Transcript Network Performance

Network Technology
COMP1664
Network Performance
Outline
•
•
•
•
•
•
•
Applications and Traffic Types
Performance Metrics
QoS/SLA
Reliability and Availability
Measuring Performance
Optimising Performance
Examples of networks and performance
Applications & Traffic Types
Analogue Telephony
Analogue sound
Bookings, reservations
Database transaction
Nightly backups - files
File transfers - batch
E-mail (inc. MM)
Text, sound, image, video
Web surfing, searching
Multimedia, real time
CAD, CAE, medical
Still Image
Internet telephony
Real time audio
Radio, VoD, TV
Real time audio, video
Electronic Data Interchange (EDI), Retail, Banking
Multimedia, data, text
Education (distance/e learning, uploads etc)
Multimedia, data, text
Types of Traffic
• Data / Files – binary, text (inc documents,
email, database queries, web-pages etc.)
• Audio – analogue, digital, varying quality
• Still Image
• Motion video - analogue, digital, varying
quality; sound-picture synchronisation
Audio and Video may be compressed
Performance Parameters and units
of measurement
•
•
•
•
•
•
•
•
Response time - sec, ms, μs
End-to-end or transfer delay - sec, ms, μs
Delay Variation (jitter) - sec, ms, μs
Throughput - bits/sec, transactions/min
Packet loss/error rate - fraction, percentage
Bit error rate (BER)
Call Blocking rate - fraction, percentage
Network access delay, transit delay etc.
How do loss and delay occur?
Packets queue in router buffers before being forwarded
• Packet arrival rate to link can exceed output link capacity
• Packets queue, wait for turn to be processed
• Memory capacity to hold the queue is finite
packet being transmitted (delay)
A
B
packets queueing (delay)
free (available) buffers: arriving packets
dropped (loss) if no free buffers
Four sources of packet delay
(for a single link)
1. Processing at Tx Node
(can be a computer or router)
- Operating system scheduling
- Check for bit errors
- Determine output link
2. Queuing
- Time waiting at output link
for transmission (depends on
congestion level of router)
transmission
A
propagation
B
nodal
processing
queueing
Four sources of packet delay (cont)
(for a single link)
3. Transmission delay:
• R=link bandwidth (bps)
• L=packet length (bits)
• time to send bits into link = L/R
4. Propagation delay:
• d = length of physical link
• s = propagation speed in
medium (~2x108 m/sec)
• propagation delay = d/s
transmission
A
propagation
B
nodal
processing
queueing
Note: s and R are very different quantities!
Tx delay - sensitive to technology. Prop delay - sensitive to distance
Delays on the Internet
Notice the delay variation – on which graph is it higher?
Global Packet loss – 30 days
Bandwidth (Bitrate)
First, some number representations and units
Powers of 10
Kilo 1000 (103)
Mega 1000000 (106)
Giga 1000000000 (109)
Tera (1012)
Fractions – e.g. seconds
Milli 10-3 (1/1000)
Micro 10-6
Nano 10-9
Pico 10-12
Analogue – measured in Hz,
Digital – bits/sec
In the computer world, we reckon in powers of 2.
210 = 1024 which is approximately 103 = 1000.
Typical BW Requirements
Analogue telephony
3-20 KHz
Digital voice
64 Kbps – 1.5 Mbps,
uncompressed
28 Kbps, 0.35 sec/page
Text (10 kbits/page)
Image (1 MByte)
Motion Video
1.544 Mbps - Digital Signal 1
(DS1)
25 Mbps to 2 Gbps
compression is essential
QoS requirements
Delay
(ms)
Jitter
(ms)
Loss
(%)
BW
(Mbps)
Voice
250
10
10
0.064
video
250
10
0.1
100
Compressed video
250
1
10-7
2-10
1 sec
-
10-7
2-10
Few sec
-
0
2-100
2-3 sec
-
0
10
Image
Data file
Transaction
Availability & Reliability
• Reliability R(t) : The probability that the network
will not break down over a period of time t
• Instantaneous Availability A: Probability of
network being available at a given instant.
• Availability A(t): The fraction of time that a
network will be available over a given period t
(e.g. 24 hours).
Calculation of Availability (A) & Reliability (R)
• MTBF – Mean Time Between Faults
• MTTR – Mean Time to Repair
• t = time period
R (t )  e
t / MTBF
MTBF
A
MTBF  MTTR
t

A(t )  A  (1  A)e AMTTR
Exponents calculator available at:
http://freeonlinecalculator.net/calculators/algebra/exponent.php?action=solve&x=2.71&expon
ent=-.876&submit=Calculate
Example: MTBF = 60 days, MTTR = 3 days, A = 0.952
0.99
1.2
A(t)
R(t)
0.98
1
0.97
0.8
0.96
0.6
A(t)
0.95
0.4
0.94
0.2
0.93
R(t)
0
t
1
2
5
10
Days
15
20
25
t
1
2
5
10
15
20
25
Days
A fault becomes more likely as the time frame considered is increased
Worked examples
1. MTBF = 10,000 hours, MTTR = 48 hours
Reliability R over 1 year (8,760 hours) = 41.64%
Availability A = 10000 / 10048 = 0.9952
A(1 year) = ? (see formula)
2. MTBF = 10000 hrs, MTTR = 168 hrs (1 wk)
A = 10000 / 10168 = 0.9834
A(8760) = 0.9834, A(350) = 0.9853
A(24) = ?
(see formula)
More examples
3. With MTBF = 20000, MTTR = 720
R(1year) = 64.53%
With MTBF = 30000, same MTTR of 720
R(1 year) = 74.67% (more reliable)
The higher the MTBF,
the higher the reliability and availability.
The lower the MTTR,
the higher the reliability and availability.
We want:
Long time to breakdown, Short time to repair.
Effect on reliability of mirroring of components
e.g. Two servers,
each one has MTBF = 1000, MTTR = 24 hours.
Application requires that only one need be available.
Instantaneous availability of one server: A = 0.9766,
Availability of any one of the two:
1 – (1 - 0.9766) x (1 - 0.9766) = 0.9994
I.e. The probability that at least one is available =
1 – probability that both will be in fail state at the same time
(Note, A = 1 would mean 100% available i.e. never fails)
A series of components
e.g. CPU, Ethernet card, cable, router etc.
All components need to be working to achieve
communication
CPU
NW Card
cable
router
R = R1 x R2 x R3 …
RCPU = 99%, REthernet card = 99%, Rrouter = 98%
System reliability = 0.99 x 0.99 x 0.98 = 0.96
I.e. The probability that none are in fail state =
the probability that all are available at the same time
Performance Analysis
When?
• Initial Design
• In operation
• Upgrading
– User needs
– Better
technology
What?
• Predict performance
• Identify Problems
• Suggest improvements
• Test proposed changes
Why do performance problems occur ?
•
•
•
•
•
•
Overload – temporary, congestion
Inappropriate choice of links, equipment
Component failure – e.g. router/switch, link
Lack of planning for applications in use
(traffic profiles, bursty traffic, statistical multiplexing)
Simultaneous reboot, broadcast storms
Poor tuning – buffers, timeout, retransmissions, packet size,
window size
Monitoring Performance
Performance monitoring and configuration inspection
tools include:
•
•
•
•
•
•
•
Ifconfig – interface configuration
ARP – address resolution
Ping – remote host availability, RTT
Traceroute – path taken
Netstat – packet statistics, loss rates
Omnipeek – packet statistics, protocols
NSLookup – DNS queries
Troubleshooting - find the bottleneck
• Consider all the stages in the source-destination
path
• Test/eliminate each link/component systematically
• Check - hardware/software/configuration
• Modern protocols are designed to reduce/avoid
congestion, but they need to be fine tuned.
Performance Prediction Techniques
• Measurement
• Mathematics – queuing theory
• Simulation – e.g. using OPNET or NS2
The Balance between
cost, performance and reliability
• Cost – H/W, S/W, staff, installation, maintenance
• Performance – quality of equipment, quality of
design, configuration settings / tuning
• Reliability – high quality equipment, security
measures – backups, mirroring, redundancy
Network Performance
Parameters and examples
Traffic Descriptors
Network traffic is in the form of frames, packets or messages.
To characterise traffic, we need to define:
• Arrival rate (packets per second)
or
Packet Interarrival time (seconds) = (1 / arrival rate)
• Packet size (bits or bytes)
Network Capacity
Depends on the network components – links, servers,
routers, switches etc, have limited capacity to
transmit or process traffic.
Capacity is measured in bytes/sec, bits/sec or
packets/sec
Performance Measures
•
•
•
•
Load, Throughput, Utilisation
Delay
Delay variation (jitter)
Packet Loss rate
Utilisation = load / capacity (aim for 70% or less).
If the load < capacity – all is well.
otherwise chaos! Long delays, packet loss etc.
Utilisation
=
Load / Capacity
For a link R = link bandwidth (‘link speed’) (bps)
E.g. R = 10 Mbps, Load on the link L = 6 Mbps
Link Utilisation = load(L) / capacity(R) = 6/10 = 0.6 = 60%
For a traffic flow, or specific packet
• ta = Interarrival time,
• a = Arrival rate = 1 / ta (packets/sec)
• PL = Packet length (bits)
• Transmission time (ts) = PL / R
E.g. R = 100Mbps, PL = 1500 Bytes = 12000 Bits
ts = 12,000 / 100,000,000 = 0.00012 Seconds = 12 µ seconds
• Utilisation = ts / ta = a PL / R
Utilisation - traffic flow example
Link speed (rate) R = 100Mbps (100,000,000)
a (Arrival rate) = 1,000 packets/S
PL (Packet length) = 500bytes * 8bits = 4,000 bits
ta (Interarrival time) = 0.001 Seconds = 1 mS,
ts (Transmission time) = PL / R = 4,000 / 100,000,000
= 0.00004 Seconds = 40 micro seconds
Utilisation = ts / ta = a PL / R
= 0.00004 / 0.001 = 0.04 = 4%
= 1,000 * 4,000 / 100,000,000 = 0.04 = 4%
Queuing delay
•
•
•
•
util ~ 0: average queuing delay small
Util  1: delays become large
Load > capacity: more “work” arriving than can be serviced. If
continues over a period of time, average delay tends towards infinite!
Aim for 70% or so.
Key to figure:
La/R = Arriving load / Capacity
Ethernet utilisation example (100 Mbps)
• Consider 6 PCs on a 100 Mbps Fast Ethernet
network, interconnected by a hub.
• Assume each PC generates 50 frames/sec on
average
• Assume average frame size = 1500 bytes
• Total load = (6 x 50 x 1500 x 8) bits/sec
=3,600,000 bits/sec = 3.6 Mbps
• Utilisation = 3.6/100 = 0.036 or 3.6%
Application
-specific
characteristics
Service level Agreement (SLA) – example
Availability: Our goal is to make the Global IP Network available to
customers, free of network outages, 100% of the time.
Latency: The average monthly latency (speed) of the North American
Global IP Network will not exceed 50 milliseconds, round-trip. TransAtlantic average monthly latency will not exceed 90 milliseconds,
round-trip. Trans-Pacific average monthly latency will not exceed 130
milliseconds, round-trip.
Packet Loss: The average monthly packet loss (reliability) of the
Global IP Network will not exceed 0.1%.
Jitter: The Average Jitter on the North American Global IP Network
will be 0.25 milliseconds or less. The Maximum Jitter on the
Backbone will not exceed 10 milliseconds more than 0.1% of a
calendar month.