- The link layer in Ethernet adds 18 bytes of header and trailer. What
does this mean for the maximum possible
throughput of data for a 10Mb Ethernet? A 100Mb Ethernet? Does this align with
real life throughput? Explain.
Suppose, for the purposes of this exercise, that there are no other overheads
(this is false). On top of a 1500 byte Ethernet frame 18 bytes are used in the
encapsulation. This is 1.2%, so a 10Mb Ethernet has just 9.88Mb of available
bandwidth, and a 100Mb Ethernet has just 98.8Mb available.
In real life we never get those rates. There are many reasons:
- There are actually more overheads consisting of an inter-packet gap
(equivalent to about 12 bytes) and a synchronisation preamble (8 bytes), giving
a total of 38 bytes of overhead;
- there are potentially collisions;
- sometimes there is overhead in the machines reading or writing the data on
the net;
- there is overhead from higher layers.
- The network layer in TCP/IP adds a 20 byte header. What does this mean
for the maximum possible throughput of TCP data for a 10Mb Ethernet? A 100Mb
Ethernet? From your understanding of encapsulation, could this layer dispense
with this header? Explain.
This will reduce maximum throughput by another 1.3%, giving 9.86Mb (98.6Mb).
But again, other factors are more important:
we will see later that TCP has an elaborate technique for maintaining
reliability: this costs in bandwidth.
This header cannot be dispensed with, it is the basis of the TCP protocol's
reliability. Besides trimming a few header bytes will not make a bit
difference in available bandwidth: the other factors are much more important.
- A wireless network is described as being 11Mb, but when used can never
seem to get more than half that. Explain why as (a) a network support
officer, (b) a marketing officer.
(a) This is all about overheads of various types, from encapsulation to
protocol overheads. Wireless networks have to be particularly careful about
the correct transmission of bits in the electrically noisy world we live in
and this translates to bandwidth overhead.
(b) I'd like to see any coherent answer to this!
- Find other examples of encapsulation in life.
This is only limited by the imagination
- a person on a bicycle on train
- a person in a car in a train in the channel tunnel
- a game of pass the parcel
- an apple in a bag in a box in a palette in delivery truck
- Compare and contrast the OSI model against the Internet model.
We can match the various layers against each other, for example Presentation in
ISO is subsumed by Application in the Internet model, but more interesting is
to see how real-life networks (in particular the Internet, mobile phone, fixed
line phone and others) align against the two models (or not, as the case may
be).
- In real implementations of the Internet model (and others) the layers
are sometimes blurred to aid efficiency. Discuss the pros and cons of doing
this.
Pros include
- speed of the resulting code;
- the IP has more upward and downward dependencies than a pure layered system
would recommend anyway (e.g., TCP headers depend on the contents of the IP
layer headers).
Cons include
- code correctness;
- code maintainability;
- ability to reimplement and replace layers independently.
And there are many more. Real life implementations tend to blur the layers
when the cost in speed is too great to ignore, but otherwise we should keep the
layers separate as much as possible to aid clear coding.
- Read about the ISO implementation of the seven layer model (ISODE,
actually just layers 4 to 7) and make notes on its main features.
The standards are available.
Also see RFC2126
- Consider broadcast TV. Classify its parts according to (a) the
OSI and (b) the Internet models. Which is a better match?
(a)
- Physical. Radio waves. Frequencies and powers.
- Data Link. Encoding of analogue or digital signal on the waves. Neither
have flow control, only digital worries about error correction.
- Network. No routing needed: just blast out the signal. If you take the
wider view of TV transmission you might want to think of how the signal is sent
to the various transmitters dotted about the countryside, but again routing is
all statically decided by the engineers designing the system.
- Transport. Analogue does nothing here, while you might regard the
multiplexing of several digital channels on to one TV channel as part of the
Transport. Whether it is connection oriented or connectionless is an
interesting debate.
- Session. Nothing.
- Presentation. All decided and fixed at design time. In particular there is
no variation allowed in digital TV transmission.
- Application. The TV programmes!
(b)
- Link. Radio waves. Frequencies and the encoding on those waves.
- Network. No routing.
- Transport. Digital multiplexing.
- Application. The TV programmes!
Perhaps the Internet layering is slightly better as there are no real session
or presentation issues, but it would be useful to have the physical and link
layers separate to allow us to describe the same transmission system on
different frequencies. This is just Tanenbaum's 5 layer classification again.
- The IEEE split the OSI Data Link layer into a logical link control
(LLC) sublayer and a media access control (MAC) sublayer. Read up on this and
discuss how it fits in with the OSI and Internet models.
The IEEE argue that a standard (such as 802.3 for Ethernet or 802.11 for
Wireless) more naturally comprises the PHY (physical) part plus MAC (how to use
the physical part) rather than the physical part then access plus other link
layer stuff.
In brief: OSI say PHY + (LLC + MAC); while IEEE say (PHY + MAC) + LLC,
and Internet says (PHY + MAC + LLC).
- When layering goes wrong: find examples (e.g., on the Web) where
insufficient attention has been paid to the presentation layer. Why do you
think that people neglect the presentation layer?
Examples abound, particularly if you use non-Microsoft products to view
Microsoft-tool generated information. Also when looking at non-English Web
pages.
Presentation was another one of those things that were ignored in the fledgling
Internet: as all the users spoke and wrote English (or variants, such as
American) it never came into consideration that different people might want
different things. A similar reason is why American websites are often the
international .com rather than the more proper .us: when
names were being devised, no thought was given to people outside of the USA.