CMPE 255: Advanced Computer Communication LECTURE 12:

1
CMPE 255 Advanced Computer CommunicationLECTURE
12

• Content Routing
• a.k.a caching

2
Bandwidth

• We think of the bandwidth of a network or link
  as the number of information bits that can be
  transmitted over it in a certain period of time
  (e.g., bits per second).
• The bandwidth of a link is really the frequency
  range tolerated by the channel without major
  attenuation.
• Telephone line is 3000 Hz (300Hz to 3300 Hz)
• Available bandwidth depends on the rate at which
  channel can change stored energy.
• We can model waveforms as sums of sine waves of
  different frequencies.
• Channel attenuates and delays each frequency
  component differently, causing distortion.

3
Bandwidth

• Signals are run through a low-pass filter, and a
  signal can have V discrete levels.
• Maximum data rate of a noiseless channel of
  bandwidth B is 2B log V bps (Nyquist,
  1924).
• Keep increasing V to achieve higher data
  rates?
• Regardless of V, the maximum data rate (capacity)
  of a noisy channel with bandwidth B and
  signal-to-noise ratio S/N is
• C B log (1S/N) bps (Shannon, 1948)
• S is the average signal power and N is the
  ave. noise power
• Example For a telephone line,
• B is 3000 Hz, with a typical S/N ratio of 1000,
  so C is 30Kbps or so
• We can achieve higher capcity only by increasing
  S/N!

2
2
4
Latency

• Latency is the time it takes for an information
  bit to propagate through a link or network.
• Latency Propagation Transmit Queue
• Propagation Distance / Speed of light
• Speed of light 3 x 10 meters/sec in a vacuum
• 2 x 10
  meters/sec in fiber
• Transmit Size in bits / Bandwidth
• Queue function of congestion in link or
  network
• We can reduce Transmit and Queue components using
  higher link speeds, but we cannot reduce the
  speed of light!
• Long distances mean long latency!

8
8
5
Bandwidth-Delay Product
Low-speed link
S
R
high-speed link
infinite bandwidth link

• Links stretching long distances have large
  storage capacity.
• Problem How to provide feedback to senders how
  do we handle latency?

6
Bandwidth-Delay Product

• The amount of data stored in the link
• Think of a link as a pipe the latency is the
  length of the pipe and the bandwidth is its
  diameter
• The BD product gives the volume of the pipe
• E.g., a channel of 50 ms latency and just 45 Mbps
  bandwidth can hold 2.25 million bits
• (as many as a PC of early 80s could store!)
• We are moving to Gigabit networks!
• Big bandwidth and big distances gt
• big aggregation and big memories at hosts

7
Implications

• Latency is the problem as the bandwidth in the
  network goes to infinity!

8
Why Web Caching
(0)
Client
Problems

• Content server is overwhelmed by clients
• Content server may be very far from client.

9
Web Caching with Client-based Hashing
(0)
Client
Limitations

• Client is modified with hash
• One cache per object
• Search uses URLs

10
Web Caching with Client-based Hashing
Client
(1)
Added Limitation

• Content server is hot spot

11
Web Caching with Hierarchies
(0)
Client

• Advantage Content server is buffered from
  requests
• Limitations
• Caches are manually organized
• Hierarchy defined per system or per URL
• Hierarchy is independent of congestion and load

12
Web Caching with Hierarchies
Client

• Limitations
• Added variable latency retrieving object
• Congestion due to search among siblings
• Client must choose the cache (static choice or
  require performance data at client)

13
Web Caching with DNS-based Hashing
(0)
Local DNS
Client
DNS

• Limitations
• Cache chosen with respect to local DNS, not
  client
• Latency and congestion limitations incurred by
  cache hierarchy

14
Web Caching with DNS-based Hashing
(0)
Local DNS
Client
DNS

• Limitations
• Cache chosen with respect to local DNS, not
  client
• Latency and congestion limitations incurred by
  cache hierarchy

15
Summary of Limitations

• Special DNS servers or modified browsers needed
• Special DNS servers receive all requests
• Search for cache-client match must be done at the
  time the request is received
• Either hashing is independent of network
  congestion or performance information must be
  sent to special DNS servers or clients
• Caches must be (manually) organized into a
  hierarchy
• DNS-based hashing optimizes with respect to local
  DNS, not the client
• Latencies incurred getting responses can be long
• Hierarchy of caches incurs additional congestion

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)