VoIP Transport Reliability on 5E-XC OIU-IP

Transcription

1 VoIP Transport Reliability on 5E-XC OIU-IP Preserving call stability on the OIU-IP-enabled 5E-XC Switch Reliable network configurations will enable VoIP services on the 5ESS switch to survive node and link failures. This paper demonstrates the reliability and OAM&P capabilities offered by Synchronous Optical NETwork/SynchronousDigital Heirarchy (SONET/SDH) compared to Gigabit Ethernet.

2 Contents Abstract...3 Introduction Meeting the challenges of VoIP...3 Reliability options...4 Packet over SONET...4 PPP considerations...5 Layer 1 protection...6 Layer 2 protection...6 Layer 3 VRRP/HSRP...8 OAM&P management considerations...9 Long-haul considerations...9 Recommendations/Conclusion...10 Acronyms...11 Defect per million methodology

3 Abstract Voice over Internet Protocol (VoIP) services are becoming more widely deployed as enterprises drive demand for lower cost voice networks. Ethernet has traditionally provided a relatively inexpensive and efficient transport for data applications that are bursty in nature over Local Area Networks (LAN). Ethernet reliability and downtime requirements are not as stringent as those of a SONET/SDH based voice network. SONET/SDH technology was deployed with voice networks in mind by providing a reliable transport that is resilience to network faults. Yet, Ethernet s simplicity, cost, increased (Gigabit) speed and fiber reach are prompting some carriers to consider LAN-based technology as an evolutionary replacement to their SONET/SDH voice networks. However, many believe that near term 1, SONET/SDH cannot be replaced by LANbased technology at the end office for VoIP applications until network reliability is addressed. Introduction Meeting the challenges of VoIP 1 One report from the Yankee Group suggests that this transition of Ethernet from LAN to displacing SONET in a service provider network could prevail in seven to ten years. Voice is a real-time application that cannot tolerate high network delay and variability within that delay, known as latency and jitter respectively. Voice reliability at the end office, which is the primary focus of this paper, cannot tolerate indeterminate network outages or equipment glitches that cause loss of service. Circuit-based Time Division Multiplex (TDM) end office engineering overcame network reliability issues by duplicating or sparing network equipment that transports voice into the long haul network. Figure 1 represents the reliability of the end office network interface. Though this paper briefly touches on Long-Haul (Metro) considerations of SONET/SDH and Gigabit Ethernet, the paper s main discussion is on the reliability factors between the end office switch (i.e., a network element like the OIU-IP in the 5ESS Switch) and the first edge device (such as an IP router) that connects to a long haul network. For a VoIP network, this long haul network is referred to as the IP core in this paper. Reliability focuses IP Core 5ESS End Office A Edge Edge 5ESS End Office B Figure 1: Point-to-Point Reliability Focus Many believe today s challenge is to merge IP technologies with today s existing SONET/SDH based voice network, rather than replacing technologies. This evolution offers packet switching efficiency over a reliable network for both data and voice services from the end office network interface. 3

4 2 Ethernet over SONET (EOS) is not discussed in this paper. 3 POS means providing IP via PPP over a SONET Optical Fiber interface whereas a GbE means (UTP electrical or Optical) 1000Mbit/sec interface. This paper discusses some possible network configuration options that combine IP over SONET/SDH and Ethernet for VoIP. The network configuration options compared in this paper focus on voice reliability that Packet over SONET (POS) 2 and Gigabit Ethernet, known as a GbE 3, provide to meet carrier grade voice network requirements. Reliability options Many vendors implementations of POS offer protection against link failures with 1+1 Automatic Protection Switching (APS). SONET APS provides a switchover time of 50msecs, which supports the preservation of stable calls. The remainder of this paper discusses some reliability options offered by POS and GbE. Ethernet lacks an equivalent SONET sub-50 msec restoration capability. One effort within standards is Resilient Packet Ring (RPR), which claims 50 msec restoration. However, RPR is far from being a standard. In addition, it is a replacement for Ethernet, so the cost to build an entire RPR network to replace Ethernet between a switch and the edge routers is difficult to justify. Packet over SONET In Figure 2, both interfaces from the Network Element (NE) connect to the same edge router. A failure of the router in this configuration will result in a loss of all the traffic served by the Network Element links even working together as a pair. From the NE perspective, the interfaces are operating as 1+1. From the router perspective, the interfaces are also operating as 1+1. A failure of a single POS interface (either on an NE or on the router) will not impact stable calls and meets the standard APS requirement for recovery and switchover. Obviously, the weakest link for reliability in this configuration is the edge router, which is the next consideration to split the protected links among different edge routers. OC- 3c (Working) OC- 3c (Protection) 5ESS NE Figure 2: POS APS Protection, Single In Figure 3, the interface from the NE connects to two different edge routers. This configuration provides protection against a failure of a single router. Many router vendors support this configuration. From the NE perspective, the interfaces are operating as 1+1. In order to provide APS across two routers, they must be interconnected to allow for communication to maintain the APS states. From a standards interoperability perspective, both sides must conform to GR-253, by appropriately populating the K1, K2 bytes for APS. 4

5 OC- 3c (Working) OC- 3c (Protection) 5ESS NE Figure 3: POS APS Protection, Two s PPP considerations Both the routers and the NE use the Point-to-Point 4 Protocol (PPP) to monitor the link group status. IP packets are sent over the selected (working) link as determined by PPP. For POS, determining the status of the non-selected side is unlike TDM. In TDM switching architecture, the non-selected side is typically in a standby state, meaning it is fully ready to accept a switch and begin carrying traffic. This is not the case for POS, where the PPP link is typically active only on the selected side. After a switch, the two ends must try to bring up the PPP link on the newly selected side. This initialization of the PPP link is in addition to the APS switchover times. Gigabit Ethernet There is no 1+1 APS equivalent at the Ethernet layer today. For IP over Ethernet there are different protection options at different layers. Some are listed below. Layer 1 Redundant physical links via hubs Layer 2 IEEE802.3ad, Link Aggregation Group L2 Switching Layer 3 Virtual Redundant Protocol (VRRP) Hot Standby Protocol (HSRP) (from Cisco) 4 The PPP protocol was chosen on the 5ESS switch, since it the most widely supported by vendors of IP routers. PPP is comprised of three main components a. A method for encapsulating multiprotocol datagrams. b. A Link Control Protocol (LCP) for establishing, configuring, and testing the data-link connection. c. A family of Network Control Protocols (NCP) for establishing and configuring different network layer protocols. Depending upon network configuration, the above methods may or may not preserve stable calls. 5

6 Layer 1 protection Figure 4 shows a simple example where physical links are used from each NE to a hub that supports an Ethernet interface back to an edge router. Hub 5ESS NE Figure 4: Layer 1 Hub Protection This example relies on the hub, but the use of a hub comes with the price of lost bandwidth to avoid collision and capital expense of using a hub. Possible reliability downtime also needs to be considered. From the NE perspective, the Ethernet link may be configured as a protection group, where only one is used for transmitting at a time. Note that the Hub sends traffic over the other two Ethernet links (back to the NE and the router). There are other configurations where a single Ethernet interface from the NE pair connects to two different edge routers via a hub on the same net. Nevertheless, the result on stable calls is the same when the link fails at the NE. Layer 2 protection Depending on switchover and detection time, the Layer 2 methods might offer suitable protection against stable calls lost on link failures. Link Aggregation is based on IEEE802.ad recommendations and Ethernet Switching is an approach relying on Layer 2 mechanisms such as Address Resolution Protocol (ARP) for publishing the active link to use for switching. Link aggregation The primary use of Link Aggregation is to increase link capacity by combining multiple links into one logical link. During normal operation, the links within a Link Aggregation Group (LAG) are load shared. In the event of failure, service is preserved and load capacity is effectively reduced and redistributed among the other links via LAG reconfiguring time to converge traffic to the remaining links. The IEEE802.ad standard set a goal for a rapid reconfiguration to occur within one second but there could be variances among vendors implementations. Another consideration is that the links within a LAG typically cannot be spilt between routers, meaning there is no protection against a router failure. 6

7 LAG 5ESS NE LAG Figure 5: Link Aggregation Group Ethernet switching In Figure 6, the Ethernet interfaces from the NE connect (via Layer 2 switches) to two different routers at the network edge. All the Ethernet interfaces are in the same Local Area Network (LAN). The Layer 2 switches may be integrated in the routers, assuming the routers support Layer 2 capabilities. If the Layer 2 switches were independent of the routers, they would also need to be considered from a reliability standpoint. L2 Switch 5ESS NE L2 Switch Figure 6: Layer 2 Switching This type of configuration makes use of Layer 2 mechanisms for recovery. Depending upon the specific configuration and capabilities of the NE, Layer 2 switches, and routers, stable calls may be preserved under failure scenarios. From the NE perspective, the Ethernet interfaces may be in a protection group, transmitting only on one Ethernet interface. It would receive traffic via standard layer 2 learning methods. On the Layer 2 switch side, the Ethernet interfaces would both be active, transmitting and receiving traffic via Layer 2 learning methods. In addition, connectivity between the Layer 2 switches is assumed to allow traffic between them. There are a number of possible configurations, each with different implications. Some possibilities would include the use of floating IP addresses, gratuitous ARPs, and/or VRRP/HSRP. Note that in the case of VRRP/HSRP, it will be necessary to have a configuration, which allows an NE Ethernet interface to have connectivity to both routers (see next section). 7

8 Layer 3 VRRP/HSRP VRRP was developed within IETF as a standard approach to eliminate single point of failure inherent in a static default routed network (RFC 2338). HSRP (RFC 2281) is a proprietary protocol developed by Cisco to address the same issue. VRRP/HSRP provides a set of routers working together to present a single virtual router to the Network Elements on the LAN. A virtual router is an abstract object managed by VRRP (or HSRP) that acts as a default router for hosts on a shared LAN. The scope of each virtual router is restricted to a single LAN. A Virtual consists of: Virtual Identifier (VRID), Set of associated IP addresses across a common LAN A VRRP router may backup one or more virtual routers Failover typically takes multiple seconds VRRP/HSRP LAN 5ESS NE Figure 7: VRRP/HSRP Figure 7 shows a NE connected to two routers via a LAN. The LAN is depicted logically, representing a physical network, which may consist of Layer 2 switches. The role of the LAN is to provide a path from a single NE Ethernet link to both routers. If a router fails, the backup VRRP/HSRP router will take over, which is transparent to the NE. 8

9 OAM&P management considerations SONET is a mature, industry-wide accepted high-speed transport that provides substantial overhead information, allowing for simpler multiplexing and greatly expanded OAM&P capabilities between SONET termination equipment. The SONET overhead is used to manage communications between endto-end paths for the STS-N 5 payload service mapping. The SONET line overhead is used for STS-N signals between STS-N multiplexers like an ADM. Lastly the SONET section is used for communications between adjacent network elements. 5 There is also a path termination when DS1 signals are mapped into virtual tributaries for STS-1 payloads. For Ethernet there is no embedded or industry wide accepted management at this transport level. There are however, a number of measurements found in MIB-II (RFC1213), RMON (RFC1757), and SMON (RFC2613) for Group function counts for the Ethernet Interface, Ethernet packet Statistics, History, Events, and Alarms. These MIBs are mature and have been approved by IETF for sometime, but this does not mean that they are universally supported. In many cases, many of the OAM&P function groups in MIB-II are ignored, and the RMON and SMON MIBs are usually provided as add-ons. This means that there is no real end-to-end OAM&P Network Management standard at the GbE, as compared with the embedded OAM&P SONET capabilities. Long-haul considerations As stated, Ethernet traditionally has been used for Local Area Networks to provide a cheaper transport within a local IP network for data applications that have not needed voice carrier grade five-nines uptime reliability. As Ethernet moves into the Metropolitan Area Network (MAN) and Wide Area Network (WAN) markets, it is still unknown if this technology will in the near-term achieve such carrier grade reliability standards. As for distance between nodes in a LAN, fiber has improved this reach, but at the edge, the Ethernet transport needs to be multiplexed over a long-haul WAN. GbE can be transported over a long haul network through a dedicated lambda (wavelength). Dense Wavelength Division Multiplexing (DWDM) networks have emerged to multiplex a GbE stream on to a single lambda wavelength. Direct layering of IP on to DWDM is not possible in the near future. As such, using GbE requires that an edge router or a next generation Optical ADM that supports G.709 optical framing be located relatively near the end office to terminate and multiplex the GbE into packet core long-haul transport, regardless of the access traffic that the office offers. 9

10 6 An SPE is also divided into two parts: the Path overhead and the revenueproducing traffic being transported over a SONET network. SONET-SDH, on the other hand, embeds the payload known as the STS-1 Synchronous Payload Envelope into the SONET STS-1 (51.84Mbps) frame. An STS-1 frame consists of two areas: The SPE and the Transport (Section and Line) overhead 6. Once data is multiplexed into the SPE, it can be transported and switched through a SONET network without being examined at each intermediate node. As such, providing IP packet over SONET offers much more flexibility to reuse existing SONET equipment that is virtually embedded into the existing long-haul WANs. SONET transport seamlessly allows the STS-1 payload to be transferred over the WAN without requiring edge routers co-located at the end office to terminate and multiplex the payload for the long haul network. Recommendations/Conclusion Lucent recognizes that today s LAN based technology is cheaper to deploy, is ubiquitous in enterprise data networks, and will mature to address the voice reliability and OAM&P required for carrier networks. Nevertheless, as Ethernet matures to meet carrier grade reliability demands, the cost to deploy Ethernet between the end office switch and the first edge device will increase. This future reliability cost makes Ethernet less attractive as a replacement for the SONET-based interface between the end office and the first edge device in a VoIP network. When factoring in the carrier grade voice reliability discussed in this paper, Lucent recommends that the network interface (i.e. 5E-XC Optical Interface Unit) at the end office be better served near term by merging IP technologies with today s existing SONET/SDH technologies. 10

11 Acronyms Acronym ADM ARP APS DPM DS-0 DS-1 DWDM EMS EO GbE HRSP ILEC IP K1/K2 LAN LCP LEC MIB NGN OAM&P OFI OIU-IP POS PPP QOS RPR RMON SMF SMON SONET/SDH STE STS TCP TDM TE TG UDP VOIP VRRP WAN Meaning Add/Drop Multiplexer Address Resolution Protocol Automatic Protection Switching Defect Per Million Digital Signal-Level 0 (64 kb/s) Digital Signal-Level 1 (1.544 Mb/s) Dense Wavelength Dvision Multiplexing Element Management System End Office Gigabit Ethernet Hot Routing Standby Protocol Incumbent Local Exchange Carrier Internet Protocol Section Overhead bytes for Automatic Protection Switching Local Area Network Link Control Protocol Local Exchange Carrier Managed Information Base Next Generation Network Operations, Administration, Maintenance and Provisioning Optical Facility Interface Optical Interface Unit Packet Over SONET Point to Point Protocol Quality Of Service Resilient Packet Ring Remote Monitoring Single Mode Fiber Switch Monitoring Synchronous Optical Network/Synchronous Digital Hierarchy SONET Terminating Equipment Synchronous Transport Signal Transmission Control Protocol Time Division Multiplexing Termination Equipment Trunk Group User Datagram Protocol Voice Over Internet Protocol Virtual Redundancy Protocol Wide Area Network 11

12 Defect per million methodology Defect Per Million (DPM) is a method commonly used for quantifying call reliability. The network elements and their network interfaces serve as the inputs to calculate the cumulative DPM ratio. The data being provided here serves as a reference on how total DPM is calculated. 7 Note that DPM results are independent of the switch size since DPM is a ratio number of calls lost per million calls processed. The total DPM 7 is the sum of two factors: the DPM contributions from the individual network elements, and the calls lost due to total downtime of each network element. For a network element, the DPM contributed by the loss of transient and stable calls is a function of the following factors: the number of switch element failure events per year; the number of calls in a transition state on the switch element at the time the failure event occurs; the number of calls in a stable state on the switch element at the time the failure event occurs; and the number of calls handled per year by the switch element. Predicted DPM is then given by: D e = defects per million due to element e T i = transient calls lost when failure i occurs on element e F i = failure rate for failure i C e = calls per year handled by element e S i = stable calls lost when failure i occurs on element e De = ( (Si + Ti )Fi ) / Ce This paper does not attempt to calculate DPM for any of the POS or GbE configurations discussed in this paper. The architecture and design detail necessary to perform these calculations is beyond the scope of this paper. Instead, different possible reference architectures for POS and GbE are considered which can help to preserve stable calls in a VoIP network. To learn more about our comprehensive portfolio, please contact your Lucent Technologies Sales Representative. Visit our web site at This document is for planning purposes only, and is not intended to modify or supplement any Lucent Technologies specifications or warranties relating to these products or services. The publication of information in this document does not imply freedom from patent or other protective rights of Lucent Technologies or others. Copyright 2003 Lucent Technologies Inc. All rights reserved VoIPTransp.v