Generating carrier class Ethernet services for business can be tricky, with a number of issues needing to be addressed if Quality of Service is to be assured. Robert Winters provides some guidance
Metro Ethernet service deployments are continuing apace on a global basis with a variety of service offerings and enabling technologies that offer 'real broadband' as an attractive alternative to lower bandwidth DSL and Cable products, high cost leased lines, ATM and Frame Relay. Depending on the region of deployment there are a number of Ethernet technology alternatives and build-out strategies in progress.
For example, in Europe many incumbent service providers are maximising their use of existing SDH transport assets by upgrading equipment to support Ethernet services. With the insertion of new Ethernet line cards a variety of non-switched pure Ethernet transport implementations such as GFP (Generic Framing Procedure) and more QoS oriented switched services such as VLANs (Virtual LAN) etc, coupled with the value add of MPLS, can now be offered and new revenue models instituted.
Alongside the transport network there are also deployments using hybrid switching and routing technologies with next generation protocols such as MPLS (Muliprotocol Label Switching) and RPR (Resilient Packet Ring).
Along with the enhancements to existing SDH transport equipment and switched Ethernet networks there also exists a growing number of European state sponsored broadband initiatives in countries such as Sweden and Ireland. These programmes encourage the rollout of dark-fibre thus enabling competitive broadband service providers to build their networks over a ready-made physical layer transport medium. This type of initiative offers a reasonably clean slate approach to building an Ethernet product offering. The competitive service provider can at least focus on a deployment technology of choice, such as Ethernet over MPLS or RPR.
However, nothing is ever that easy. As can be imagined when business class services (as opposed to best effort home consumer type) are being guaranteed on an end-to-end basis between two major metropolitan areas, or indeed within the confines of a particular metro ring, there are challenges where 'Carrier Grade Ethernet QoS' is required. In this situation, service providers are expected to offer high bandwidth Ethernet services but also reliability, redundancy and high quality business class applications. Applications are increasingly delay and jitter sensitive, such as multicast video, time sensitive e-commerce web solutions and voice over IP (VoIP). This article focuses on the requirements of carrier grade Ethernet QoS at a layer 2 service and IP application level and assumes other carrier grade issues related to hardware redundancy (for example, MPLS fast reroute guarantees, inherent SDH protection and RPR protection) are addressed.
To capture the enormous enterprise business market with differentiated Carrier Grade Ethernet QoS products requires an understanding of the capabilities of Ethernet Services and the applications being transported over Ethernet. What to look out for when offering Carrier Grade Ethernet QoS:
1. Understand the performance of QoS and CoS (Class of Service)
The IEEE 802.1p,q standards for Virtual LAN (VLAN) services offer a method for identifying a service stream, setting bandwidth and assigning a priority setting that determines class of service (CoS), rather than it being a pure QoS parameter in network implementations such as ATM which offers attributes such as CBR (constant bit rate) settings etc. So, in order to benefit from inherent Ethernet cost effectiveness and high bandwidth, plus offer QoS, there is also a need to bring additional quality metrics into the mix such as those offered through connection admission control (CAC) for end-to-end bandwidth and using MPLS signalling and traffic engineering capabilities. Ethernet industry-focused organisations, such as the Metro Ethernet Forum (MEF), have defined service types including Ethernet Line and LAN Services for point to point and point to multipoint/multipoint to multipoint services.
CoS identifiers within these services can include specific source and destination MAC addresses, customer edge VLAN ID/IEEE 802.1p. Inspection of these packet headers requires processing power from network devices that may impact performance and requires verification of performance on a per service basis. The MEF has defined traffic profiles per CoS identifier that include Commited Information Rate (CIR), Peak Information Rate (PIR) and associated burst sizes. The provider can thus offer a greater number of service options to their customers. For example, a subscriber may connect to a metro Ethernet service at one location with 10Mbps user to network interface and another location at 100Mbps. The CIR in this case could be 10Mbps. More is to come. With the development of VPLS and loss-less packet transmission in metro Ethernet networks, the number of network options will continue to increase.
2. Check ability to guarantee service stream and IP application flow quality.
In the past it has been difficult to test on a per service and per application flow basis since traditional test methods relied on packet blasting at layer 2 only.
Basically, if the layer 2-service pipe was rated by RFC2544 throughput tests, this was generally viewed as a sufficient guarantee of quality. However, to really guarantee carrier grade Ethernet QoS, a far more granular approach is required. Service providers need to be confident that each service, each user and each IP application flow using that service are thoroughly tested for quality.
Therefore, a pragmatic approach to testing is required whereby corporate Ethernet service and application flow models can be quickly built, then emulated and analysed for quality issues throughout the network under test with varying load and network status conditions. Using this test method, QoS boundaries can be realistically determined for both network services and application layers.
3. Guarantee end-to-end QoS
Ethernet services invariably start out their 'circuit life' as a layer 2 service (e.g. VLAN) originating at the customer premise into some point of aggregation and transport such as MPLS/RPR. The transport method can be a layer 3 VPN such as MPLS RFC2547 and then converted out the 'other side' back to the layer 2 VLAN and into the remote customer premise. It is important to test on an end to end basis. For example, with the possibility of an MPLS misconfiguration the number of hops and propogation time can change and requires end-to-end test for different traffic engineered service configurations. Also, it is important that each CoS priority assignment for 802.1 VLANs effectively maps onto MPLS EXP bits (equivalent quality metric) and back again. In situations involving MPLS fast reroute, how long does it really take for an individual end to end Ethernet service to get back to normal if a disruption occurs? Another consideration is restoration of service when normal conditions resume.
4. Understand the effects of TCP/IP application flows on 'guaranteed' Ethernet services bandwidth
Yes, we all know that Ethernet is a layer 2 service and you should not care about the IP and application layers above. However, when it comes to offering bandwidth guarantees you need to pay attention. It is extremely important to consider the effect of multiple TCP/IP application traffic flows running over a given layer 2 service and the potential side effects such as a drop in effective bandwidth. Due to TCP congestion notification schemes, layer 4-7 performance can rapidly degrade leaving customer bewildered and confused about the service specification and network performance. Rather than facing an irate customer who believes in getting the bandwidth pipe they paid for, it is worth testing a variety of scenarios with voice, video and data traffic in advance that can cause an excessive amount of dropped packets. In this way a service provider can better understand how and why this occurs, but also explain to customers why, for example, a 20Mbps service at layer 2 does not necessarily translate into the equivalent 'application bandwidth'. Of course, with full RFC2544 tests, throughput can be guaranteed at layer 2, but add real application TCP flows into the mix and see what happens.
5. Have a method of isolating Ethernet quality service issues from customer application problems
The ability to offer end-to-end carrier grade Ethernet QoS usually assumes the customer has the perfect set of well-behaved applications. Again, as an Ethernet services provider, the last thing needed is blame for a service issue caused by application problems. Aside from bandwidth hogging applications such as peer-to-peer (P2P) transactions that sponge bandwidth at an enormous rate, even standard IP applications such as Web, E-mail, VoIP, Multicast and Streaming applications can contribute to latency and loss of bandwidth. The ability to quickly isolate a problem source - and to prove it - is a key element in customer satisfaction.
Pressure on service providers and the equipment vendors supplying them to provide carrier grade Ethernet quality of service (QoS) guarantees are being heightened with the introduction of an array of value added applications such as Video on Demand, response time critical Web applications and VoIP. In order to improve competitive advantage there are a number of areas in which QoS issues can be determined and mitigated with practical quality boundaries worked out in which premium level business class services can be more effectively and confidently guaranteed.
Robert Winters is Chief Marketing Officer and Co-founder Shenick Network Systems Limited, and can be contacted via tel: +353-1-2367002; e-mail: firstname.lastname@example.org [l=www.shenick.com/]http://www.shenick.com/[/l]