March 1998

GUARANTEEING QoS OVER AN ATM NETWORK

BY JOHN M. GILES

One advantage of Asynchronous Transfer Mode (ATM) networks over other packet switching networks such as Frame Relay or X.25 is its high level of predictability. ATM cells traveling over a net-work are of a small, fixed size, as opposed to variable length frames. This allows ATM network switches to interweave a sub-scriber’s cells with other network users’ traffic, and still maintain tight tolerances on end-to-end network delay variances.

The innate ability to prioritize traffic and predict the arrival time of cell pay-loads allows network users to subscribe to services that are suitable for voice and video traffic, in addition to data traffic. In the past, companies have leased trunk lines from the telephone company to handle voice and video applications where 100 percent of the leased line capacity is available to the user. This solution is expensive and does not lend itself to meshed "any-to-any" network connectivity.

ATM networks offer an inexpensive solution based on actual utilization and can provide meshed network connections over a single network drop using virtual connections. Users will migrate voice and video conference applications to ATM networks only if the service provider can guarantee a quality of service (QoS) that is similar to that of the leased line, and at a lower cost.

This article explores the types of service categories offered by public net-work service providers and the associated traffic contract parameters. For each combination of service category and traffic contract parameters, particular measurements can be made to ensure that the ATM circuit under test meets the desired QoS levels specified in the service contract. Specific ATM QoS testing methodologies are also discussed.

ATM FORUM SERVICE CLASSES
Service classes describe the high-level requirements of each virtual circuit operating on a network. Classes of service specify how various ATM net-work elements prioritize traffic for unique application requirements.

Constant Bit Rate (CBR) Service
This service provides a sustained, continuous flow of bits from the sending station to the receiving station. A fixed amount of bandwidth is reserved throughout the entire virtual circuit and is guaranteed by the service provider to always be available to the user. This service is ideal for interactive digitized voice or video applications requiring a constant stream of digital information. Some application examples include teleconferencing, telephony, distance learning, and pay-per-view video.

Variable Bit Rate (nrt-VBR and rt-VBR) Service
Non-real-time Variable Bit Rate (nrt-VBR) service is ideal for bursty traffic that does not require tight tolerances on arrival time. Typical business-class local-area network (LAN) traffic fits this pro-file. Some application examples are tick-et reservation systems and banking transactions. Real-time Variable Bit Rate (rt-VBR) service allows for bursts of traffic, but requires tight tolerances on arrival time. Some application examples are SNA (Systems Network Architecture) traffic, packetized voice, and some types of multimedia retrieval systems.

Unspecified Bit Rate (UBR) Service
This service provides no guarantee on bandwidth availability. Only unused bandwidth is available to the user and data is transferred on a best-effort basis. Only data applications that are not time sensitive should use this service. Some examples include e-mail, remote terminals, and file transfer.

Available Bit Rate (ABR) Service
This service uses a flow control mechanism to regulate bit rates at the source of traffic. Applications can only transmit traffic at rates limited by feedback from the network. The difference between ABR and UBR is that when congestion exists in a virtual circuit, the ABR service will cause the transmission rates to throttle down, whereas the traffic on a UBR service may get discarded by any switch in the network path experiencing congestion. Some application examples are LAN emulation and LAN interconnection/ internetworking services. TRAFFIC CONTRACT PARAMETERS Public ATM network service providers offer contract parameters to subscribers that specify limits. These limits impact the price of the service and include the following:

Peak Cell Rate (PCR)
This is the maximum bandwidth for the virtual circuit that can be guaranteed by the service provider. Cells sent in excess of this rate may be discarded or marked as discard eligible by the ingress ATM switch. ATM cells traveling through the network marked discard eligible may be discarded by any switch in the network experiencing congestion. For CBR service, this rep-resents the guaranteed constant band-width for the virtual circuit.

Sustainable Cell Rate (SCR)
Applicable only to VBR services, SCR specifies the minimum bandwidth available at any time to the customer’s VBR traffic on a virtual circuit.

Maximum Burst Size (MBS)
Applicable only to VBR services, MBS specifies the maximum number of cells that can be transmitted at the PCR rate and still comply with the traffic contract.

Cell Delay Variation Tolerance (CDVT)

This applies to time sensitive services and specifies the maximum allowable tolerance of Cell Delay Variation (CDV) between two end stations. When performing QoS measurements, it is important that the same service class and traffic contract levels be emulated by the tester to ensure that performance levels are met. ATM QOS MEASUREMENTS ATM QoS measurements exist primarily for time-sensitive applications. If a customer has no guarantee by a service provider that an ATM virtual circuit will deliver voice or video at a comparable quality to a leased line, that customer is not likely to cut over to a switched ATM service — even if the service is less expensive. To address this issue, the ATM Forum has established the following QoS measurements:

Cell Delay Variation (CDV)
Because the time it takes for a cell to be switched within the ATM switch fabric varies slightly from cell to cell due to queuing delays, a small difference in arrival time between cells will occur. When this is amplified by the cell travelling through many switches to reach its destination, the resulting jitter effect can degrade the reception quality of voice and video payloads. If cells arrive too soon (clumping), the PCR could be exceeded and cells may be discarded. If the cells arrive late (gaps), it may noticeably affect audio or video quality. Two types of CDV measurements have been standardized: the CDV 1-point test and the CDV 2-point test.

The 1-point CDV measurement relates to the early arrival time of cells relative to a calculated arrival time. Received cells are examined to see if they arrive early or late compared to an expected arrival time (1/PCR) which is an indication of cell clumping. This measurement is only applicable to CBR traffic, but can be performed between any two network elements and is not restricted to only end-to-end measurements. Average early arrival times that are above the PCR indicate a problem.

The 2-point CDV measurement relates to the actual measured delay through the network end-to-end. Unlike the 1-point CDV measurement, this test is applicable to any class of traffic and takes into account the effect of lost, erred, or misinserted cells. The difference between the earliest measured cell time delay (CTD) and the latest measured CTD is the peak-to-peak value. A guide-line for maximum CDV recommended by the International Telecommunications Union (ITU) is 250 milliseconds peak-to- peak for DS1 and DS3 voice circuit emulation. For MPEG video or HDTV, a maximum CDV of 1 millisecond peak-to- peak is recommended.

Cell Transfer Delay (CTD)
The time between when the first bit of a cell has left the sending station and when the last bit of the cell has arrived at the receiving station is CTD. This is an important value for constant bit rate applications because if a cell takes too long to traverse a network, it is considered lost or late by the receiving station and the entire packet being reassembled gets thrown away. This affects the voice or video quality. A guideline recommended by the ITU is 130 millisecond maximum CTD for voice or teleconference video applications, and 1 second for streaming video applications.

Cell Loss Ratio (CLR)
CLR is the number of cells lost during travel to the receiving station, divided by the total number of transmitted cells. Cells may never arrive at a destination due to a number of reasons: a switch sends a cell to the wrong destination; a switch is severely congested; the sending station bursts above its contract limits and cells are discarded; or the cell takes too long to traverse the network, exceeding the maximum CTD, and arrives too late to be processed. This is usually an all or nothing proposition. Either no cells arrive at a destination or nearly every cell arrives within the time limit. An ATM virtual circuit should have a maximum CLR of 10E-7, or 1 lost cell in every 10 million transmitted cells.

Cell Misinsertion Rate (CMR)
CMR is a count of the number of cells misinserted, divided by the number of seconds the test ran. A misinserted cell is delivered to an unintended destination. The usual cause of this is when a cell header is corrupted, the destination address may appear valid to a switch and the cell is sent to the destination address in the corrupted header. The unintended recipient of this cell does not recognize the payload of the misinserted cell as being valid data. This can cause an entire data packet that was being assembled in the ATM layer to be discarded. This error is rare. A rule of thumb is the CMR should be ten times less than the CLR.

The last three QoS measures specified by the ATM Forum and the ITU are Cell Error Rate, Severely Errored Seconds, and Severely Errored Cell Block Ratio. These standards have recently been developed and are not yet widely implemented by switch vendors.

RELATIONSHIPS
After reviewing the definitions of each parameter that can be negotiated within an ATM service provider’s contract, it can be seen that a virtual circuit can be provisioned to fit the needs of nearly any conceivable application requiring net-work bandwidth.

ATM QoS Testing Methodology
When an ATM service is installed, the service provider delivers the service over a standard transmission medium such as DS1, DS3, or OC3. These mediums sup-port low to high bit rates to accommodate a customer’s throughput requirements and are priced accordingly. The tester must accommodate the interface type selected and perform conformance tests on the physical layer and ATM layer first to ensure that ATM cells can be transported reliably between the user and the edge switch.

To determine if the ATM circuit delivers the guaranteed QoS, the test device must be able to transmit ATM traffic of the same ATM service class and at the maximum bit rate specified in the con-tract. Once configured to simulate the traffic profile that the end user intends to transmit over the virtual circuit, measurements can be made on the received stream of ATM cells to determine if the QoS measures are acceptable.

It is important to remember that many virtual circuits may be provisioned on one physical interface, so multiple streams of ATM cells must be generated to replicate the simultaneous loads that will be placed upon the switch port. QoS performance is then measured on any one of the virtual circuits under load.

If QoS performance is not measured on a virtual circuit under load, it could appear to perform within the contract when in use by itself. However, when all of the virtual circuits are in use, the edge switch or other network components could become congested. The result is the circuit that tested within agreed QoS limits would experience severe performance problems when the user’s critical applications were executed across the link.

Figure 1 illustrates a common scenario where three Permanent Virtual Circuits (PVCs) are provisioned within one DS1 (Digital Service, level 1, with a 1.544 Mbps signal speed in North America) circuit. If all three of the PVCs were constant bit rate service types, then the total PCR sum could not exceed 1.024 Mbps.

If all three of the PVCs were variable bit rate applications, the total SCR sum could not exceed 1.024 Mbps and the total PCR could not exceed 1.35 Mbps. If the end user needs more bandwidth, a DS3, fractional DS3, Inverse Multiplexer ATM (IMA), or OC3 physical interface would be required.

End-to-end testing over a PVC can be accomplished in one of two ways. First, a tester at each end of the PVC can be used to run two tests with each tester having a turn at being the traffic source and being the traffic receiver. The advantage of this method is each direction of the PVC can be tested independently.

Second, a physical or virtual loopback can be placed at the far end of the PVC and a single tester can be used to both source the traffic and receive the traffic by looping back to itself .

CIRCUIT QUALIFICATION
An end user has ordered ATM service between the headquarters and remote branch office that contain two PVCs — one to carry data, and another to carry interactive real-time video teleconference information. The PVCs have been "nailed up" by the service provider so that all of the ATM switches in the net-work carrying traffic have the proper configuration. The local loops on each end of the PVCs are DS1 lines that have been installed, and the physical layer testing is complete. Table 2 is an example of the customer’s contract parameters with the service provider. Referring back to Figure 2, an ATM tester that sources multiple shaped traffic streams and makes ATM QoS measurements that meet the ATM Forum specifications, is connected. The tester is configured to source CBR traffic at 512 Kbps on VPI/VCI 0/97. Additionally, the tester is configured to source protocol data unit traffic that simulates network file transfers on VPI/VCI 0/98.

Finally a receive filter is configured in the unit to receive the stream of traffic for the QoS measurements on VPI/VCI 0/97. This is the PVC that carries the CBR traffic being looped back to the tester.

Once the traffic generation is initiated, the user begins measuring the QoS parameters of interest. According to Table 1, the following QoS measure-ments should be made for a CBR circuit: CDV, CTD, CLR, and CMR.

The first test measures both the cell loss and cell misinsertion ratios simultaneously. Cells are transmitted out on the PVC and the received cells are examined to see if they match up with the known transmitted cells total count and source address. Errors are recorded as a ratio.

Our next test is the 2-point CDV test. The time deltas between the transmitted and received cells are measured over time and the average becomes our mean CTD value. In addition, the minimum and maximum arrival times are also measured and the difference between them becomes our peak-to-peak CDV value. This completes the requirements for CDV and CTD measured values.

The QoS tests are complete. If the mea-sured values are within the service provider’s specifications, then the loop-back is removed and the customer can cut over the applications intended for these circuits with confidence. If the measured values are outside the service provider’s specifications, a technical specialist is usually assigned to solve the problem by altering the customer’s network route to a less congested path and retesting, or uses other troubleshooting techniques.

Early adopters of ATM service include physicians who share medical imaging data, engineers who share CAD/CAM data, and companies that are interested in reducing travel through video conferencing. It may seem from these examples that the average enterprise network may not yet require similar service guarantees. However, most long-distance voice traffic in the United States now travels over an ATM network backbone. In the near future there will be one cost-effective service that will satisfy all the voice, video, and data needs. Standards-based QoS testing of these mission critical communication links will aid greatly in the migration to ATM service for wide-area applications.

John M. Giles is a research and develop-ment network manager at Fluke Corporation, a world leader in compact, professional electronic test equipment. The Fluke Networks Division provides products and services for the installation, maintenance, and troubleshooting of data communication networks, with hand-held test tools for both LAN and WAN applications including ATM, xDSL, Fast Ethernet, Token Ring, Category 5 cable, and switches. For more information, call 800-44-FLUKE or visit the company’s Website at www.fluke.com/nettools.