2. Units and scales
Be familiar with the units & scales involved
Time
Milliseconds : 10-3 (or ms)
Microseconds : 10-6 (or μs)
Nanoseconds : 10-9 (or ns)
Units
MB = megabytes (unit of storage)
Mb = megabits (unit of transmission)
Time is the key to operation & measurement
3. Time nomenclature
Asynchronous
No time relationship between two end-points
Synchronous
Clocks at each end-point are synchronized
Isosynchronous
Synchronization is achieved by start and stop bits in
the data sent between end-points
Self-clocking
Example: Manchester encoding – clock embedded
with the data
4. Time nomenclature
RTT or Round-trip Time
The time to send a datagram from one entity to another and then
back (the two one-way trip times are not necessary equal)
Bit time – see bandwidth (also called “wire time”)
Jitter
The variation in periodicity of packet or bit arrival (usually due to
network congestion or routing changes)
NTP
Network Time Protocol – A protocol to synchronize a computer
clock to a external time source (see RFCs 778, 1119, & 1305)
5. Bandwidth
Stated as megabits per second or mb/s or mb
sec-1
The bit time is 1/bandwidth
Bandwidth measures capacity not speed
Bytes are the unit of storage, not bandwidth
and are usually written with B, while bits are
written with b
6. Delays
Propagation delay
Function of distance and the type of media
Queuing delay
Mostly a function of traffic through the forwarding
device, queuing discipline, & available memory
Processing delay
Complicated due to routing/switching architecture,
hardware/software processing, packet inspection,
filtering, and so on
7. Delays
Transmission delay
A function of bandwidth (sometimes called the
wire time)
The time it takes to put a unit of data completely
onto the transmission media ( 1
𝑏𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ
𝑥 𝑏𝑖𝑡𝑠 𝑝𝑒𝑟 𝑢𝑛𝑖𝑡 )
Latency
End-to-end total delay
8. Time revisited
Measuring time on computers
Time intervals (how long does it take)
Absolute time (time stamps)
Granularity
Drift
Jiffies – time between two successive clock ticks
on a computer (LINUX)
See LINUX man page on LEARN/Blackboard
9. About time and other things
Variance of times for a repeated operation is
many times more important than the average
time (we will see this with the TCP RTT
estimator)
Time measurements can be synchronous to
some external source or relative to the entity
making the measurements
It is important to remember all of these terms
and how to use them
10. Computer vs Network Time Scales
Distinct Time Scales
Processor
Network
Interactive devices
What this means
Network is slow from
the CPU viewpoint
Network times may
or may not affect the
user experience
1 ns 1 μs 1 ms 1 s
Time Scale ~1 GHz CPU
Transmission delay
Prop delay SwRI-
UT Austin
Integer add
Screen refresh
Log scale
12. More definitions
Utilization
Channel utilization is the percentage of capacity
used to transfer actual data (derived from rate of
sending, overhead bits, acknowledgement
datagrams)
Throughput
Rate of successful message delivery over a
channel (retransmissions reduce throughput)
13. And a few concerning
statistics
Inter-arrival time
Time between successive arrivals of messages at
a network device; usually these times are drawn
from some distribution function (example:
Poisson)
Heavy-tail – example: Pareto distribution
Self-similar
Statistics look the same over different time or
spatial scales
15. Message nomenclature I like to
use
Physical layer: bits/symbols
Link layer: frame
Network layer: datagram
Transport layer: segment
Application layer: message
Layer independent: protocol data unit (PDU)