Quality-of-service (QOS) parameters
Throughput
Throughput refers to the total bandwidth delivery capability of a network
measured in bit/s (that is, routing, flow control, congestion control, etc.)
and the requirements for both transmission and switching. In OSI, it enables
Transport and Network layer entities (that is, functional units in these layers)
to choose between alternative ongoing connections and, in the Transport layer,
for example, it is used to decide whether to multiplex a new connection over
existing Network-layer connections.
Transit Delay
Transit delay or the maximum delivery delay for data is also an important
parameter to end systems. It enables them to choose between alternative connections,
and also to allocate internal resources to satisfy the requirements of the particular
application supported (that is, interactive or file transfer).
Network connections priority
This parameter relates to the occasional congestion problems that arise in
packet networks. It enables
a connection that supports several applications to be degraded or discarded
as a result of this congestion. Priority is expressed by three subparameters:
- priority to obtain a network connection (for example, in order to obtain diagnostic information regarding the congestion),
- priority to hold and keep a connection until the completion of certain data transfers, and
- priority for certain types of data traffic over (for example, queued data-traffic.
Others
Other parameters such as acceptable cost, residual error rates, etc., are
also important considerations related to routing, management and maintenance
of the network connection.
16 - SLIDE NET 16 Throughput of X25 links
- links in series: throughput of slowest link
- links in parallel: depends on load sharing
- limiting factors for each link:
3.1 Data capacity of link
ie 48K,9.6K,4.8K,2.4K
data rate is reduced by overheads such as:
Packet Headers, Flags, bit stuffing.
3.2 Level 3
Assuming end-to-end significance (Public Network does not support D bit) the window of each virtual circuit may close if either DTE does not acknowledge packets fast enough or packets are held up in the network due to switches or links.
The throughput depends on,
- loop delay (depends on transmission queue on each line)
- the window size
Window Size
Window size recommended for Public Network for high throughput.
| highest DTE | Window size |
| 48000 | 7 |
| 9600 | 7 |
| 4800 | 4 |
| 2400 | 2 |
This limitation is per virtual circuit, so if this is a limitation the throughput
could be increased by setting up more than one virtual circuit.
Loop delay
depends on:
speed of DTEs at each end
loading on links
level 2 factors
3.3 Level 2
the window in either direction of a link may close if the DTE or DCE does
not acknowledge frames fast enough. A highly loaded line would limit the throughput.
A high error rate on the line would limit the throughput.
4 Limiting factors for switches,hosts and PADs
Limitations in processing power or memory could cause the switch to limit throughput by allowing windows to close or by use of RNR.
Maximum Achievable Throughput (including overhead octets at
levels 2 & 3
Assuming an average Network loop delay of 0.7s
/-------------------------------------------------------------------------------------------\
| Maximum data | Line Signalling Rate (kbit/s) |
| Packet size |----------------------------------------------------------------------------|
| octets | 2.4 | 4.8 | 9.6 | 48 |
|-------------------------------------------------------------------------------------------|
| 128 | 1.51(2) 2.4(4)| 2.27(2) 4.8(5)| 3.04(2) 9.6(7) | 4.16(2) 14.56(7) 48(24) |
|-------------------------------------------------------------------------------------------|
| 256 | 1.77(2) 2.4(3)| 2.88(2) 4.8(4)| 4.16(2) 9.6(5) | 6.48(2) 22.7 (7) 48(15) |
|-------------------------------------------------------------------------------------------|
| 512 | 1.95(2) 2.4(3)| 3.33(2) 4.8(3)| 5.16(2) 9.6(4) | 9.17(2) 32.09(7) 48(11) |
|-------------------------------------------------------------------------------------------|
| 1024 | 2.06(2) 2.4(3)| 3.64(2) 4.8(3)| 5.88(2) 6(4) | 11.64(2)40.74(7) 48(9) |
\-------------------------------------------------------------------------------------------/
the values expressed represent - throughput, kbit/s (window size) respectively
Assuming an average Network loop delay of 2s
/-------------------------------------------------------------------------------------------\
| Maximum data | Line Signalling Rate (kbit/s) |
| Packet size |----------------------------------------------------------------------------|
| octets | 2.4 | 4.8 | 9.6 | 48 |
|-------------------------------------------------------------------------------------------|
| 128 | 0.9 (2) 2.4(6)| 1.11(2)3.89(7)| 1.27(2) 4.44(7) | 1.43(2) 5.01 (7) 48(67) |
| | | 4.8(9) | 9.6(16)| |
|-------------------------------------------------------------------------------------------|
| 256 | 1.23(2) 2.4(4)| 1.67(2) 4.8(6)| 1.97(2) 6.91(7) | 2.39(2) 8.37 (7) 48(48) |
| | | | 9.6(10) | |
|-------------------------------------------------------------------------------------------|
| 512 |1.517(2) 2.4(4)| 2.23(2) 4.8(5)| 2.93(2) 9.6(7) | 3.89(2) 13.61(7) 48(25) |
|-------------------------------------------------------------------------------------------|
| 1024 | 1.72(2) 2.4(3)| 2.69(2) 4.84) |3.77(2) 9.6(6) | 5.52(2)19.32(7) 48(18) |
\-------------------------------------------------------------------------------------------/
the values expressed represent - throughput, kbit/s (window size) respectively
Public Network
..............................
x25 dte . . x25 dte
-- t1 -----?. -----------t3-------------? . ---- t1 ------?
. .
?- t2 ----- . ?----------t4------------- . ?--- t2 ------
. .
................................
2t1+t3 = time taken to transfer a full data packet between two
X25 DTEs
2t2+t4 = time taken to transfer an RR packet between two X25 DTEs
t3+t4 = network loop delay
assumptions
- both the sender and receiver are X25 DTEs
- The time for the receiver to accept, process and acknowledge a data packet is negligible.
- there is no queuing at the DTE/DCE interfaces (neither in the DTE nor in the network)
- The time for the sender to process a P(R) and prepare a new data packet is negligible.
- The data signalling rate, D, is the same at both X.25 interfaces.
- The maximum attainable throughput indicates the overhead octets introduced at x25 levels 2 and 3.
- Successive information frames transmitted on a customer access links and inter-exchange links are separated by a single flag.
- That P(R)s are transported by RR packets.
Calculation of maximum attainable throughput, D.
values of D were calculated using the formula :-
D= w * fm
--------------------
2(t1+t2) + ( t3+t4)
where w = window size
fm= length of I-frame carrying a full data packet
Calculations of the network loop delay (t3+t4) in table 1 were based on an
estimated network transfer time t3 of 270 ms for an 128 octet data packet in
what is expected to be a maximum length call in the UK packet-switched network.
For inter-network calls (table 2) the network loop delay was taken as :-
2 * (corresponding value in table 1) + 500 ms.
Values of T3+T4 are listed below:-
Max packet Network Loop Delay (t3+t4) ms
size (octets) table 1 table 2
| 128 | 475 | 1450 |
| 256 | 560.5 | 1621 |
| 512 | 731 | 1964 |
| 1024 | 1072 | 2644 |
Note: we cannot guarantee or assure throughput outside the Public Network network
Throughput related aspects of the quality of service which X25 users can expect
from Public Network (assuming End-to-end significance of P(R)
- for national calls, the use of a maximum data packet size of 128 octets and modulo 8 numbering will enable X25 users accessing Public Network at data signalling rates of up to 9.6 kbit/s to utilise up to the whole capacity of their access lines for a single virtual call. customers gaining access at 48 kbit/s will be able to achieve a maximum throughput of 14.6 kbit/s per call. Bearing in mind that in general, the X25 DTE will be capable of engaging in a number of simultaneous calls, the above quality of service is felt to be more than adequate, however if necessary the throughput for 48kbit/s access may be increased towards an upper limit of 40 kbit/s by using larger packet sizes up to 1024 octets.
- For international calls, the anticipated quality of service using a maximum Data packet size of 128 and modulo 8 numbering will be somewhat lower, X.25 users being able to devote 100%, 82%, 47% and 10% of the capacity of their access lines to a single call when using data signalling rates of 2.4, 4.8, 9.6 and 48 kbits respectively.
- The performance quoted in 2 assumes end-to-end signifi- cance, however where the foreign network only provides local significance Public Network users can expect to achieve similar throughput to those quoted for national calls in para 1.
- Bearing in mind that X72 currently only allows the use of a single packet
size (128 octets) the performance of an international connection in which
end-to-end significance was maintained could be improved by using large window
sizes (>7) and modulo 128 numbering.
17 - Public Network quality of service parameters
General
All delays are specified for normal busy hour load and expressed, where appropriate,
in terms of:
- the mean value
- the 95% probability value ie, the time interval in which 95% of packets
sent to the network will be responded to by the network.
Delays across Public Network are measured from the receipt at the Exchange (PSE) of the
last bit of the packet from the DTE to the insertion of the first bit of the
packet into the output queue of the destination PSE. All figures given only
apply to calls wholly set up within the National Public Network network.
17.1 Call request delay
The call set-up time is defined as the time from the receipt of the originating
exchange of the last bit of the call request packet to the placing of the first
bit of the call connected packet on the output queue of the calling DTE, but
excluding the time from the placing of the first bit of the incoming call packet
on the output queue to the called DTE to the receipt by the destination exchange
of the last bit of the call accepted packet.
The network has been designed to offer a mean call set-up time for a call request packet of length 38 bytes, including the time taken to progress and switch the packet and to record accounting and statistical information, routed over a connection consisting of 5 * 48 Kbit links in tandem, of less than 400ms and for 95% of such calls of less than 480ms.
17.2 DATA packet transfer delay
The delay of the DATA packet across Public Network may include one or several retransmissions
for error correction. This delay is variable and depends on the routing, the
trunk speed, traffic loading etc.
The network has been designed to provide a mean delay of a DATA packet of
(3+128) octets across Public Network of less than 360ms and the 95% probability of less
than 400ms.
17.3 Established Call Clear
The network may clear a call during the DATA TRANSFER phase because of congestion
by sending a CLEAR INDICATION packet to the DTE. The network has been designed
to ensure the probability of this occurrence should not exceed 6E-6 per second.
17.4 Spontaneous Resets
The network may reset the call because of congestion by sending the appropriate
indication to the DTE. The network has been designed to ensure the probability
of this occurrence should not exceed 6E-6 per second.
