CALL PERFORMANCE STUDIES ON THE ATM FORUM UNI SIGNALLING IMPLEMENTATIONS

Transcription

1 CALL PERFORMANCE STUDIES ON THE ATM FORUM UNI SIGNALLING IMPLEMENTATIONS Su Kwe Long, R. Radhakrishna Pillai 1, Jit Biswas 1, and Tham Chen Khong Department of Electrical Engineering National University of Singapore 1 Kent Ridge Crescent, Singapore gisskl@sgh.gov.sg; {pillai,jit}@iss.nus.sg; cktham@cesun1.ee.nus.sg KEYWORDS Computer systems, Telecommunications, Emulation, Performance analysis, Communications ABSTRACT Asynchronous Transfer Mode (ATM) is widely accepted as the transfer mode for Broadband Integrated Services Digital Networks. Since ATM networks provide connection-oriented services, signalling is an essential component of these networks. The performance of the signalling system is crucial in determining the scalability of the network. Call performance measurements on two different ATM switches were carried out using a broadband network analyser. The performance measures considered are call setup and release delays and the call throughput of the switches. Calls are generated according to a deterministic/poisson distribution and various components of the call setup delay are analysed. The probability distribution of the delays are obtained and the correlation between the setup delay and its different constituents are established. It is observed that the delays at the calling and the called user entities of the UNI form a major component of the call setup delay. The average call setup time with the switches are found to be 21 and 27 msecs respectively. The call setup delay does not seem to increase linearly with the number of ATM switches involved whereas the call release delay increases linearly with the number of ATM switches involved. With simultaneous call arrivals, though with higher delays, both the switches are found to establish calls at the rate of 1 calls/sec. INTRODUCTION Asynchronous Transfer Mode (ATM) is widely accepted as transfer mode for Broadband Integrated Services Digital Networks (B-ISDN). Since ATM networks provide connection-oriented service, signalling is an essential component of such networks. The performance of the signalling system is crucial in deciding the scalability of a communication network. Today several switch vendors have implemented the ATM Forum UNI specifications for signalling. This paper reports the call performance studies on the ATM Forum UNI 3. signalling implementations. The measurement was carried out on Fore ASX-1 and ASX-2 ATM switches using a Hewlett-Packard Broadband Series Test System (BSTS). The BSTS was connected to the ATM switch through OC-3 links operating at 155 Mbps and call requests were generated using its UNI emulation function (Figure 1). The call requests followed a Deterministic/Poisson distribution. The signalling response from the switch was analysed to determine the delays involved in call setup and release. The call throughput of the switch is determined by stressing the switch with simultaneous call requests. Tx Rx Tx R Rx Attenuator ATM Switch HP Broadband Series Test System Figure 1 Hardware setup for the measurement ATM FORUM UNI SIGNALLING The procedure involved in setting up a switched virtual circuit (SVC) according to UNI 3. specification [ATM Forum 1995] is shown in Figure 2. Call Request The calling party (client) initiates call establishment by transferring a SETUP message on the signalling virtual channel 2 across the interface and starting timer T33. This SETUP message contains several information elements (IE) to identify the message, specify various ATM Adaptation Layer (AAL) parameters, calling and called party addresses, Quality of Service (QOS) requirements, and a number of other elements if needed. Following the transmission of the SETUP message, it is considered to be in the Call Initiated state. If no response to the SETUP message is received by the user before the first expiry of timer T33, the SETUP message may be retransmitted and timer T33 is restarted. If the user does not receive any response to the SETUP message after the second expiry of timer T33, the user shall internally clear the call. 1 Institute of Systems Science, National University of Singapore, Heng Mui Keng Terrace, Kent Ridge, Singapore According to ITU-T draft Recommendation Q.2931, the Signalling virtual channel uses VPI=, VCI=5.

2 Call Proceeding If the network and called party (server) can determine that access to the requested service is authorised and available, the network and called party may send a CALL PROCEEDING message to the calling user to acknowledge the SETUP message and indicate that the call is being processed, and enter the Incoming Call Proceeding state. In this state, no more call establishment information is needed, nor will any call request be accepted. When the calling party receives the CALL PROCEEDING message, the user shall stop the timer T33, start timer T31, and enter the Outgoing Call Proceeding state. ATM Client ATM Switch ATM Server P cell PD SETUP CALL PROCEEDING SW SETUP PD - Propagation Delay P cell - Cell Transmission Time SW - ATM Switch switching time PR ser - Delay between receiving SETUP message and sending CONNECT message at server PR cl - Delay between receiving CONNECT message and sending CONNECT ACK message at client PR sw - Delay between receiving and sending CONNECT message at switch Call Setup Delay T CONNECT PR sw CALL PROCEEDING CONNECT CONNECT ACK PR ser Call Processing Time R PR cl CONNECT ACK Time Figure 2 UNI Signalling for connection setup Call Confirmation Upon receiving an indication that the call has been accepted, the called party shall send a CONNECT message across the interface to the calling party and enter the Active state. This CONNECT message contains some parameters such as call reference, message type, accepted AAL parameters and other identifiers created as a result of the IEs in the SETUP message. On receipt of the CONNECT message, the calling party shall stop timer T31, send a CONNECT ACKNOWLEDGE and enter the Active state. If the calling party has received CALL PROCEEDING message, but does not receive a CONNECT or RELEASE message prior to the expiration of timer T31, then the calling party shall initiate clearing procedures towards the network. Call Release

3 If the calling party wants to release the connection, the party shall send a RELEASE message to the called party, start timer T38, disconnect the virtual circuit and enter Release Indication state. This message only contains the basic information to identify the message across the network. Once the virtual channel used for the call has been disconnected, the called party shall send a RELEASE COMPLETE message to the network and enter the Idle state. On receipt of the RELEASE COMPLETE message, the calling party shall stop timer T38 and enter the Idle state. The called party can initiate the call release too, and all the responses will occur in the reverse direction. SETUP message to the server and CALL PROCEEDING message to the client. Similarly, there is a small delay when the switch sends CONNECT to the client and CONNECT ACKNOWLEDGE to the server. In order to make the measurement more accurate, these two delays must be deducted from the Equation (1). Throughput The throughput is defined as the ratio of the number of successful connection requests over the number of connection requests made per unit time. In another words, the throughput is the ratio of the number of successful call setup over total number of call setup requests generated..1.8 Probability Fore ASX-1 Fore ASX-2 RESULTS AND DISCUSSION Call Setup Delay Measurement In this measurement, the effect of the following factors on call performance are examined: The characteristic of ATM switches The effect of call holding time The effect of number of hops on ATM connection setup delay Call Setup Delay Call Setup Delay Figure 3 Call setup delay distribution The call setup delay is defined as the delay between the connection request (sending SETUP message by the client) and connection confirmation (sending CONNECTION ACKNOWLEDGE by the client). It is denoted by T in Figure 2. In order to measure the delay, the following assumptions are made: The propagation delay PD is negligible as the distance from the user to switch is less than 1 m. The Segmentation and Reassembly (SAR) delay at an ATM user is negligible as most of the UNI messages can be segmented into one ATM cell. The cell transmission time P cell is small and ignored because of the small cell size (53 bytes). The time stamp is made at ATM layer. Let the delay between receiving CONNECT and sending CONNECT ACKNOWLEDGE at client be PR cl, the delay between receiving SETUP and sending CONNECT at server be PR ser and the delay between receiving CONNECT and sending CONNECT ACKNOWLEDGE at switch be PR sw. When the switch receives SETUP message, it takes SW to route the SETUP message to the destination. Thus the call setup delay T can be expressed as: T = SW + PR ser + PR sw + PR cl (1) where the sum of PR ser and PR sw is termed call processing time (R) at server. The Fore ATM switch was connected to two Optical Line Interfaces of HP BSTS, where each of them was emulated as either ATM client or server (Figure 1). The call setup requests were generated at a constant rate of 1 call per second and later the calls were released at the same rate. The results obtained from a sample size of 1 for the ATM switches Fore ASX-1 and ASX-2 are listed in Table 1 and Table 2. Delay Mean Pr ser PR sw PR cl R T Table 1 Call setup delays at Fore ASX-1 Delay Mean Pr ser PR sw PR cl R T Table 2 Call setup delays at Fore ASX-2 The call setup delay at ASX-1 is shorter than that of ASX-2. Comparison of these two sets of results shows that PR sw makes the difference. This delay is the only switch-related delay in the measurement, other delays are user-related. The results show that ASX-2 switch needs 3.36 ms more to assign VPI and VCI values for the calling (client) and called party (server) compared to ASX- 1 switch. To calculate the value of SW, Equation (1) is used. The following calculation is used for ASX-1: Equation (1) is valid only if the ATM switch can handle UNI Signalling messages simultaneously. However, due to the scheduling of CPU in ATM switch, there is a small delay when the switch sends

4 Call Setup Delay Mean = ms Std. Dev. = Figure 4: The effect of call holding on call setup delay SW = T - (PR ser + PR sw + PR cl) = ( ) ms = 6. ms Similarly, the value of SW at ASX-2 is calculated as 1.43 ms. However, due to the scheduling delay mentioned earlier, the actual value of SW should be smaller. After synchronising the ATM client and server, the scheduling delay was found to be 1.44 ms at ASX- 2. Thus, SW = = 8.99 ms Min. = ms Max. = ms Number of Unreleased Calls As the ASX-1 switch does not respond with CALL PROCEEDING message back to client when it receives SETUP message, this scheduling delay does not involve in SW measurement at ASX-1 switch. From the calculation, it shows that the value of SW is 2.99 ms shorter at ASX-1 switch than that at the ASX-2 switch. Again, the value of SW is switch-related. In order to find the relationship between call setup delay and various delay parameters in Equation (1), the correlation coefficients 3 between call setup delay and the various delay components are calculated and listed in Table 3. at ASX-2 has a stronger relationship with PR cl. The call setup delay distribution for a single switch is plotted in Figure 3. To measure the effect of call holding on call setup delay, the ATM connections were setup one after another and not released. The call setup delay for every call was measured. Figure 4 shows the mean call setup delay (each data point is the mean of 5 measurements) under different number of unreleased calls in the system. There was no difference in call setup delay with increase in the the number of unreleased calls. This shows that the call holding has no effect on the call setup delay. Next, in order to measure the effect of number of hops on the call setup delay, two switches are used. The emulated server was connected to ASX-1 and the emulated client was connected to ASX-2. The two switches were connected to each other. The measurement results are shown in Tables 4 and 5. Delay Mean Pr ser PR sw PR cl R T Table 4 Call setup delays with two switches Probability Call Setup Delay Figure 5 Call setup delay distribution with two switches Delay Correlation Coefficient ASX-1 ASX-2 PR ser PR sw PR cl Table 3 Correlation coefficients between call setup delay and various delay components (single switch) The results show that call setup delay has a stronger linear relationship with PR ser at ASX-1 switch, while the call setup delay Mean Call Setup Delay Fore ASX-1 Fore ASX-2 3 If Correlation is r xy, µ x and µ y are mean of x and y, then Correlation r xy µ xµ y Coefficient ρ xy =, where σ x and σ y are standard deviation of x and y. σ σ x y Number of Simultaneous calls Figure 6 Call setup delay under different number of simultaneous calls

5 Mean Call Processing Time Fore ASX-1 Fore ASX-2 Number of Successful Calls Number of Simultaneous calls Number of Call Requests Figure 7 Call processing time under different number of simultaneous calls Figure 8 Throughput under simultaneous call generation Delay Correlation Coefficient PR ser PR sw PR cl Table 5 Correlation coefficients between call setup delay and various delay components (two switches) The results show that the call setup delay with two switches is higher than the sum of the setup delays measured previously at two individual switches. By comparing the results with single and two switches, it is noticed that the delay parameters in Equation (1) do not vary much. From the correlation coefficients, it shows that the call setup delay has little relationship with the delay components. The main contribution to the call setup delay with two switches is the delay involved between two ATM switches. The delay related to ATM network (ATM switches) can be calculated as: delays are not discussed in this paper. However, the results show that NNI introduces a significant delay in call setup involving multiple switches. The call setup delay distribution with two switches is plotted in Figure 5. Throughput Measurement In this measurement, simultaneous 4 call requests were generated by the analyser to measure the throughput for each switch. Number of call requests and number of successful calls were recorded, as well as the call setup delay of the successful calls. Because of the software and hardware limitation, a maximum of 1 simultaneous calls were generated. The focus was on the increase in call setup delay when the number of simultaneous calls increased. The measurement results are listed in Figure 6, Figure 7 and Figure 8. Each data point is the mean value from 5 experiments. T - PR ser - PR cl = ms = ms This delay is much larger than that of a single switch; 1.87 ms for ASX-1 and 16.1 ms for ASX-2. Since the communication and control between two ATM switches are handled by Network-Network Interface (NNI), the details of NNI Notice that the two ATM switches performed differently in this measurement. The call setup delay at ASX-1 increased exponentially when the number of simultaneous calls increased, but that of ASX-2 increased logarithmically when the number of calls was beyond 3. The increase in delay was caused by the queuing delay at the switch as the switch can handle only 1 calls per second[fore Systems]. The server call processing time also increased in the same manner as the call setup delay because the incoming call request rate was faster than the server response rate. Though the delays are different, both the Fore ATM switches yielded 1% throughput. To find out which delay parameter causes the increase of the call setup delay, the correlation coefficients between call setup delay and various delay components calculated as shown in Tables 6 and 7. 4 Because of hardware limitation, the ÒsimultaneousÓ calls are in fact generated one after another with interval of 5 ms.

6 Delay Correlation Coefficient 1 call 1 calls 1 calls PR ser PR sw PR cl Table 6 Correlation coefficients between call setup delay and various delay components at Fore ASX-1 Delay Correlation Coefficient 1 call 1 calls 1 calls PR ser PR sw PR cl Table 7 Correlation coefficients between call setup delay and various delay parameters at Fore ASX-2 When the number of simultaneous calls is small, it is the ATM user causing the increase in delay for both switches because the switch can still handle the connection requests. However, when the number of simultaneous calls is large, it is the ASX-1 ATM switch causing the increase in delay, and the ASX-2 switch directs the delay to the ATM user. This difference in delay distribution is caused by the difference in switch architecture [Fore Systems]. Call Release The delay between a release request (sending RELEASE at client) and release confirmation (receiving RELEASE COMPLETE at client) was also measured at both FORE ASX-1 and ASX-2 ATM switches. The results are shown in Table 8. The call release at FORE ASX-2 takes longer time than that of ASX-1, whereas the call release delay with two switches was close to the sum of the individual delays. Mean ASX ASX Double Switches Table 8 Call release time CONCLUSION Call performance measurements on two different ATM switches were carried out using a broadband network analyser. The probability distribution of the delays are obtained and the correlation between the setup delay and its different constituents are established. It is observed that the delays at the calling and the called user entities of the UNI form a major component of the call setup delay. The average call setup time with the switches are found to be 21 and 27 msecs respectively. The call setup delay does not seem to increase linearly with the number of ATM switches involved whereas the call release delay increases linearly with the number of ATM switches involved. With simultaneous call arrivals, though with higher delays, both the switches are found to establish calls at the rate of 1 calls/sec. Future work involves conducting similar studies on PNNI implementations. References [ATM Forum 1995] The ATM Forum ÒATM User Network Interface (UNI) Specification Version 3.1Ó. Prentice Hall: pp , [Black 1995] Uyless Black ÒATM: Foundation For Broadband NetworksÓ. Prentice Hall: pp [Fore Systems ]Fore Systems ÒForeRunner ASX-1 ATM Switch Architecture Manual Release 2.1Ó. Fore Systems Inc., Pittsburgh. [Hewlett Packard 1995] Hewlett Packard IDACOM Telecommunications Operation ÒHP Broadband Series Test System: UNI Signalling UserÕs GuideÓ. Hewlett Packard Company, Alberta Canada. Biography Su Kwe Long received his Bachelor Degree (Honours) in Electrical Engineering from the National University of Singapore in Currently he is working as an IS specialist and is in charge of the ATM network infrastructure at Singapore General Hospital (SGH). His research interests are on ATM networks. R. Radhakrishna Pillai is a research staff at the Institute of Systems Science, National University of Singapore. His research interests are in signalling and service management in ATM networks, MAC protocols, and modelling and performance analysis of communication networks. His current work include wireless ATM, open signalling, and ATM network and service management. Dr. Pillai received the M.E and Ph.D degrees from the Indian Institute of Science, Bangalore, in 1989 and 1993 respectively. Prior to joining ISS, he was with Tata Elxsi (India) Ltd., Bangalore, where he was involved in the design and development of networking and multimedia products. Jit Biswas is a Member, Research Staff at the Institute of Systems Science (ISS), Singapore. His areas of work include Parallel and Distributed Computing, Network Management and Open Signalling. Dr Biswas led a project entitled "ATM Enterprise Network Management System", which was a joint collaboration with ITRI, a Taiwanese Research Institute. He was also responsible for a Network Management project which provided know how and expertise for the Management of Services and Networks for the Singapore National High Speed Network Testbed. Dr Biswas obtained his Bachelors degree in Electrical and Electronics Engineering from Birla Institute of Technology and Science, his Masters degree in Computer Science from the Southern Methodist University, and his Ph.D. degree in Computer Science from the University of Texas at Austin in Dr Biswas also has a Diploma in Industrial Engineering from NITIE, India and has worked with a major automobile manufacturer in India.