Characteristics of Packet-Switched Networks; Protocols

Tom Kelliher, CS 325

Jan. 31, 2011




Read 2.1-2.3.

From Last Time

The Internet's edge and core.


  1. Packet-Switched network characteristics.

  2. Protocol layers and models.

  3. Security.

  4. History

Coming Up

Introduction to application layer, HTTP and FTP.

Packet-Switched Network Characteristics


Packet transmission delay model:


Nodal delay (one switch):

d_{\rm nodal} = d_{\rm proc} + d_{\rm queue} + d_{\rm trans} + d_{\rm prop}

Delay components:

  1. Processing delay: integrity checking, routing, etc.

  2. Queuing delay: Waiting in output buffer prior to transmission. Variable.

  3. Transmission delay: Getting the entire packet ``out the door.''

    Let packet contain $L$ bits and link transmission rate be $R$ b/s. Transmission delay is then $L/R$.

  4. Propagation delay: Time for one bit to traverse the medium between two switches.

End-to-end delay with $n$ switches along the route:

\sum_{i=1}^{n} d_{{\rm nodal}_i}

Using traceroute to see the route between two hosts:

% traceroute
traceroute to (, 30 hops max, 40 byte packets
 1 (  1.383 ms  1.552 ms  1.674 ms
 2 (  30.927 ms  232.192 ms  250.335 ms
 3 (
    67.405 ms  122.690 ms  140.896 ms
 4 (
    49.226 ms  85.665 ms  104.371 ms
 5 (
    159.112 ms  177.257 ms  195.600 ms
 6 (
    213.743 ms  257.386 ms  266.623 ms
 7 (
    275.783 ms  249.083 ms  247.324 ms
 8 ( 238.350 ms  229.374 ms (  219.795 ms
 9 ( 210.880 ms ( 202.249 ms ( 193.167 ms
10 (  184.294 ms (  175.361 ms  166.344 ms
11 (  157.196 ms  10.448 ms  160.945 ms
12 (  26.664 ms  6.023 ms  16.402 ms
BTW, Level 3 is a Tier 1 ISP.

Queuing Delay and Packet Loss

If packets arrive more quickly at the switch than we can send them, we have a couple problems:

  1. Packets will begin to queue up in the switch's buffers -- increasing queueing delay.

  2. If the buffer fills completely, packets will be dropped -- lost forever.

Traffic intensity is a metric used to describe queuing delay:


where $a$ is the packet arrival rate, per second.


Throughput and Latency

Two throughput measures:

  1. Instantaneous throughput -- throughput at a particular instant.

  2. Average throughput.

    Transferred 5 MB MP3 file in 12 sec.: $(5 * 2^{20} * 8) / 30 = 1.4~{\rm

The ``slowest'' link determines the overall throughput -- the bottleneck:


Internet resources are shared:


Latency: How long it takes the first bit to make it from end-to-end.

Rules of thumb:

  1. Latency matters for realtime applications: gaming, telephony. Throughput may or may not matter -- low-fi audio, no; video, yes.

  2. Throughput usually matters for file transfer.

  3. Neither particularly matter for email; maybe IM.

Protocol Layers and Models

Protocols (services) are layered on top of each other:



  1. Only the physical layers on two hosts communicate directly.

  2. Higher layers communicate through lower layers -- abstraction.

    Think of two heads of state communicating through their ministers, and the ministers communicating through under-secretaries.

  3. Going down, layers add headers with information specific to that layer (nested envelopes).

  4. Going up, layers examine and discard headers.

Layer synopses:

  1. Application layer: What ordinary think of the Internet as: HTTP, SMTP, FTP, etc.

    Unit of exchange: message.

  2. Transport layer: message transport service between hosts.
    1. TCP: connection-oriented service; guaranteed delivery; segmentation of messages; congestion control.

    2. UDP: connectionless service; delivery not guaranteed; no congestion control.

    Unit of exchange: segment.

  3. Network layer: segment delivery service.

    IP protocol. No guarantee of delivery. Routing services occupy this layer.

    Unit of exchange: datagram.

  4. Link layer: delivery of datagrams between two adjacent nodes.

    Sometimes delivery is guaranteed. The mediums between links may vary.

    Unit of exchange: frame.

  5. Physical layer: move frames from one node to the next.

    Unit of exchange: bit.

TCP/IP example:



The ``bad guys'' can:

  1. use viruses, worms, malware on web sites, etc. to take control of hosts.

    Botnets -- ``148,000 hosts added to botnets daily.''

  2. attack Internet infrastructure. DDOS attack:


  3. Read, modify, or delete packets.

  4. Masquerade as legitimate hosts.

    Hosts files, DNS cache poisoning attacks.


  1. The original ``Internet'' in 1969 consisted of four nodes.

    Today? Reachable hosts according to the Internet Systems Consortium:


    This is a gross undercount.

  2. An early IMP (router):


  3. A modern ``router,'' the Linksys WRT54GL:


Thomas P. Kelliher 2011-01-31
Tom Kelliher