Whitepaper Enabling Triple Play

Enabling Triple Play in Access Networks

A Guide to Implementing Quality of Service for Triple Play Services in the Access Networks with the Xpeedium2®pro Family of Switch/Router Devices. © March, 2006, SwitchCore Corp. DOC00332, V1

Enabling Triple Play in Access Networks

Contents

1

Introduction ..........................................................................................................................4

2

Background ..........................................................................................................................4

3

State of the Market ...............................................................................................................4

4

Migration from ATM to IP over Ethernet ............................................................................5

5

QoS ........................................................................................................................................6

6

7

5.1

Bandwidth Reservation................................................................................................................ 6

5.2

Label Switching ............................................................................................................................ 6

5.3

Classification ................................................................................................................................ 7

5.4

Congestion Control ...................................................................................................................... 7

5.5

Queuing ....................................................................................................................................... 10

Implementing QoS for Triple Play.....................................................................................10 6.1

General features ......................................................................................................................... 10

6.2

VLANs .......................................................................................................................................... 11

6.3

DiffServ ........................................................................................................................................ 11

Summary .............................................................................................................................12

SwitchCore Corp.

April, 2006

3

Enabling Triple Play in Access Networks

1 Introduction This white paper from SwitchCore explains the current development of Triple Play services in the Access Network. It explains some of the issues that arise and how they can be solved by using SwitchCore’s Xpeedium2®pro family of Access Aggregation Switches.

2 Background Triple Play services (voice, video, and data) offer service providers new revenue opportunities and ways of lowering deployment costs..These new real-time services (like VoIP and IPTV) require strong Quality of Service (QoS) support per subscriber. QoS is a general term which includes the ability to differentiate types of traffic and to ensure a certain level of performance across a network. QoS not only gives service providers a means of offering different services, but also a better way for managing the billing and control of the different Service Level Agreements (SLAs). IP and Gigabit Ethernet switching are leading the technology that allows these services to be deployed in the Access and Aggregation networks. This white paper explores current trends and identifies the most relevant QoS features in Ethernet switching for implementing Triple Play services.

3 State of the Market Service providers are constantly facing the challenge of high cost of equipment (CAPEX) and operating costs (OPEX). Traditionally, high service quality like five 9’s uptime has meant legacy platforms were expensive to deploy and provision. Relatively flat revenues over the last few years have service providers searching for ways to increase revenues. They are doing this by offering increased residential and business service which bundle voice, video, and data (i.e. they offer Triple Play services). Video is a generic term which includes broadcast TV, Pay-per-view, Video on Demand and, going forward, will include HDTV. These services are deployed through the Access Networks primarily using cable and DSL, as well as FTTx (Fiber). Triple Play over wireless is also driving the interface of wireless networks to IP over Ethernet. Triple Play services require greater bandwidth compared to traditional data services. The Multiple Service Operators (MSOs) who provide cable service are well-positioned, with the cable infrastructure and with the upgrades they have done, to provide enough bandwidth for all services. Incumbent Local Exchange Carriers (ILECs), better known as the phone companies, must move from ADSL to ADSL2+ and VDSL in order to offer the necessary bandwidth. FTTx deployments, through access technologies like Passive Optical Networks (PON), offer plenty of bandwidth from 100Mbps to 1Gbps. The major challenge for fiber and wireless is the capital expenditures involved in deploying a new infrastructure. Wireless

PSTN Cable Data Video

DSL

FTTx

Figure 1 Broadband Deployment Structures

SwitchCore Corp.

April, 2006

4

Enabling Triple Play in Access Networks

4 Migration from ATM to IP over Ethernet In order for Service Providers to offer Triple Play services, the underlying network infrastructure must be able to provide sufficient bandwidth and flexible control over the distribution of this bandwidth (provisioning) using QoS. The following are the bandwidth requirements for Triple Play services. Service HDTV Broadcast TV HSI VoIP

Bandwidth 12Mbps 2-6Mbps 3-10Mbps 64-750kbps

Table 1 Bandwidth Requirements for Broadband Services

The total bandwidth required is ideally in the range of 16 to 22Mbps. The median for Triple Play, however, is around 8 to 10Mbps. QoS can provide bandwidth guarantees, as well as maximum limits based on service and/or user. At present, most of the access and aggregation networks are based on ATM. ATM-based DSLAMs work well for first generation Internet traffic; however, they have limitations. Although ATM has built-in QoS, ATM-based DSLAMs are costly. The high cost is not just the cost of equipment but also the complexity of provisioning and maintenance. ATM also cannot meet the increasing amounts of bandwidth required, especially for uplinks. ATM-based DSLAMs are mainly Central Office-based, which limits their widespread deployment. Regardless of which type of access network (e.g. Cable or DSL) is deployed, a shift is taking place in the underlying technologies that support these Access Networks. IP over Ethernet provides a more costeffective solution, resulting in a major shift away from legacy TDM and ATM. By moving from ATM to IP over Gigabit Ethernet-based equipment, network providers can lower their costs and deploy higher bandwidth equipment. IP-DSLAM, Next-gen DLC, BLC, EPON, GEPON, and MSPPs are examples of IP based Access equipment. This type of equipment allows telecom and Internet companies like Vonage and Skype to currently offer low-cost VoIP services. Similarly, IP Television (IPTV) allows digital television service to be delivered over broadband connections using IP. PSTN

Triple Play ● Voice ● Video ● Data

Local Exchange

Video Server

Gig Ethernet Switches Service Provider #1

IP DSLAM

Service Provider #2

B-RAS Access Routers

Access

Aggregation

IP-Service edge

Service Networks

Figure 2 Access Network Infrastructure SwitchCore Corp.

April, 2006

5

Enabling Triple Play in Access Networks

5 QoS QoS is a top-level measure of service quality a user receives and can be separated into bandwidth, bit error rates, latency and jitter. The QoS considered in this document is restricted to bandwidth, latency and jitter (as bit error rates are a physical characteristic of the network). By default, the Internet delivers traffic as Best Effort due to the underlying network infrastructure being packet-switched and connectionless. The size of packets varies and the paths packets take from source to destination also vary. This and other details make Ethernet less deterministic than ATM which has built in QoS due to its fixed cell sizes and connection-oriented paths. To make up for this deficiency, the IEEE and IETF have defined standards (IEEE 802.1 P, Q and IETF Diffserv) which help provide different aspects of QoS over Ethernet for both Local Area Networks and the Internet. For Triple-play real time services like voice and video, QoS is mandatory. For QoS to be effective it has to operate end-to-end. If there are issues of lack of bandwidth or congestion at any point from source to destination, service is impacted. QoS on the Internet can be implemented using many techniques including bandwidth reservation, label switching, classification, congestion control and queuing techniques.

5.1

Bandwidth Reservation

Protocols for reserving bandwidth across a network, such as Int-Serv and more specifically the RSVP protocol developed under it, are not scalable and are difficult to deploy. Alternatively, service provisioning per subscriber at the edge and bandwidth deployment throughout the network can ensure enough bandwidth "reservation" for the voice and video services. Making sure that the sum of the voice and video services is not over-subscribing the networks is an important factor in controlling the latency and jitter of the services as well as ensuring the traffic reaches its intended destination.

5.2

Label Switching

Label switching such as MPLS has been designed mainly for core networks, so MPLS only solves part of the QoS problem unless it can be implemented out to the subscriber. This is not typical and so an alternative method of QoS must be available from the edge of the core out to the subscriber. Using VLANs as labels is being standardized in access applications (for example, IEEE 802.1ad and WT101). Traditionally, VLANs (defined by IEEE 802.1Q) provide an effective way to segment traffic on a LAN into different broadcast domains. They limit traffic to being switched between ports in the same VLAN. In IPaccess applications however, the use of the VLAN has expanded. The VLAN tag can be used to identify and isolate traffic from individual customers. The use of the 802.1p field within the VLAN tag can be used to retain the ATM priority information when the packet is converted from a legacy ATM network to the Ethernet networks of today. The VLAN ID field is 12 bits which limits the number of VLANs to 4K. If a VLAN was used to identify a subscriber to the edge of the WAN, this would be a severe limitation for Service Providers (SPs). SPs can get around this limitation by adding a second outer VLAN tag (S-VLAN) to the customer's inner VLAN (CVLAN). This Double VLAN-tag support is also called Stacked VLANs or Q in Q encapsulation and is standardized under the IEEE 802.1ad standard. The second VLAN ID is added to the header on the edge device and later removed on the destination device. For DSL applications the WT-101 standard recommends the following addressing: In the upstream direction, the subscriber is identified with the CVLAN and optionally the priority. In the downstream, the subscriber is identified with a combination of SVLAN, C’-VLAN, and MAC address (depending on service). SwitchCore Corp.

April, 2006

6

Enabling Triple Play in Access Networks

DSL Line/ VCC

DSLAM Line Card

DSLAM Switch Card Backplane

"B-RAS" (Router) Uplink

R

"C’-VLAN" CPE

"C-VLAN" + UserPriority (Logical line #)

L2 Termination "S-VLAN" VLAN rearrangement

Figure 3 VLAN Structure in an IP DSLAM

5.3

Classification

To address the QoS needs of Triple Play services, traffic has to be differentiated and grouped for treatment. In order for the traffic to be differentiated, a switch or router must be able to parse the packet header for different fields including source and destination addresses, type, priority and more. This parsing is known as classification. The groups resulting from the classification can be called classes or traffic, or Traffic Classes. It may also be possible to group Traffic Classes together to form Traffic Class Aggregates. This may be useful if, for example, the Traffic Classes group traffic for a single service and the Traffic Class Aggregate group all services handled by a service provider. Once the traffic has been classified into a category it can be accepted or discarded under congestion conditions / assigned a priority etc.

5.4

Congestion Control

Congestion control includes congestion avoidance and congestion management. Both approaches are an important way to address congestion and can be implemented both in hardware and software at different layers of the network protocol stack. This chapter considers congestion control in the hardware. Congestion avoidance controls the flow of traffic from one point in a network to the next so that the next hop is not oversubscribed. This may involve shaping the traffic to limit the output to the absolute maximum rate that the receiver can accept or it may involve policing traffic to limit it to an average rate that the receiver can accept. If policing is used, the receiver must be able to store a limited burst of traffic. If shaping is used, the burst must be stored at the local node. Policing may be used in front of shaping to ensure the load on the local traffic buffer is not too great. The diagrams below show the difference between the shaper and policing functions. As can be seen, the output from the remote node is the same in both cases. The key to the diagrams is: Received rate: Transmit rate from local switch: Transmit rate from remote node:

SwitchCore Corp.

April, 2006

7

Enabling Triple Play in Access Networks

Bandwidth

Bandwidth Data is buffered In local buffer

Max

Max

Data is output for remote buffering

Data is discarded

Data out from remote buffer

Local buffer is drained

Time

Time Figure 4a: Local Device Shaping

Figure 4b: Local Device Policing

Congestion management deals with congestion in a network node. It ensures that the important traffic is passed through the node and it clears congestion by discarding some of the less important traffic. Congestion management may involve assigning bandwidth guarantees and limits on groups of traffic (Traffic Classes and / or Traffic Class Aggregates) and / or on physical ports. This allows the traffic (or bandwidth) manager to consider anything within the guarantee as not contributing to congestion, whilst all traffic over the imposed maximum limits automatically adds to the congestion in the switch. This allows the execution of SLAs within the switch. Diff Serv offers an alternative method of congestion management and SLA execution:

Differentiated Services (Diffserv) In order to switch and route different types of traffic on the Internet, the IETF developed the DiffServ standard. The basic idea behind this is to move from a best effort network to one which can differentiate different classes of traffic, assign those appropriate priorities and condition the traffic so that it conforms to some agreement. The information about congestion at any particular point in the network can be passed to all nodes in the network. The benefit to real time traffic is that it is guaranteed an aggregate behavior based on the way the traffic is identified. A DiffServ node does this by taking the TOS field in the IP header and marking the packet with a certain priority called a DiffServ class. This DiffServ class is a 6 bit DSCP (DiffServ Code Point) which marks packets based on groups of classes. These classes are Best-effort, Assured Forwarding, Expedited Forwarding and Network Control. Assured Forwarding offers different levels of service across four classes. Assured Forwarding provides each user with a profile which determines how much traffic they are allowed to send through the network. If there is congestion and traffic has to be dropped, Best Effort will be dropped rather than Assured Forwarding traffic. Expedited Forwarding is a premium class of service comparable to having a dedicated leased line. DSCP

Precedence

Purpose

00000000

0

Best effort

00001000 00010000 00011000 00100000 00101000 00110000 00111000

1 2 3 4 5 6 7

AF Class 1 AF Class 2 AF Class 3 AF Class 4 Express forwarding Control Control

SwitchCore Corp.

April, 2006

8

Enabling Triple Play in Access Networks

A DiffServ Node has the following components which are implemented in Ethernet Switches and Routers at the edge and core of the network.

Meter

Packets

Classifier

Marker

Shaper/ Dropper

Figure 4 Logical View of Packet Classifier and Traffic Conditioner

Classifier The task of a Classifier is to look at the header information and select which flows deserve a service. A simple classifier, called a Behavior Aggregate (BA) selects packets based on the DSCP value only. Due to it’s simplicity it gives little to no flexibility in identifying and handling different types of traffic. A more flexible and powerful classifier that selects packets based on the content of various fields is called a Multi-Field Classifier. It classifies the packet looking at various fields in the header including SA, DA, DSCP, Protocol field, source and destination ports. Meter Monitors and measures the traffic flows identified by the Classifier. It determines if packets are in-profile or out-of-profile. Marker Sets the DSCP value which in turn determines the treatment the packet will receive, in the current network switch and also potentially in any future network switches it has to pass through. Shaper Delays packets in a traffic flow to bring them into compliance with agreed SLA. Dropper Discards packets in a traffic flow (called policing) to bring them in compliance with the SLA. In DiffServ every packet belongs to a “service” and has a color indicating the drop precedence for that packet. It is important to be able to do the following: 1. Detect and Act on the service and the color – select outgoing Queue and Assign a Traffic Class. 2. Set the Service and Color to defined one based on Classification 3. Meter the traffic and be aware of the color. Set color based on the Metering result. 4. Drop traffic using Drop Precedences (colors) SwitchCore Corp.

April, 2006

9

Enabling Triple Play in Access Networks

5. Schedule traffic based on service – using SP and /or Weighted Round Robin (WRR) 6. Shape outgoing traffic – to adjust traffic streams to meet SLAs

5.5

Queuing

Queuing guarantees a certain performance across the network. This results in a tiered structure which we experience in life e.g. in mailing packages. Best Effort could be analogous to sending a package by parcel post, which is sometimes called the “send and pray” philosophy. A faster way to send packages is by first class, and then there are premium services like Priority mail (two to three day delivery), and Express mail which guarantees next day delivery. Similarly email and other non-priority data can be sent using Best Effort. If there is congestion this is the type of traffic that will be dropped first. Even if there is no congestion, Best Effort traffic will be delivered after all other types of traffic. Voice should be offered at the highest service level (Express mail) to ensure the traffic reaches its destination with the minimum latency. This will also ensure low jitter rates. Video needs to be recognized as "Priority mail" to ensure traffic reaches its destination. Latency and jitter on video traffic are acceptable within strict limits thanks to buffering in the set-top box. Packets can be assigned priorities at different layers of the network protocol stack. The IEEE 802.1p standard defines 8 priority levels for delivering traffic across a LAN. This priority is only valid at Layer-2 and has to be removed or translated into a Layer-3 priority over the WAN. At layer three the IETF has defined the DSCP field for Diff-Serv (Differentiated Services). The DSCP is a six bit field which allows Internet traffic to be placed in different service classes. 802.1p and Diff-Serv are simple ways of assigning different priorities but are coarse ways of providing QoS, without guarantees.

6 Implementing QoS for Triple Play In order to offer IP based Triple Play services it is important for Ethernet switching equipment to offer the following:

6.1

General features •

Flexible bandwidth provisioning per service or subscriber.

•

Enforce the SLAs for each customer/service guaranteeing a minimum and maximum bandwidth. Distributing the remaining bandwidth among customers (this increases efficiency and enables SPs to sell over-provisioned services.)

•

The ability to scale customers and services into the thousands as technology improves the ability to add more customers per system. As bandwidth increases also need ability to add more services per customer.

•

Provide a way to group traffic, subscribers, services, and/or service providers to provision SLAs. Include bandwidth (min, max, RED curves) and other QoS features. Traffic can be classified into groups called Traffic Classes (TCs).

•

Recombine the groups (TCs) to assign QoS. This regrouping can be called “Traffic Class Aggregates (TCA)”. TCA’s simplify provisioning allowing multiple providers per system. Bandwidth allocation and policing can be managed per Aggregate. 128 to 256 TCAs is a good

SwitchCore Corp.

April, 2006

10

Enabling Triple Play in Access Networks

target figure. •

Provide three separate SLA enforced service levels for voice, video and Internet data to each subscriber. If each subscriber has 4 services, 1K subscribers will need a total of 4K services. This would mean 4K TCs and 2K or 4K meter/markers.

•

Detailed statistics per subscriber, service, traffic class to allow billing. Need statistics for incoming, discarded, and forwarded traffic per color, per traffic class, per port, and per priority. The greater the granularity and number of counters, the more flexibility for service providers from an accounting and billing perspective. The ability to keep track of thousands of customer makes it possible to enforce SLAs higher up in the aggregation network.

6.2

VLANs •

Flexibility in identifying customers/services: Many different fields may identify the customer, like VID, IP SA, IP DA, MAC SA, MAC DA, Port, etc. There must be an efficient yet flexible ways to identify the customers.

•

Extensive VLAN Support, including large number of VLANs (up to 4K), VLAN swap capability, and ability to add/remove Service Provider-VLANs. These operations are called Swap and Push (when an ID is added) or Pop and Swap (when ID is removed).

•

In ATM-to-Ethernet migration, custom VLAN tagging is a good substitute for VC/VP in ATM. Custom VLAN tagging and forwarding simplifies the migration by providing ATM-primitives in an Ethernet environment.

•

DSL applications need the ability to handle One Subscriber to one VLAN (1:1), Many Subscribers to one VLAN (N:1) or Many nodes acting as one VLAN (called Transparent LAN Services (TLS)), all defined in the WT-101 standard. Each type of LAN mapping and direction of traffic flow upstream versus downstream needs the mapping of Subscribers, VLANs, MAC addresses (in some cases), ports, 802.1p priority.

6.3

DiffServ •

Provide full DiffServ support including Policing and Traffic Shaping, providing per hop quality of service

•

A color classifier which assigns incoming packets with a color based on 802.1P priority and/or the DSCP

•

Color aware bandwidth management, metering and marking for each subscriber and service provider to provide different levels of service to each customer.

•

Ability to shape traffic that exceeds its quota using programmable algorithms which prevent congestion downstream. Need to be able to delay traffic based on multiple shapers per port.

•

Ability to discard traffic that exceeds its quota using programmable algorithms which prevent congestion downstream. Need to be able to discard based on appropriate factors.

SwitchCore Corp.

April, 2006

11

Enabling Triple Play in Access Networks

7 Summary In order for Service Providers to offer Triple Play services, IP-based access equipment needs to offer a wide and flexible range of QoS features. Ethernet switching technology provides the QoS benefits, the high bandwidth and the low cost that is required in different types of networking equipment. Flexible bandwidth provisioning, support for DiffServ, VLANs and detailed statistics are some of the key features that are required to provide services for voice, video and data. All the features discussed should be implemented in a one chip solution to lower the number of components in the Bill of Materials. For more information please refer to “Building Next-Generation IP-DSLAMs using the Xpeedium2®pro Access Aggregation Switches,” at www.Switchcore.com.

SwitchCore Corporation

SwitchCore AB

2077 Gateway Place, Suite 220 San Jose, CA 95110 USA Phone: +1 408 436 7200 Fax: +1 408 436 7206

Emdalavägen 18 SE - 223 69 Lund . Sweden Phone: +46 46 270 2500 Fax: +46 46 270 2581

www.switchcore.com

email: [email protected]

Xpeedium2® is a US registered trademark of SwitchCore.

SwitchCore Corp.

April, 2006

12