability when facing an
intelligent adversary
21
Is Gnutella a power-law network?

• Later, larger networks display a bimodal
  distribution.
• Implications
• High tolerance to random node failures preserved
• Increased reliability
• when facing an
• attack.

May 2001
22
Traffic analysis

• ? 6-8 kbps per link over any connection.
• Traffic structure changed over time.

23
Total generated traffic

• 1Gbps (or 330TB/month)!
• Note that this estimate excludes actual file
  transfers
• Q Does it matter?
• Compare to 15,000TB/month estimated in US
  Internet backbone (Dec. 2000).
• Reasoning
• QUERY and PING messages are flooded. They form
  more than 90 of generated traffic
• predominant TTL7
• gt95 of nodes are less than 7 hops away
• measured traffic at each link about 6 to 8kbs
• network with 50k nodes and 170k links

24
Topology mismatch

• The overlay network topology doesnt match the
  underlying Internet infrastructure topology!
• 40 of all nodes are in the 10 largest Autonomous
  Systems (AS).
• Only 2-4 of all TCP connections link nodes
  within the same AS.
• Largely random wiring.
• Entropy experiment gives similar results.

25
Course Topics

• Internet Architecture and Design Principles
• Flat Pricing vs. Prioritized Traffic
• Internet Measurements
• Availability in Wide-Area
• Patterns in Real Networks
• Modeling the Internet Topology
• Internet Services DNS
• Web Caching, Content Distribution Networks
• Overlay Networks
• Peer-to-Peer systems
• Computational Grids
• Security Issues
• Sensor Nets
• Wireless Networks
• XML SOAP and Web Services

26
Course Topics

• Internet Design Principles
• How do I deliver Internet services end-to-end
  vs. within the network?
• Flat Pricing vs. Prioritized Traffic
• How do I determine which traffic to pass over the
  Internet?
• Internet Measurements
• What does the Internet really look like?

27
Course Topics

• Availability in Wide-Area
• How reliable is the Internet?
• Patterns in Real Networks
• What does Internet traffic look like?
• Modeling the Internet Topology
• How can I construct realistic models of Internet
  structure?

28
Course Topics

• Internet Services DNS
• How well does DNS work?
• Web Caching, Content Distribution Networks
• How do we optimize Web content mgmt?
• Overlay Networks
• Improving routing performance

29
Course Topics

• Peer-to-Peer systems
• Gnutella, etc., etc.
• Computational Grids
• Globus, etc.
• Security Issues
• Authorization, etc.

30
Course Topics

• Sensor Nets
• How do I structure program networks of
  lightweight devices?
• Wireless Networks
• How do I route in ad hoc networks?
• XML SOAP and Web Services
• What are Web services anyway?

31
Projects

• Literature surveys, real implementations,
  analytical evaluations
• Can be performed individually or in a team of two
• Your project ideas appreciated (to be discussed
  before proposal due date)
• Primary goal is to do something interesting and
  to do it well

32
Example Project

• Gnutella network analysis
• Develop a crawler that traverses network,
  collects membership connectivity info
• Analyze structure
• Characterize structure

• See, e.g.
• Mapping the Gnutella Network Properties of
  Large-Scale Peer-to-Peer Systems and Implications
  for System Design, M. Ripeanu, I. Foster, A.
  Iamnitchi, in IEEE Internet Computing Journal,
  vol. 6(1), 2002

33
Project Ideas

• http//dsl.cs.uchicago.edu/Courses/cs347-2002/cs34
  7_projects.htm
• Gnutella network measurements
• Topology discovery for 500K nodes
• Structural analysis with 500K nodes
• Study impact of overlay networks
• Etc.

34
Project Ideas

• Overlay networks build unstructured or
  semistructured self-organizing overlays
  optimizing different cost functions
• Topology-aware map onto physical infrastructure
• Usage-aware map onto usage patterns
• Analysis of Sloan Digital Sky Survey logs to
  explore access patterns
• What files are accessed how often
• What community usage patterns emerge?
• How can we exploit these in content distribution
  networks?

35
Project Ideas

• Compare qualitatively and analytically current
  file-location solutions (CAN, Chord, Gnutella,
  Napster, etc.) in the context of scientific
  file-sharing collaborations.
• Evaluate sharing patterns based on real usage
  traces in a scientific collaboration
• Use these patterns to evaluate benefits/drawbacks
  and propose better alternatives
• Expand existing simulator to evaluate request
  forwarding techniques for resource location in
  grid environments

36
For More Information

• Contact me
• Ian Foster, foster_at_cs.uchicago.edu
• Email or set up a meeting
• Contact Anda, our TA
• Adriana Iamnitchi, anda_at_cs.uchicago.edu
• Monitor the class web page
• http//dsl.cs.uchicago.edu/Courses/cs347-2002/

37
Next 2 Classes

• Friday
• Discuss
• J. Saltzer, D. Reed, and D. Clark, End-to-end
  Arguments in System Design. ACM Transactions on
  Computer Systems, Vol. 2, No. 4, pp. 195-206,
  1984.
• D. Clark and M. Blumenthal, Rethinking the design
  of the Internet The end to end arguments vs. the
  brave new world, Workshop on Policy Implications
  of End-to-End. December 1, 2001.
• Leading group Ian 2 volunteers (who?)
• Wednesday
• Leading Group Anda 1-2 volunteers (who?)