As Ethernet services grow in popularity with enterprises, the rapid deployment of voice over Internet protocol (VoIP) is a top priority for many service providers. And when customers switch from private voice lines to VoIP, their expectations are high; they expect toll-quality voice and top-quality data services running over their Ethernet access lines. Service-level agreements (SLAs) between a service provider and customer outline quality terms in a legal contract. SLAs help service providers attract and retain customers, but there are penalties associated with substandard service: poor customer satisfaction, increased spending on maintenance, and often direct financial penalties.
When best effort Ethernet services are used (e.g., for Internet and email access), SLAs often focus on link availability by specifying maximum acceptable downtime and meantime to repair. SLAs get a lot more detailed when delay-sensitive, mission-critical services are deployed over an Ethernet link: VoIP, videoconferencing, and remote data storage require more than best effort access. In addition to availability, these more detailed SLAs also outline expectations related to quality of service (QoS) by specifying acceptable Ethernet performance criteria.
Service providers can confidently commit to these agreements if the link is thoroughly tested when commissioned. A simple test plan that only validates the basic functionality of an Ethernet link (using ping, connectivity verification, etc.) leaves an operator exposed to poor network performance once live customer traffic begins. Knowing this problem but lacking test capabilities, some providers over-commission bandwidth to ensure SLAs will be met. This expensive practice can be avoided by diligently confirming a link’s performance before activating a service.
To completely test an Ethernet access line requires an understanding of the type of applications running over the link. For basic Ethernet services such as Internet access, virtual private LANs, and remote backup applications, the Internet Engineering Task Force (IETF) standard for benchmarking network interconnect devices, RFC-2544, is an excellent testing guideline. RFC-2544 outlines specific tests to validate throughput, latency, frame loss, and back-to-back (burst) performance, all of which are commonly cited in sophisticated SLAs.
Delay-sensitive applications such as VoIP and IP video are sensitive to additional performance parameters such as inter-packet delay (packet jitter) and packet sequencing (out-of-order packets). Because VoIP coexists with other types of data traffic, bandwidth-demanding applications such as remote backup or file transfer protocol can frequently degrade the quality of VoIP services. This type of problem can be avoided by simulating real traffic patterns during commissioning and adjusting network parameters to provide priority for time-sensitive VoIP packets.Before turning up a VoIP service, the Ethernet access line should be tested for performance with a focus on the five key parameters listed in the Table. These same tests are critical for troubleshooting VoIP services during or after deployment. Various combinations of parameters and their duration/frequency point toward different network problems. For example, intermittent frame loss points toward core switching and routing problems, while continuous frame loss infers backhaul network congestion. With comprehensive test results, an experienced deployment team will be able to quickly optimize a customer link. The most common sources of poor VoIP performance are:
- Route flapping. A network frequently changes the path that packets use to get to their final destination. That can be caused by congested networks or poor traffic engineering. Route flapping causes several call quality issues and is often related to packet jitter, latency, and out-of-order packets.
- Access link congestion. If VoIP packets are not given dedicated bandwidth or sufficient priority over the access link, other bursty or sustained bandwidth-consuming applications can rapidly degrade voice quality or the ability to sustain a call. Common examples of these interfering applications include large file downloads, offsite remote backup, or videoconferencing services. Poor throughput and packet loss are both indicators of access link congestion. These problems can often be alleviated by reconfiguring policies on the access switch.
- Backhaul congestion. Bandwidth limitations in the aggregation of access lines are often indicated by high latency and low throughput.
By analyzing the types of errors measured and comparing them to their frequency and duration (intermittent, periodic, continuous), a service provider can properly assess the adjustments required to commission a link satisfying SLA-defined quality levels.
To realistically simulate and analyze the interaction of multiple services being transported over a single link, the test instruments used must be able to recreate the client’s network environment, both on a LAN and WAN level. To simulate customer premises equipment, parameters such as flow control and throughput need to be emulated. To ensure that generated test traffic is carried with the same priority and routing scheme as the customer’s data, virtual LAN tagging of test traffic is required.
To qualify an Ethernet link carrying various services, multi-stream test traffic analysis is required. This function allows a deployment team to simulate typical link traffic by generating unique test streams for each of the several services typically sharing an access line. For instance, a high-priority stream with variable bandwidth and frequent bursts of data replicates VoIP traffic, while a low-priority continuous throughput stream simulates offsite backup applications. Since simulated client traffic is fundamental to an accurate test plan, the Ethernet tester used must be able to support a reasonable number of independently configurable streams. Normally, a maximum of 10 streams is sufficient for commissioning applications.
When stressing a network with multi-streams, detailed measurements are performed on one stream, while controlling the priority, bandwidth, and characteristics of the others; this technique permits network traffic replication over a wide variety of typical boundary conditions.
Specialized VoIP analysis tools are often inadequate for Ethernet link commissioning. These instruments typically focus on perceptual voice quality testing (e.g., mean opinion scores, R-factor, and other PESQ algorithms). These measurements provide a single aggregate score that typifies the quality of voice conversations experienced by end users. These consolidated figures of merit are useful in confirming that links have been properly commissioned once stress-testing and network analysis is completed but do not offer the granularity required to help deployment teams troubleshoot or fine-tune an access link’s network parameters. Although VoIP test sets are often capable of measuring basic performance parameters, they often lack key features required for link commissioning: full line-rate traffic generation, network emulation, and detailed Ethernet (RFC-2544) conformance testing.
A final key consideration in VoIP pre-deployment testing is the effect of traffic direction. Since traffic or bandwidth over an access link tends to be asymmetrical (downloads exceed uploads, or vice versa), quality problems often occur predominantly in one direction (i.e., transmission or reception is affected). One-way effects are often experienced in telephone conversations using VoIP-one caller hears echo, dropouts, and delay, while the other hears a perfectly clear conversation.
To ensure that unidirectional effects are accounted for, it is important to use a test setup that permits parameter measurement in either direction independently. Although all commissioning schemes involve two test sets, some only perform measurements in loopback configuration (see Figure). This method results in send and receive test scores combined together into a single averaged measurement; upload or download issues cannot be independently identified.
It should be noted that one parameter, latency, cannot be measured unidirectionally without sophisticated clock synchronization schemes (involving GPS signals, for example). Latency is typically tested in loopback. All other performance metrics, including throughput, burst, frame loss, and packet jitter should be measured unidirectionally to obtain accurate test results.
SLAs are a key differentiator for service providers offering Ethernet services. SLAs help retain customers and attract new ones. In the case of VoIP, many potential customers will only migrate from PSTN services when SLAs guarantee toll-quality calls. Commissioning Ethernet to carry VoIP requires a carefully designed test plan. The payoffs are significant: Customers are satisfied and SLAs are met.
Scott Sumner is partnership program manager at EXFO (St. Laurent, Quebec).