APPLICATION NOTE Alcatel-Lucent Virtualized Service Router - Route Reflector Creating a new benchmark for performance and scalability
Abstract Network functions virtualization (NFV) has attracted much recent attention. NFV transforms specific network functions that run on specialized, dedicated hardware platforms into software functions implemented on general-purpose computing platforms. NFV offers added flexibility and faster time-tomarket, enabling new ways to deliver services. Alcatel-Lucent has over a decade of industry leadership and innovation in IP routing platforms, Service Router OS (SR OS) software, and network/service management. The company is embracing the move to NFV and is committed to delivering virtualized solutions that help network operators reach their NFV goals. Leveraging the SR OS and designed for x86 server environments, the Alcatel-Lucent Virtualized Service Router () is the industry s first virtualized IP/MPLS router. BGP route reflection is an ideal function for virtualization. The - Route Reflector (-) leverages key innovations and has set a new benchmark for performance and scale compared to alternate virtualized implementations.
Table of contents Introduction / 1 BGP s / 1 Virtual BGP s / 2 Alcatel-Lucent : Architected for NFV / 2 Parallel SMP / 3 64-bit OS support / 4 -: Flexible deployment options / 4 -: Raising the bar for performance and scale / 5 - reference information / 6 Standards and protocols / 6 Host, hypervisor and licensing information / 6 Conclusion / 7 Acronyms / 7 References / 8
Introduction Network functions virtualization (NFV), defined by the European Telecommunications Standards Institute (ETSI), 1 aims to transform the way that network operators design networks by evolving network equipment functions onto industry-standard, high-volume servers. The application of NFV introduces many benefits, including rapid scaling of services, faster time-to-market, and in some cases reduced costs from economies of scale by using standard IT-server virtualization technology. Building on over a decade of investment and industry recognition with its Service Router (SR) portfolio, Alcatel-Lucent has introduced the industry s first virtualized IP/Multiprotocol Label Switching (MPLS) router: the Alcatel-Lucent Virtualized Service Router (). The is highly flexible, enabling rapid service innovation, elastic scalability, and lower operating costs with a homogenized physical infrastructure and NFV management and orchestration. Based on the Alcatel-Lucent Service Router Operating System (SR OS) and architected for x86 server environments, the delivers the IP service richness, reliability and resiliency required to gain a competitive edge in delivering IP networking services. The provides a suite of virtualized functions and services: 2 This application note focuses on the Route Reflector (-). The - is functionally and operationally equivalent to a 7750 SR hardware-based BGP, with the added benefits of delivering higher flexibility, scale and performance. For information about other virtualized functions, see the most recent Alcatel-Lucent data sheet. BGP s In a standard BGP configuration, a BGP route learned from one Internal BGP (ibgp) peer is not re-advertised to another ibgp peer. This rule exists because of the assumption of a full ibgp mesh within an autonomous system. A full ibgp mesh imposes scaling challenges when many BGP peers exchange large amounts of routing information. BGP route reflection eliminates the need for a full ibgp mesh by using one or more designated BGP s to re-advertise routes from one ibgp peer to another ibgp peer. A and its ibgp peers (clients) form a cluster. To prevent single points of failure, a cluster may implement multiple s for redundancy. In addition, route reflection duties can be distributed over different BGP s based on address families. In today s networks, the function is predominantly run on an IP router that is dedicated for route reflection or that performs the function in addition to other IP routing and services functions. The function is primarily run on the router control card, and the performance and scale of route reflection is mostly determined by the control card s hardware specifications (CPU type, speed and memory). 1 Margaret Chiosi et al., Network Functions Virtualisation: An Introduction, Benefits, Enablers, Challenges & Call for Action. SDN and OpenFlow World Congress, Darmstadt-Germany, October 2012. http://portal.etsi.org/nfv/nfv_white_paper.pdf 2 Please refer to the Virtualized Route Reflector () datasheet. Link available in the References section of this document 1
Higher-end routing platforms with next-generation control cards are often required to meet requirements in today s networks. A router used solely for functions is underutilized in the data plane because functions require minimal data-plane resources. Conversely, a router that shares the function with other IP routing functions may not have sufficient resources (CPU and memory) to support scalable route reflection. A router-based model is restricted in its ability to flexibly scale up improve performance and scale with evolving network requirements. A new router platform may be required to support additional capabilities. In addition, some network operators may want to deploy separate s for specific applications, such as Internet routing, Layer 3 (L3) virtual private network (VPN), and Layer 2 VPN. This inflexibility increases costs related to deploying and scaling s. Virtual BGP s BGP route reflection is mainly a control-plane function with minimal traffic in the data plane, making it ideal for virtualization. operation is memory and compute-intensive and is well suited to run on x86-based server platforms. A virtualized (v) removes reliance on dedicated hardware, which lacks sufficient performance and scale in the control plane and/or which may be underutilized in the forwarding path. A v offers more flexible deployment options and upgrades for improved scale and performance. Scale and performance levels can be adjusted up or down as needed by flexibly allocating server (host) resources to the v. A v can be deployed as a single for all applications or with dedicated s for specific address families, such as Internet routing, L3 VPNs and L2 VPNs. This flexibility enables innovative deployment and operational models: for example, a v instance could be automatically restarted on a new server in case of failure of the v that previously executed the function. Alcatel-Lucent : Architected for NFV Moving to an NFV deployment model requires that virtualized network functions take full advantage of the underlying IT-server virtualization technologies (with extensive multicore support and significantly larger memory capacity offered by host machines). The Alcatel-Lucent Virtualized Service Router () is based on a real-time, modular, highly faulttolerant design. It features several industry-leading architectural innovations for implementing virtualized network functions. Robust and highly scalable, the is purpose-built for x86-based hosts and leverages the widely deployed and field-proven SR OS. The has been designed for reliability and High Availability (HA), and it features several innovations that take advantage of the unique attributes provided by x86-based hosts. 2
The and its network applications are managed by the Alcatel-Lucent 5620 Service Aware Manager (SAM), which also manages traditional Alcatel-Lucent 7750 SR network applications. Alcatel-Lucent offers a comprehensive NFV and software defined networking (SDN) portfolio that includes: CloudBand Management System Nuage Networks Virtualized Services Platform With a broad ecosystem of partners, Alcatel-Lucent provides network operators with a complete, fully integrated NFV infrastructure. Table 1 lists key features and benefits. Table 1. features and benefits FEATURE/BENEFIT Flexible deployment models Designed for scale Real-time OS Modular process architecture Fault-tolerant design Best practises in large-scale system design Strict separation of control-plane and data-plane functions DESCRIPTION Deployment options: Integrated system (single virtual machine [] supports control-plane and data-plane functions) Distributed system (control-plane and data-plane functions distributed across multiple s) - is implemented using the integrated system model. -PE and -AA can be implemented using the integrated or distributed system model. ing power scales up as [virtual] CPUs and memory are added to a specific and scales out as s are added in a distributed deployment model. Fine-grained scheduling prioritizes high priority processes for faster convergence and control-plane responsiveness while maintaining fairness. separation protects individual processes. Highly reliable Non-Stop Services for mission-critical applications, with HA and in-service software upgrades for the distributed system model Optimized custom OS, rigorous design practises, strict APIs, and memory protection that scales across diverse architectures Supported for the distributed deployment model In addition to the capabilities listed in Table 1, two innovations provide significant benefits for virtualized network functions: parallel Symmetric Multi ing (SMP) and 64-bit OS support. Parallel SMP Multicore CPUs provide highly scalable processing power that improves routing convergence times for high-performance routing applications and services. To further enhance the power of multicore processing, Alcatel-Lucent has created a unique approach to software OS design. The supports parallel SMP, a multi-threaded software approach that enables the concurrent scheduling and execution of different processes on different processor cores. With SMP supported in the SR OS since 2009 the unleashes the power and performance of multicore processing to deliver unmatched performance. SMP maximizes performance by taking full advantage of the multicore capabilities of Intel x86-based hosts (servers), as shown in Figure 1. Running processes in parallel across multiple CPU cores provides impressive performance benefits. In contrast, a solution that does not support multiprocessing runs on a single core, significantly impacting performance and providing no scale-up capability. In a single-processor implementation, the underlying OS cannot take advantage of the additional CPU cores. 3
Figure 1. Maximizing performance with SMP Virtualized Core 0 Core 1 Core 2 Core 3 Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 Core 6 Core 7 Core 4 Core 5 Core 6 Core 7 SMP with Alcatel-Lucent Single processor implementation with Vendor X 64-bit OS support With a 64-bit OS, processes can access a much larger memory address space compared to a 32-bit OS. This capability has been proven for 7750 SR deployments. The latest 7750 SR control processing cards (SF/CPM5 with 16 GB of RAM) support 8 million unique IPv4 routing information base (RIB) routes with a 32-bit OS image and 46 million RIB routes an almost six-fold increase with a 64-bit OS image. x86-based servers can support hundreds of GB of memory space. A 64-bit OS can fully exploit this additional space with the potential for enabling very high scaling for specific virtualized functions. -: Flexible deployment options The - is deployed as an integrated single system that includes the control-plane and dataplane functions. Each is assigned virtual resources CPU, memory and network interface cards from the host s available resource pool. For maximum scaling flexibility, resources can be scaled up or down as required to meet evolving performance and scaling needs. The - is modeled as a router, with a control card and a line card that supports five Gigabit Ethernet ports and one or more CompactFlash devices. The - supports the same routing features as a 7750 SR-based (relevant to a control-plane ). Virtual route reflection can be deployed with a single - that supports all applications or, as shown in Figure 2, with separate -s for separate address families and applications Internet, L3 VPN, L2 VPN with the option to deploy multiple -s for redundancy. -s can be independently scaled up or down to meet specific application requirements (e.g., L3 VPN - may be configured to support more resources than the Internet or L2 VPN -). Figure 2. Dedicated hardware and virtual s Dedicated Route Reflectors Virtual Route Reflectors Internet Internet L3 VPN L2 VPN Internet L3 VPN L2 VPN Virtualization L3 VPN L2 VPN Hypervisor Hypervisor Host Host The - is managed under the 5620 SAM and NFV orchestration systems, which enable the dynamic, rapid and flexible introduction of - functions. 4
-: Raising the bar for performance and scale The - leverages parallel SMP and a 64-bit OS support to set a new industry benchmark for performance and scale: Increasing CPU resources increases performance for route learning and. Increasing memory resources increases the scale and number of BGP peers and routing entries. performance was tested with the - and was compared with an alternate commercially available v solution ( Vendor X ) that is not optimized (does not support SMP) for x86-based deployments. Test methodology The v test setup matched real network-deployment scenarios, including route learning and route reflection (515,000 IPv4 prefixes) for the full IPv4 routing table for increasing number of BGP peers (100 to 1000 peers). The test measured convergence time: the time taken by the v to learn and reflect all routes for an increasing number of BGP peers. Test environment Test equipment from a leading test equipment vendor emulated the BGP peers, providing the Internet prefixes and the emulated BGP clients. Equipment/function specifications for the v tests are listed in Table 2. Table 2. Host and hypervisor specifications EQUIPMENT/FUNCTION Host Hypervisor SPECIFICATIONS Intel Xeon E5 v2 host platform with a dual CPU motherboard (single CPU used: Intel Xeon E5-2687W v2, 3.4 GHz with 8 cores/16 threads) Support for Intel Hyper-Threading Technology, with two processing threads per physical core 3 Host OS: Linux Linux Kernel-based (K) Each was configured to support: 8 cores (16 virtual CPUs [vcpus]/16 threads) 24 GB RAM Test results The performance of the two v implementations is shown in Figure 3. The test results illustrate the significant advantage of the x86 optimized design of the - implementation over an alternate v. Figure 3. Test results: - and alternate v implementations Number of clients 1000 500 250 100 41 23 Vendor X 88 180 172 346 Alcatel-Lucent 720 1534 Alcatel-Lucent - 8x faster 0 500 1000 1500 2000 Convergence time (seconds) 3 Highly threaded applications can complete more work in parallel, completing processes sooner. 5
For the test setup with 100 peers, the - converged in just 23 s while the alternate solution required 172 s to learn and reflect all routes. The - performance advantage was maintained when the number of clients (BGP peers) increased to 250, 500 and 1000. In each scenario, the - performed approximately eight times faster than the alternate implementation. This significant performance advantage can be attributed to SMP support on the -. SMP dynamically allocates cores (vcpus) on demand and runs processes in parallel. In Figure 4, a command-line interface (CLI) screen shows how the - took advantage of the multicore capabilities of the Intel x86 host. The alternate v solution, which was not architected for SMP, used a single core and was therefore unable to match the performance of the -. Figure 4. Host CPU utilization: - and Vendor X v 4 Alcatel-Lucent - vcpu utilization 0 [ IIIIIIIIIIIIIIIIIII 33.3% ] 1 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100.0% ] 2 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 78.7% ] 3 [ I IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 83.6% ] 4 [ II 3.3% ] 5 [ I IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 76.5% ] 6 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100.0% ] 7 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100.0% ] Mem [ IIIIIII 4656/128743HB ] Swp [ 0/0HB ] 8 [ IIIIIIIIIIIIIIIIIIIIIIIIIIII 52.5% ] 9 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100.0% ] 10 [ I 1.7% ] 11 [ II 3.3% ] 12 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 83.6% ] 13 [ II 3.3% ] 14 [ I IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100% ] 15 [ IIII 6.8% ] Tasks: 52, 29 thr; 7 running Load average: 5.55 283 1.18 Uptime: 1 day, 15:34:31 PID NI VIRT RES SHR MEM% PRI - - S CPU CPU% TIME+ Command 30306 0 23.8G 3421M 6540 2.7 20 - - S 8 656 22:36.10 /usr/libexec/qemu-kvm -name -... Vendor X v vcpu utilization 0 [ IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 100.0% ] 1 [ 0.0% ] 2 [ 0.0% ] 3 [ II 3.2% ] 4 [ II 3.3% ] 5 [ 0.0% ] 6 [ 0.0% ] 7 [ 0.0% ] Mem [ IIIIIIIIIIIIIIIIIIIIIIII III 24811/128743HB ] Swp [ 0/0HB ] 8 [ 0.0% ] 9 [ 0.0% ] 10 [ 0.0% ] 11 [ 0.0% ] 12 [ 0.0% ] 13 [ 0.0% ] 14 [ 0.0% ] 15 [ 0.0% ] Tasks: 50, 32 thr; 2 running Load average: 0.17 0.09 0.09 Uptime: 1 day, 13:35:34 PID NI VIRT RES SHR MEM% PRI - - S CPU CPU% TIME+ Command 23305 0 10.7M 896 756 0.0 20 - - S 0.0 656 00.00.0 /usr/libexec/qemu-kvm -name vrr-vendor X... - reference information Standards and protocols For information about - routing/policy standards and protocol support, see the Standards and protocol support sections of the 7750 SR product guides. Host, hypervisor and licensing information Please refer to the latest Virtualized Service Router () datasheet. Link is available in the References section of this document. 4 The CLI screen shows host utilization. Cores (vcpus) that show minimal loading are allocated to Linux host OS processes. 6
Conclusion Alcatel-Lucent is embracing virtualization for IP/MPLS service routing. To help network operators realize this vision, the company has introduced the Alcatel-Lucent Virtualized Service Router (): the industry s first full-service virtualized SR. For virtualized IP routing functions to deliver true flexibility, scale and performance, the virtualized function implementation must take full advantage of underlying IT-server virtualization technologies. The - supports innovations such as parallel SMP and a 64-bit OS. These attributes help deliver superior performance and scale compared to alternate v solutions. Alcatel-Lucent leverages over a decade of industry leadership and innovations in IP routing platforms, SR OS software, and network/service management to enable best-of-breed virtualized network functions and services. For more information about the Alcatel-Lucent, visit: www.alcatel-lucent.com/products/virtualized-service-router. Acronyms 5620 SAM Alcatel-Lucent 5620 Service Aware Manager 7750 SR Alcatel-Lucent 7750 Service Router NFV OS network functions virtualization Operating System AA Application Assurance PE provider edge API application programming interface RAM random access memory BGP Border Gateway Protocol RIB routing information base CLI command-line interface route reflector CPU central processing unit SDN software defined networking ETSI European Telecommunications Standards Institute HA High Availability ibgp internal BGP IT information technology K Kernel-based L2, L3 Layer 2, Layer 3 LAN local area network MPLS Multiprotocol Label Switching SMP SR SR OS vcpu VPN v - Symmetric Multi ing Service Router Alcatel-Lucent SR OS virtual CPU virtual machine virtual private network virtualized Alcatel-Lucent Virtualized Service Router Route Reflector 7
References 1. Alcatel-Lucent. Alcatel-Lucent Virtualized Service Router, Release 13. Data sheet. March 2015. http://resources.alcatel-lucent.com/?cid=182483 2. Alcatel-Lucent Virtualized Service Router Product page www.alcatel-lucent.com/products/virtualized-service-router 3. Alcatel-Lucent. Virtualization realized. IP Routing optimized. ebrochure. November 2014. http://resources.alcatel-lucent.com/asset/182575 4. Chiosi, Margaret, et al. Network Functions Virtualisation: An Introduction, Benefits, Enablers, Challenges & Call for Action. SDN and OpenFlow World Congress, Darmstadt-Germany. October 2012. http://portal.etsi.org/nfv/nfv_white_paper.pdf www.alcatel-lucent.com Alcatel, Lucent, Alcatel-Lucent and the Alcatel-Lucent logo are trademarks of Alcatel-Lucent. All other trademarks are the property of their respective owners. The information presented is subject to change without notice. Alcatel-Lucent assumes no responsibility for inaccuracies contained herein. Copyright 2015 Alcatel-Lucent. All rights reserved. PR1504010665EN (April)