Fujitsu, Ltd. commissioned The

T H E TOLLY G R O U P No. 202146 September 2002 Fujitsu, Ltd. R920/R980 IP Switching Node Performance Evaluation Test Summary Premise: Multi-gigabit routers designed for the backbone of the Internet must exhibit the capability of integrating cell- and packet-based traffic while delivering high throughput and low latency for IPv4 and IPv6 traffic flows. Such routers should also be capable of interoperability with existing routers for MultiProtocol Label Switching Virtual Private Networks (MPLS-VPN) while effectively implementing advanced features, such as QoS and MPLS-TE. Fujitsu, Ltd. commissioned The Tolly Group to evaluate its R920/R980 IP Switching Node, a high-performance router designed for use in Internet service provider backbone networks. The Tolly Group conducted a battery of tests ranging from IPv4 and IPv6 forwarding and latency measurements, quality of service (QoS), and broadband remote access server (BRAS) capacity and performance, to an evaluation of MPLS traffic engineering (MPLS-TE) capabilities demonstrating the 's high availability and load balancing capabilities as well as its MPLS-VPN interoperability with Cisco Systems and Juniper Networks routers. Engineers also evaluated the R920/R980's support for Virtual Private LAN Services (VPLS), which allows the connection of multiple sites in a single bridged domain over an MPLS network, thereby enabling individual sites to appear to be on the same LAN segment, regardless of their geographic location. Testing took place in September 2002. Test Highlights Delivers full wire-speed performance for 64-, 512- and 1,518-byte packets in IPv4, IPv6, and mixed IPv4 and IPv6 environments with eight Gigabit Ethernet ports Successfully demonstrates Virtual Private LAN Services (VPLS) Demonstrates MPLS-VPN interoperability with Cisco Systems and Juniper Networks routers Delivers Broadband Remote Access Server (BRAS) capacity of 32,000 simultaneous sessions on two BRAS modules Achieves Label Switched Path (LSP) fail-over within 1 second Percentage of theoretical maximum throughput Percentage of theoretical maximum throughput R920 Zero-Loss Throughput 8-port, Gigabit Ethernet as Reported by the Agilent RouterTester 10 8 4 2 10 10 10 10 10 10 10 10 10 64-512- 1,500-64- 512-1,500-64- 512-1,500- IPv4 IPv6 IPv4/v6 Frame size (bytes) and traffic type Source: The Tolly Group, September 2002 Figure 1 2002 The Tolly Group Page 1

Results IPv4 and IPv6 Forwarding and Latency Measurements Test results show that the R920, equipped with eight Gigabit Ethernet SC multimode fiber interfaces, forwarded 10 of line rate for 64-, 512- and 1,518-byte IP packets with zero loss ( 0.001%) and delivered average latency of 48.4, 71.2 and 133.4 microseconds for each packet size, respectively, at the zero-loss full-mesh rate for IPv4 traffic. When traffic was changed to IPv6, again the R920 forwarded 10 of line rate for 64-, 512- and 1,518-byte IP packets with zero loss. Average latency for the IPv6 traffic was recorded at 60.3, 77.2 and 138.6 microseconds for 64-, 512- and 1,518-byte packets, respectively. Finally a 50/50 mix of IPv4 and IPv6 traffic was benchmarked. The R920 forwarded 10 of line rate for 64-, 512- and 1,518-byte IP packets with zero loss and delivered average latency of 57.7, 76.3 and 137.7 microseconds for each packet size, respectively, at the zero-loss rate. (See Figure 1.) Gigabit Ethernet Fast Ethernet L2- VPN #1 L2- VPN #2 Layer 2 Switch Virtual Private LAN Services (VPLS) Tests demonstrated that the supports VPLS in an MPLS environment. This proves that the is capable of mapping VLANs in an MPLS network by offering secure support for multipoint-to-multipoint configurations across an MPLS Network. (See Figure 2.) VPLS Test Environment R920 Layer 2 Switch Juniper M20 R980 R920 Layer 2 Switch Source: The Tolly Group, September 2002 Figure 2 MPLS-VPN Interoperability Verification Test results demonstrated that the R920/R980 could interoperate successfully with Cisco Systems 7206 and Juniper Networks M5 and M20 routers in an MPLS-VPN environment based on RFC 2547bis. Specifically, testing was conducted with two MPLS-VPNs supported simultaneously. (See Figure 3.) L3-VPN #1 CE R980 Juniper M20 Juniper M5 Cisco 7206 CE CE CE L3-VPN #2 Source: The Tolly Group, September 2002 Figure 3 2002 The Tolly Group Page 2

Broadband Remote Access Server (BRAS) Capacity Since engineers were limited by the number of test tools available for this test, testing was conducted on two BRAS modules where the maximum number of sessions capable of being established was an aggregate of 32,000 sessions. Test results demonstrate that the R980 established 16,000 sessions per BRAS module, or an aggregate of 32,000 simultaneous sessions. BRAS Throughput Tests results demonstrated that the R980 forwarded 10 of line rate of 64-byte upstream and 72-byte downstream and 10 of line rate for 1,510-byte upstream and 1,518-byte downstream packets. (See Figure 4.) MPLS-TE LSP (Label Switched Path) Fail-over When engineers simulated an LSP failure by removing a cable, MPLS-TE fail-over to the backup LSP happened in 916 microseconds on average. Similarly, after a command from the command line was issued, fail-over to the backup LSP occurred in 452 microseconds on average. (See Figure 5.) MPLS-TE Load Balancing Testing demonstrated that the R920/R980 effectively enforced the desired load balancing distribution requested across two separate LSPs. (See Figure 6.) DiffServ EF QoS In Diffserv EF tests, engineers sought to determine the capability of the to give priority to a high priority traffic stream. Tests demonstrated that the forwarded 10 of the high priority traffic stream when two competing streams of traffic designated as 'high-priority" and "best-effort" created oversubscription on the egress port. (See Figure 7.) DiffServ AF QoS Diffserv AF tests demonstrated the 's capability to prioritize routed IP traffic into different service classes. Test results showed that the met the Class-of- Service requirement of the highest priority traffic, guaranteeing delivery of the offered load. For testing, three streams of data, each representing a Gold, Silver and Bronze traffic class, converged on the. The was configured for a ratio of 4:2:1 for Gold, Silver and Bronze traffic classes, respectively. Testing shows that the effectively enforced the Class-of-Services and suffered no variation from the designated ratio. This shows the router is capable of delivering bandwidth according to the assigned traffic classification ratio defined on the. (See Figure 8.) Analysis While service providers today handle IPv4 traffic almost exclusively, IPv6 looms on the horizon. It is critical that next generation and mixed networks provide the same level of performance as today's IPv4 networks. The demonstrated that wire-speed performance of eight Gigabit Ethernet interfaces could be expected regardless of the IP traffic mix. Fujitsu, Ltd. R920/R980 IP Switching Node Performance Evaluation Fujitsu, Ltd. R920/R980 IP Switching Node Product Specifications* Feature Switching capacity: 20Gbit/s (R920), 80G/160Gbit/s (R980 ) Expansion : Four 2.5 Gbits/s slot (R920), Fourteen 2.5 Gbits/s slot (R980), Eight 2.5 Gbits/s and six 10Gbits/s slot Redundancy: All hardware component, 1+1 APS (POS/ATM) Interface: POS: OC-3c/STM-1, OC-12c/STM-4, OC-48c/STM-16, OC-192/STM-64 ATM: OC-3c/STM-1, OC-12c/STM-4 Ethernet: 1000Base-SX/LX, 10/100Base-T, 10GBase-LR BRAS sessions: up to 192,000 per system Protocols: IPv4: Static, RIP/RIPv2, OSPFv2, BGP4, IGMP2, PIM-SM IPv6: Static, RIPng, OSPFv3, BGP4+ MPLS: LDP, RSVP-TE Traffic engineering: MPLS-TE (RSVP-TE) QoS: Diffserv (EF/AF/CS/BE), CR-LSP (TE tunnel), Diffserv over MPLS (E-LSP), VPN: BGP/MPLS-VPN, VLAN-VPN mapped MPLS Power: DC -48V (R920/R980), AC 100/200V (R920), AC 200V (R980) Power consumption: Max. of 1.200 watts (R920), Max. of 2,500 watts (R980) For more information contact: Fujitsu, Ltd., Global Business Group Solid Square East Tower 580 Horikawa-cho, Saiwai-ku, Kawasaki 212-0013 Japan Phone: 81.44.540.4072 Fax: 81.44.540.4132 URL: http://telecom.fujitsu.com *Vendor-supplied information not verified by The Tolly Group 2002 The Tolly Group Page 3

Fail-over time in milliseconds The Tolly Group Fujitsu, Ltd. R920/R980 Value-add services like VPLS are also on an upward growth path as service providers look for new revenue streams. The demonstrated the capability of providing such services. Furthermore, it is important to service providers that new equipment interoperate effectively with existing infrastructure. Not long ago, removal of existing infrastructure devices and implementation of new ones was commonplace procedure but today's economic conditions are not conducive to an overall forklift approach. Testing of the demonstrated multi-vendor interoperability with the Cisco Systems 7206 and the Juniper Networks M5 and M20 routers in an MPLS-VPN environment. Service providers need to support large numbers of simultaneous BRAS connections while maintaining high levels of performance. Testing of the successfully demonstrated the system's capability to support an aggregate of 32,000 sessions on two BRAS modules. The also demonstrated wire-speed performance with a zero-loss tolerance of 0.01% (actual loss occrred at 0.007%) with 64-byte upstream and 72-byte downstream frames and 1,510-byte upstream and 1,518-byte downstream frames. The traffic engineering features of MPLS switches also play a prominent role in service provider networks. High-reliability is a critical factor for carrier-class gear, along with reliable fail-over protection for LSPs and load balancing that ensure the capability of meeting service level agreements (SLAs). The successfully demonstrated LSP fail-over within one second while also demonstrating the capability to effectively load balance traffic in accordance with configured specifications. R980 BRAS Zero Loss ( 0.01%) Throughput Gigabit Ethernet as reported by Fujitsu NextStream Percentage of theoretical maximum 10 8 4 2 64-byte upstream 1,510-byte upstream Meeting SLAs is critical for service provider gear. With data networks viewing all packets as equal, implementation of DiffServ, a method for classifying traffic priority types, can be used for traffic prioritization to help meet SLAs. However, utilizing DiffServ in a packet is meaningless unless 10 10 10 10 72-byte downstream 1,518-byte downstream Source: The Tolly Group, September 2002 Figure 4 Fail-over time in milliseconds R980 MPLS-TE LSP Fail-over Time Gigabit Ethernet as reported by the Agilent RouterTester 1,000 750 500 250 0 the switching infrastructure can effectively support this type of prioritization. The demonstrated effective prioritization of high-priority tagged traffic while also meeting class-of-service requirements. 2002 The Tolly Group Page 4 916 Cable Pull LSP fail-over time 452 User Initiated via Command Line Source: The Tolly Group, September 2002 Figure 5

Test Configuration and Methodology R920/R980 MPLS-TE Load Balancing Gigabit Ethernet Interface as reported by the Agilent RouterTester The Tolly Group tested the following: a Fujitsu R920, software version E10V02L02C23, outfitted with either up to four Gigabit Ethernet SC fiber interfaces or three OC-48 interfaces and alternately substituting a 48 port Fast Ethernet blade for one Gigabit Ethernet blade; and a Fujitsu R980 IP Node, software version E10V02L02C23, outfitted with either up to eight three Gigabit Ethernet interfaces or eight OC-48 interfaces depending on the test scenario listed below. Percentage of throughput 10 8 4 2 75% 75% 75% 75% 25% 25% 25% 25% Path 1 Path 2 Path 1 Path 2 For performance tests, the was connected to an Agilent Technologies RouterTester equipped with multiple four-port Gigabit Ethernet E7904A modules. For BRAS capacity testing, the was connected to an Agilent Technologies QA Robot E7299B. All Agilent devices were controlled by Agilent software version 5.0.7.2.1. Due to limitations of available test tools, the BRAS performance was conducted with Fujitsu NextStream NXS1400, software version V04L01C00, which allowed generation of Layer 3 traffic with appropriate BRAS information. For tests of IPv4 and IPv6 forwarding and latency tests, the R920 was equipped with four two-port Gigabit Ethernet line cards connected to eight Agilent RouterTester modules. (See Figure 9.) For tests of MPLS-TE, the R980 and R920 were equipped with two Gigabit Ethernet line cards for MPLS-TE and four Gigabit Ethernet line cards for BRAS. Transmit For tests of Quality of Service, the R920 was equipped with three OC-48 line cards. Receive Traffic classification and weight Desired Actual Source: The Tolly Group, September 2002 Figure 6 Percentage of Throughput R920 Quality of Service DiffServ EF OC-48 Interface as reported by the Agilent RouterTester 10 8 4 2 67% 67% 10 10 Best Effort () Priority (1) Traffic type and bandwidth utilization offered Actual Desired Source: The Tolly Group, September 2002 Figure 7 For tests of zero-loss ( 0.001%) throughput and latency tests of IPv4, IPv6, and mix of IPv4 and 2002 The Tolly Group Page 5

IPv6 traffic, engineers sent packets to each interface on the test bed, with traffic destined for every other interface, creating a full mesh traffic pattern. RouterTester tests sent traffic flows consisting of either 64-, 512-, or 1,500-byte Ethernet frames. Zero-loss throughput tests were at 10 of line rate and decreased by 1% increments of line rate until zero loss was obtained. In Virtual Private LAN Services tests, engineers connected six Fujitsu PCs running Windows 2000 Professional operating system to a Layer 2/3 switch, either a Fujitsu A380, Cisco Systems Catalyst 3550 or Extreme Networks Summit 48i representing the customer edge equipment which in turn connected via Gigabit Ethernet to either one of two R920s or a R980. All three s fed into a Juniper Networks M20, running the JUNOS 5.4R1.4 operating system. (See Figure 2.) For tests of MPLS-VPN interoperability, engineers connected six Fujitsu PCs running Windows 2000 Professional operating system to a Layer 3 switch, either a Fujitsu R920, Cisco Systems 3600 or a Juniper Networks M5 representing the customer edge equipment which in turn connected via Gigabit Ethernet to either a R980, a Cisco Systems 7206, running a Cisco IOS12.2(11)T or a Juniper Networks, M20, running the JUNOS 5.4R1.4 operating system. (See Figure 3.) For maximum BRAS capacity tests, two Agilent QA Robots were attached, one to each of the two Gigabit Ethernet interfaces of the R980. Attached to the R980 was a R920 interconnected via a Gigabit Ethernet interface to a RADIUS server connected via Fast Ethernet Percentage of throughput R980 Quality of Service Diffserve AF OC-48 Interface as reported by the Agilent RouterTester 4 2 57% 57% to the R920. The QA Robot was configured to setup 16,000 sessions per BRAS module. Once established, the sessions were maintained for 10 minutes and then systematically torn down. (See Figure 10.) For MPLS-TE LSP protection tests, engineers configured the device 29% 29% 14% 14% Gold : 4 Silver : 2 Bronze : 1 Traffic classification and weight Desired Actual Source: The Tolly Group, September 2002 Figure 8 G e o S t r e a m R 9 2 0 4 Gigabit Ethernet interface modules (2 interfaces per module) Source: The Tolly Group, September 2002 Figure 9 under test with redundant Gigabit Ethernet paths. Engineers then simulated an LSP fail-over by either removing a cable or by executing a command at the device interface that rerouted the traffic across the backup LSP. Agilent's RouterTester recorded the number of lost packets during fail-over. If a packet is transferred at a known rate, then 2002 The Tolly Group Page 6

Upstream R980 R920 NXS1400 NXS1400 Downstream Radius Server Source: The Tolly Group, September 2002 Figure 10 the time during which the packet cannot be passed through the device under test can be estimated based on the number of lost packets during fail-over. (See Figure 11.) For MPLS-TE load balancing tests, engineers configured a R920 and R980 in a back-to-back configuration, each with a pair of Gigabit Ethernet paths interconnecting the two devices. The s were configured to offer 75% of the traffic on one of the paths and 25% of the traffic on the other path. Agilent's RouterTester forwarded line rate traffic and counters on the s confirmed the 75%/25% balance. For Quality of Service Diffserv EF (prioritization) tests, two OC-48 interfaces were configured as ingress ports on the R920 and a single OC-48 as an egress port. Agilent's RouterTester forwarded 1 of the theoretical maximum of each link across two of the ingress links. These traffic streams were configured to offer priority traffic. The remaining two ingress links were configured for of the theoretical maximum of each link. These links were configured for "best effort" traffic. Two ingress links were destined for a single egress link offering a 4 over subscription scenario. Agilent's RouterTester captured a per-link analysis of the received traffic and priority traffic was confirmed. (See Figure 12) For Quality of Service Diffserv AF tests, engineers configured the R980 to provide queues of Gold, Silver and Bronze levels of service and to assign weighted round-robin ratios of 4:2:1 for the three queues. To create congestion, engineers configured two OC-48 links, each offering 99% of theoretical maximum traffic streams consisting of 33% of each of the three traffic types. Classification of packets into service levels was performed on ingress. R980 Path 1 P a t h 1 R920 Path 2 R920 Source: The Tolly Group, September 2002 Figure 11 2002 The Tolly Group Page 7

BE Priority 1 OC-48c POS BE Priority R920 BE Priority 8 2 OC-48c POS 1 Priority: Ingress side 0.2 Gbit traffic (1 of 1 Gbit traffic x2) Egress side 0.2 Gbit traffic (2 of 1 Gbit traffic) Best Effort: Ingress side 1.2 Gbit traffic ( of 1 Gbit traffic x2) Egress side 0.8 Gbit traffic (8 of 1 Gbit traffic) Source: The Tolly Group, September 2002 Figure 12 The Tolly Group gratefully acknowledges the providers of test equipment used in this project. Vendor Product Web address Agilent Technologies Router Tester http://www.agilent.com/routertester/ Agilent Technologies QARobot http://www.agilent.com Fujitsu Kyusyu Digital Technology Ltd. NextStream1400 http://www.qdt.fujitsu.com Tolly Group Services With more than a decade of testing experience of leading-edge network technologies, The Tolly Group employs time-proven test methodologies and fair testing principles to benchmark products and services with the highest degree of accuracy. Plus, unlike narrowly focused testing shops, The Tolly Group combines its vast technology knowledge with focused marketing services to help clients better position product benchmarks for maximum exposure. The company offers an unparalleled array of reports and services including: Test Summaries, Tolly Verifieds, performance certification programs, educational Webcasts, white paper production, proof-ofconcept testing, network planning, industry studies, end-user services, strategic consulting and integrated marketing services. Learn more about The Tolly Group services by calling (732) 528-3300, or send E-mail to info@tolly.com. For info on the Fair Testing Charter, visit: www.tolly.com/about/ftc.asp Project Profile Sponsor: Fujitsu, Ltd. Document number: 202146 Product Class: Carrier-class IP Router Products under test: Fujitsu Ltd. R920/R980 Testing window: September 2002 Software versions tested: E10V02L02C23 Software status: Generally Available For more information on this document, or other services offered by The Tolly Group, visit our World Wide Web site at http://www.tolly.com, send E-mail to info@tolly.com, call (800) 933-1699 or (732) 528-3300. Internetworking technology is an area of rapid growth and constant change. The Tolly Group conducts engineering-caliber testing in an effort to provide the internetworking industry with valuable information on current products and technology. While great care is taken to assure utmost accuracy, mistakes can occur. In no event shall The Tolly Group be liable for damages of any kind including direct, indirect, special, incidental, and consequential damages which may result from the use of information contained in this document. All trademarks are the property of their respective owners. The Tolly Group doc. 202146 rev. kco 27 Nov 02 2002 The Tolly Group Page 8