Download ppt

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Deep packet inspection wikipedia , lookup

Network tap wikipedia , lookup

Distributed operating system wikipedia , lookup

Computer network wikipedia , lookup

Internet protocol suite wikipedia , lookup

IEEE 1355 wikipedia , lookup

CAN bus wikipedia , lookup

Airborne Networking wikipedia , lookup

Recursive InterNetwork Architecture (RINA) wikipedia , lookup

UniPro protocol stack wikipedia , lookup

Zero-configuration networking wikipedia , lookup

IEEE 802.1aq wikipedia , lookup

Routing in delay-tolerant networking wikipedia , lookup

Transcript
Communication (II)
Chapter 4
5/25/2017
1
Topics
 Fundamentals
 Stream
 Multicast

Overlay (application layer multicast)


5/25/2017
Application layer MC vs IP layer MC
Epidemic and gossip
2
Multicast
 Multicast: a source sends a
message to the subset of network
nodes (say a multicast group)



Applications: Video conferences,
network games, database
A source could send through many
unicast
However, sender may not know
individual receivers and efficiency
issue
 Unicast and Broadcast
5/25/2017
3
Network multicast
 A “join” protocol (so to send or receive from) multicast
group - Internet Group Management Protocol (IGMP)
-
Routers forward mcast-addressed datagrams to hosts that
have “joined” that multicast group
-
Multicast Routing protocols -- different trees for diff groups
Group-shared tree
5/25/2017
Source-based trees
4
Overlay multicast
 An application forms their own multicast
network - overlay network


A network that has only hosts who wants to
join the MC application -- “routers” are hosts!
Network links are TCP connections.
 The network can be a tree or a mesh or …
5/25/2017
5
Overlay Construction
 The relation between links in an overlay
and actual network-level routes.
5/25/2017
6
Overlay multicast metrics
 Problem: logical links vs physical links
 Metrics:
 Link stress (per link): how often a packet crosses the
same link

Closer to 1, the better

Stretch or relative delay penalty (RDP):
 Measures the ratio in the delay between two nodes in
the overlay, and the delay that those two nodes
connecting in the underlying network.
 Smaller the ratio (minimum be 1), the better

Tree cost: minimum spanning tree.
5/25/2017
7
Application architectures
 Rendezvous node

Well know node, keep the tree members
 For a single source based applicaiton

Find a best parent node


5/25/2017
Direct to the source (the stretch is 1 !)...
Disadvantage?
Consider nodes’ load (degree), e.g., k neighbor.
8
Topics
 Fundamentals
 Stream
 Multicast


Overlay (application layer multicast)
Epidemic and gossip

5/25/2017
9
Data Dissemination
 Simple techniques for spreading information
in very large-scale distributed systems.
 Want efficiency, robustness, speed, scale
 Tree distribution is efficient, but fragile (plus
configuration is difficult)
 Flooding is robust, but inefficient
 No central control
 Gossip is both efficient and robust, but has
relatively high latency

5/25/2017
Or epidemic.
10
Probabilistic multicast




Distributed local information (local view)
Loose or no synchronization
Scalable
Reliable



Probabilistic guarantees on full delivery
No delay upper bound
Graceful degradation in the presence of failure
 Delete message
 Replication, fault-tolerant
5/25/2017
11
Gossip-based Protocols
 Anti-entropy propagation model
 Node P picks another node Q at random
 Subsequently exchanges updates with Q
 Differ by the number of time they gossip the same
message or the number of gossip targets they select
each time.
 Some “epidemic” related terms:
 Infected – holds the data and willing to spread further
 Susceptible – not seen the data yet.
 Removed – has the date but will not send to others.
5/25/2017
12
Push and Pull
 Approaches to exchanging updates
 P only pushes its own updates to Q
 P only pulls in new updates from Q
 P and Q send updates to each other
 (1) when many are infected, push may induce long
delay for spreading

Pull is better: a susceptible asks around, high chance to
hit a infected.
 (2) when data entries are big, send a “digest” (instead
of the data directly) of the state, and the recipient can
request anything it doesn’t already have.

5/25/2017
Apply to push, pull and push-pull.
13
Centralized algorithm
 Round: picking node, picking a piece of data

5/25/2017
log2(n), for n nodes.
14
Gossip-based protocol: k fanout
4
3
0
5
1
2
9
8
6
5/25/2017
7
15
Analysis
 Probability of infection
n nodes, k member infected,
 Anybody can infect anyone else with equal probability.
 What is the probability Pinfect(k, n) that a particular uninfected
member is infected in a round if k are already infected?

5/25/2017
16
Analysis
 For simple epidemic


Gossip propagation time
Expected number of
rounds
Expected # of rounds
# rounds
5/25/2017
17
Analysis
 s: fraction of nodes remain uninfected:
 1/k: probability of stopping infecting others.
5/25/2017
18
Further design issues
 How to know the n? Or , scalable
membership protocol?

Paper, [SCAMP: lightweight membership service for
gossip-based protocols]
 Removing data:


Early deletion cause restore old data
keep record - Death certificates


5/25/2017
Repeat spreading Death certificates
When to clean up?
19