Transcript Packet Network Performance

Frame Relay Performance
Peter Hicks
Acting Rapporteur Q2/17 “ Data Network
Performance”
Tel: + 61 3 9253 6308
Fax: + 613 9253 6777
Email: [email protected]
Slide 1
Why worry about Network Performance
• Importance:
– Network Management & Planning
– Both customers & network operators have an interest
(contractual arrangements / SLAs)
• Benefits of understanding performance
– Able to provide Performance Guarantees for customers
– Accurate and cost effective dimensioning and network
provisioning
– Customer reporting
– Compliance with International Standards / National Regs.
Slide 2
ITU-T Performance Objectives Framework
• End to end performance
• 3 x 3 matrix
Network
(X.25, FR, ATM, IP)
TERM.
TERM.
Speed
Criteria
Speed
Accuracy
Dependability
Function
Access
•Call setup delay
Access/call setup
Info Transfer
Disengagement
call clearing
• Call setup error
probability
Information
Transfer
• Residual error rate
•Packet, Frame,
• User information
cell transfer delay
misdelivery prob.
•Throughput
Disengagement
• Call Clearing
(Delay)
• Premature
disconnect
probability
• Call setup failure
probability
• User information
loss probability eg Frame Loss
Ratio
Dependability
Only the Info transfer
stage is applicable to IP
networks
• Call clear failure
probability
Availability (function of the primary parameters
Slide 3
Accuracy
Availability
Packet Network Performance
• ATM, Frame Relay, X.25 all examples of Packet Switching
technologies
• Connection oriented; simultaneous circuits (virtual
circuits) able to be supported on a single access line
• For FR and ATM, no acknowledgment of frames / cells
sent into the network (requires transport layer protocol
to ensure end to end data integrity)
• Can essentially use the same techniques for measuring
and quantifying performance
• Performance (Loss & delay) is very dependent on the
network architecture, the capacity provided within the
network (buffering and transmission trunk speed) and
traffic loading.
Slide 4
What are the important parameters that can
be readily measured?
• X.25:
– throughput and packet transfer delay, Call setup delay
• Frame Relay: for CIR or EIR traffic
– Frame transfer delay , Frame delay jitter & Frame loss ratio
• ATM
– Cell transfer delay, Cell delay variation, Cell loss ratio
• IP (best effort / connectionless)
– packet transfer delay, traffic flows
– packet loss ratio
– packet delay jitter
Slide 5
Frame Relay Performance Parameters
• User information transfer performance
parameters for the FR PVC services defined in
X.144
• Key Primary Performance Parameters:
– User information Frame Transfer Delay
– Frame Delay Jitter
– Frame Loss Ratio
– Residual Frame Error Ratio
– Extra Frame Rate
Slide 6
Frame Relay Performance Parameters
• Connection set-up performance parameters for
the FR SVC services defined in X.145
• Key Performance Parameters
– Connection Set-up Delay
– Disconnect Delay
– Release Delay
– Connection set-up error probability
– Connection set-up failure probability
Slide 7
Frame Relay Performance Parameters
• Frame Transfer Delay
– time taken for a frame to traverse the network. Time taken
commences when the first bit of frame is transmitted and
ends when the last bit of the frame is received
• Frame Loss Ratio (FLRc or FLRe)
– The ratio of lost frames to total number sent for either the
committed or excess traffic streams
• Frame Delay Jitter = FTDMax – FTDMin
• Residual Frame Error Rate
– The rate of errored frames arriving at the destination
• Extra Frame Rate
– The rate at which extra frames which were not part of the
source traffic are detected in the destination traffic stream
Slide 8
Frame Relay Performance Parameters
• Frame Based Conformant Traffic Distortion:
– distortion from the traffic contract, measure of clumping or
spacing of traffic bursts
• Connection Set-up Delay
– Time interval between the occurrence of a Set-up message
and the occurrence of the corresponding return Connect
message
• Dis-connect Delay
– one way delay based on the transport of the a disconnect
message from the clearing to the cleared end terminal.
Slide 9
FR Performance Objectives
• Objective values for frame transfer delay, frame loss
ratio & frame delay jitter specified in ITU-T
Recommendation X.146
• End to end objectives (not including contributions of
the access line) apply to an international frame relay
data connection.
• National Network (portion) allocated 34.5% of the end
to end objective for FTD & FLR
• International Network (portion) allocated 31% of the
end to end objective for FTD & FLR
• 4 Quality of Service classes specified
Slide 10
X.146: FLR, FTD FDJ Objectives
Service
Class
0
Slide 11
FLR
1
No upper
bound
< 1 X 10-3
2
< 3 X 10-5
3
< 3 X 10-5
FTD: 256
byte Frames
No upper
bound
95 %
< 400 ms
95 %
< 400 ms
95 %
< 150 ms
FDJ
Not applic.
95 %
< 52 ms
95 %
< 17 ms
95 %
< 17 ms
FR Performance Objectives - Notes
• All values are provisional and they need not be met by networks
until they are revised (up or down) based on real operational
experience. The FTD objectives apply edge-to-edge.
• For FTD performance, all objectives apply to frames of size 256 (i.e.
to frames with user information fields of 256 octets). If frames of
size 128 are used to estimate compliance with these objectives, then
the following tighter 95th percentile objectives of for FTD should be
used; 380 ms for classes 1 & 2, and 130 ms for class 3.
– Frame Relay Forum have specified 128 octet frame size
– Networks with high speed backbone should meet objective when
using 512 frame size
• In the case of service class 3, if the international portion route length
exceeds 9300 km, an allowance of 6.25 ms per 1000 km of route
length is allocated to the international portion.
Slide 12
Frame Transfer Delay
• Highly sensitive to network topology/architecture,
internode trunk transmission speeds & traffic levels
• Can predict delay by simple model
• Can measure using echo / loop back techniques
– avoids use of synchronised real-time clocks at remote sites
– simple to setup, approach favoured / adopted by ITU
– extensive testing of X.25 (and Frame relay & ATM)
• Can also use OAM techniques specified in X.148 and
FRF.19 (Requires OAM to be implemented)
Slide 13
Impact of performance objectives on
network design
• The transfer delay performance objectives and
geographic span impose a maximum transit node
limit .
– Network Architecture and Infrastructure impact
• For example the mean frame transfer delay (Class 3)
objective across a national FR network (excluding
the customers access lines) is specified in
Recommendation X.146 as 34.5% of 150ms, and can
be used to calculate the maximum number of nodes
that a frame can transit.
Slide 14
Simple model to calculate transit delay
Core
Switch
Access
Switch
DTE
t Propag
Core
Switch
t Propag
t Propag
t Clocking
N
L+
tP
t Node
t Clocking
N
(k-1)L+ (k-1)
Access
Switch
t Propag
t Node
t Node
t Clocking
Core
Switch
t P +KN
N
t Node
t Clocking
L+
tP
DTE
t Propag
t Clocking
N
Notes
1. Model can be applied to X.25, Frame Relay, ATM or IP networks
2. Propagation delay may or may not be a dominant component of the overall
delay. This depends on distance, trunk transmission speeds and pkt size.
Slide 15
Transit Delay Model
• The model consists of a concatenation of nodes and internode
transmission links.
• Each FR switch can be characterised by a mean (or worst
case) processing delay of N ms.
• The transmission time across the internode trunks is
dependent or the size of the data frame and the transmission
speed and can also be expressed as a fixed delay of L ms.
• The propagation delay is distance dependent (5ms/1000 km)
but can be readily calculated for each internode transmission
section as tpi. Propagation delay may or may not be a
significant component of the overall delay.
• Hence overall delay depends on distance (propagation delay),
trunk transmission speeds & frame size.
Slide 16
Transfer Delay Model (cont)
• Using this model an active connection can be
shown as:
–a series of (k-1) transmission links (l1 to lk-1)
through k switching nodes (n1 to nk)..
–each link li has a clocking delay of L ms
–each link has a propagation delay of tpi &
–each switch ni has a processing & queuing
delay of N ms.
Slide 17
Expression for mean transit delay
• The mean or alternatively the upper-bound - worst case packet
transfer delay across the network is readily calculated as
mean delay approx = (k-1)L + Propagation Delay + kNmean
worst case delay < (k-1)L + Propagation Delay + kNworst-case
• With a high speed backbone (34 Mbits/s) the transmission (clocking)
_
delay for a 1024 byte data
Frame is 240 ms. This delay reduces to
53ms if the transmission backbone is 155 Mbits/s.
• For a 48 byte Frame the clocking delay is approximately 3ms. The
switching and queuing delay (the variable parameter) through a high
speed ATM/FR switch is of the order of 1ms.
• Over long distances the dominant factor will be the propagation
delay (Melbourne to Perth 17 ms, Melbourne to Sydney 5ms, Perth to
Brisbane 28 ms).
• For “old style” X.25 networks with low speed (64kbit/s) transmission
trunks, switching/clocking delay may dominate
Slide 18
Expression to calculate maximum
number of switching nodes
From the above expressions we can also derive an
expression for the maximum number of switching (or
routing) stages within a network
( Delay Objective + L – Prop-delay )
Max number of hops k =
(L+ N)
For example for a National FR network, delay objective = 52ms
Assume Geographic span 4000 km -> Prop Delay = 20ms
Frame size 256 Bytes, Trunk transmission 34 Mb/s -> L = 61ms
For N = 2 ms: k=15.6 maximum node of switches = 15
Slide 19
Clocking delay for various transmission
rates and frame sizes
Transmission
Speeds
64
128
256
512
1024
1544
2048
34368
44736
155520
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
kbit/s
Frame Size (FR Information Field)
64 bytes
8 ms
4 ms
2 ms
1 ms
0.5 ms
0.35 ms
0.25 ms
16 ms
12 ms
3.5 ms
128 bytes
16 ms
8 ms
4 ms
2 ms
1 ms
0.68 ms
0.5 ms
31 ms
24 ms
7 ms
256 bytes
32 ms
16 ms
8 ms
4 ms
2 ms
1.35 ms
1 ms
61 ms
46 ms
13 ms
512 bytes
64 ms
32 ms
16 ms
8 ms
4 ms
2.67 ms
2 ms
120 ms
92 ms
27 ms
1024 bytes
128 ms
64 ms
32 ms
16 ms
8 ms
5.3 ms
4 ms
240 ms
184 ms
53 ms
2048 bytes
256 ms
128 ms
64 ms
32 ms
16 ms
10.6 ms
8 ms
480ms
367 ms
106 ms
Table 1
Clocking delay for various transmission rates and frame sizes.
The frame size refers to the size of the FR information field.
Also a 2 byte FR header plus a 2 byte FCS are assumed
Slide 20
Effect of transmission delay
and frame size on FTD
• Consider a national network which has a geographic span of
4000 km, consisting of 8 switching stages and inter-trunk
transmission speeds of 2Mbit/s. Each switch contributes 1ms
of queuing delay. Propagation delay 5ms / 1000km.
• For a 256 octet test frame, each trunk will contribute a clocking
delay of 1 ms. (see Table 1). The total FTD is calculated as 8 x 1
ms + 4000 km x 0.005 + 7 x 1 ms = 35 ms. This network meets
the national portion FTD objective allocation of 51.75 ms for
Class 3 FR Services.
• For a 512 octet test frame, each trunk will contribute a clocking
delay of 2 ms. (see Table 1). The total FTD is calculated as 8 x 1
ms + 4000 km x 0.005 + 7 x 2 ms = 42 ms. This network meets
the national portion FTD objective allocation of 51.75 ms for
Class 3 FR Services.
Slide 21
Effect of transmission delay
& frame size on network architecture
• Question: Can a network with 10 switching stages
and a trunk speed of 1.544 Mbit/s meet the national
portion FTD objective allocation for class 3 if a frame
size of 512 octets is used?
– what is the maximum frame size allowed in order to meet
the objective
• Question: Show that that only in the case where the
number of switching stages exceeds eight (8), and the
inter-node trunk transmission speed is than 1.544
Mbit/s or less will the national portion FTD objective
be exceeded when the test frame size is 512 octets
Slide 22
Practical Measurement of transit delay
• Accurate measurement of transit delay requires
synchronised real time clocks located at appropriate
locations. Very expensive
• alternative technique / practical low cost method
required.
• measure the round trip delay time of a 256 byte test
frame sent to an echo facility or loop back
– echo facility receives a frame and retransmits the frame on
the same virtual connection
– echo technique standardised in Rec X.139 & ( X.148)
• Use OAM (FRF.19 frames) techniques as per X.148
Slide 23
Measurement of Network Transit Delay
using echo technique
X.36
FR network
X.36
Test DTE
Access lines can be at different speeds
Ideally echo device retransmits after set period of time.
Use 256 octet test frame.
Slide 24
Echo
device
Measurement of round trip delay time
• Define:
Tr = round trip delay time to the echo facility
Td = access line transmission delay (1ms for 256 byte
frame transmitted at 2.048 kbit/s)
Tnw = Network transit delay
Tech = Echo facility delay
• Assuming the echo facility is connected to
destination pkt exchange by 2.048 kbit/s line
• Tnw ~ Tr/2 - 2Td - Tech/2
Slide 25
Application of echo technique to measure
national & international transit delay
• National network delay can be made by locating echo
device at the international gateway exchange
• establish logical channels to national gateway
exchange and to a destination international gateway.
Define
• t1 = round trip delay time to national gateway
• t2 = round trip delay time to destination international
gateway
• tint = (t2 - t1)/2 - tgw
where tgw = transit delay of the destination gateway
Slide 26
155 Mbit/s Trunk between core switches
Brisbane
34 Mbit/s Trunk from core to access switch
Core
Switch
loopback
(128 kbit/s)
Legend for transmission links:
Access
Switch
loopback
(128 kbit/s)
d = distance [km]
tp = propagation delay [ms]
d = 1000 km
tp = 5 ms
Access
Switch
Perth
d = 500 km
tp = 3 ms
Core
Switch
Canberra
Core
Switch
Sydney
Core
Switch
Access
Switch
B
Access
Switch
d = 500 km
tp = 3 ms
d = 1000 km
tp = 5 ms
d = 3500 km
tp = 17 ms
loopback
(128 kbit/s)
Melbourne
Core
Switch
Access
Switch
B
loopback
(128 kbit/s)
Arrangements for Performance
testing for ATM & FR
Slide 27
A
Access
Switch
Access
Switch
Central Measurement
Test Equipment
A
256 kbit/s access
loopback
(128 kbit/s)
loopback
(128 kbit/s)
Some Frame Relay Transfer Delay Results
• FR traffic test source located at Melbourne,
• used 512 octet test frame
• one way delay to other capital cities
Sydney (1000 km)
Brisbane (2000 km)
Canberra (500km)
Perth (3500 km)
11 ms
18 ms
11 ms
24 ms
• each switch contributes in the order of 1 - 5 ms
• end to end delay dominated by propagation delay,
but switching & clocking delay of the edge / access
switches make noticeable contributions
Slide 28
Frame Loss Ratio Performance
• Rec X.146 specifies 1 X 10-3 for Class 1 network
• Measurement of Frame loss ratio for CIR traffic
– mean monthly figure < 1 X 10-5
– worst case 4 X 10-5
» Have we over dimensioned the network? Is our Class 0 network
providing Class 3 service?
• Loss ratio very dependent on network dimensioning
and traffic levels
• Also noting the large number of small frames (voice
and TCP acknowledgment) causing some functional
processors to be very heavily loaded. If the switch
becomes overloaded it discards frame
Slide 29
What’s new from Q2/17
• New Recs covering:
– FR Network Availability
– Metrics for FR/ATM Service Inter-working
– Performance of IP over FR
– FR OAM
Slide 30
FR Network Availability
• Availability is the percentage of time that the network
can successfully transfer frames. Availability is a
Key parameter often specified in SLAs
• Traditional ITU approach is to choose a significant
primary performance parameter (eg FLR) and assess
the performance of a connection against a defined
threshold for that parameter.
– For example: If the FLR > 10% over a period of time the
connection is declared to be unavailable.
• New Rec X.147 provides a number of options for
assessing availability – based on use of OAM Frames
or Status messages
Slide 31
Availability vs Connectivity
• Availability (ITU)
– represents the point at which the IP service is so bad as to be
unusable: eg. extremely high packet loss will impact on the
achieved transfer rate (FTP or HTTP) but the transport layer
protocols will still work.
– (a digit bearer is consider unavailable if BER>10-3 for 10
consecutive seconds : Rec G.826)
• Connectivity (IETF IPPM)
– defines the period when there is no working route between
source and destination - nothing gets through.
• Perhaps we need both
Slide 32
Performance of IP over FR
• What is the performance of an IP network when the
backbone infrastructure (connectivity) is provided
by FR connections
• Can the IP service classes defined by Y.1541 be
supported?
• Propagation delay dominant for long distances.
Difficult to achieve a user-to-user IPTD of 100ms on
long IP Paths.
• New Rec X.FRIP provides guidelines for use of
frame relay as the lower layer transport.
Slide 33
Metrics to characterise FR/ATM
Service Interworking Performance
ATM
DTE
FR
DTE
ATM
FR
IWF
The ATM DTE has no
knowledge that it is
talking to a FR DTE.
The FR DTE has no
knowledge that it is
talking to an ATM DTE.
FR / ATM Service Interworking
Slide 34
FR/ATM Service Interworking Metrics
• End-to-end performance or the performance of
the IWF can be characterized by the following
user layer parameters:
–Data Block Delivery Ratio
–Data Block Transfer Delay
–Data Block Delay Jitter
• Parameters are independent of the FR or ATM
Traffic Contracts
Slide 35
FR OAM
• SG13 developed I.620 (1998) covering FR OAM
–Only basic functionality defined - detection of fault
conditions using loop-back frames
–SG13 have agreed to withdraw I.620 in favour of
FRF.19
• Frame Relay Forum have developed FR.19
–Extensive capabilities to monitor primary
performance parameters FTD, FLR (Data Delivery
Ratio) and fault detection
• For completeness of the FR Recommendations
propose new Rec X.FROAM (text to be technically
aligned with FRF.19)
Slide 36
Any Questions?
Slide 37