/
Go with the flow
Subscribe to the High-Speed LANs newsletter
This time, we grounded the study in basically the same test suite to broaden the base of vendors and increase participation in optional tests. The final lineup included switches from six vendors, with newcomers Alcatel, Extreme Networks and NBase-Xyplex joining returning vendors Cisco, Enterasys and Foundry Networks. The returning vendors brought either upgraded gear or entirely different products. Last year, almost half of the dozen products tested were Layer 2-only devices. This year, the devices were Layer 2 and Layer 3 with the exception of the NBase-Xyplex switch. Four of last year's participants did not return. IBM has gotten out of the switching business. Hewlett-Packard, Avaya and Nortel Networks declined, saying interoperability issues with their offerings have changed little since Round 1. Among the prominent switch vendors that have missed the first two rounds are 3Com, Alteon, Intel, Marconi and Network Peripherals. Intel planned to participate in this round but withdrew just as the test began, citing lack of resources to support the test. The bottom line on this round: Thumbs up virtually across the board, because all vendors proved interoperability in our mandatory tests. Additionally, participation in the more advanced "optional" tests increased dramatically -- a good sign for network managers looking for deeper levels of interoperability from their network gear. Mandatory tests are those considered absolutely essential in enterprise networks: autonegotiation, IP routing, Routing Information Protocol (RIP) 1.0 and RIP 2.0 interoperability. When disparate switches can work together on these levels, network managers are afforded more choices when designing and upgrading their infrastructures. The optional tests stress important functions indeed, but ones that might not be necessary on every network. These tests included link aggregation, accelerated spanning tree reconvergence, Gigabit Ethernet uplink support, IPX routing and IP Virtual Router Redundancy Protocol (VRRP). Only one test conducted last year, Layer 2 flow control, was dropped from this year's test suite. The suite of tools used to carry out these tests comprised Netcom Systems' Advanced Multiport Performance Tester/Analyzer/Simulator, SmartBits' SMB-200 test tool, Agilent's Internet Advisor, NetIQ's (formerly Ganymede Software) Chariot, and Wavetek Wandel Goltermann's Domino Fast Ethernet Internetworking Analyzer DA-350. While the majority of last year's participants were reluctant to venture beyond the mandatory tests, all of the Round 2 participants took on these more challenging tests. Some even ventured into nonstandardized areas where their support for accelerated convergence for spanning tree and VRRP were scrutinized. Foundry stepped up to the plate for both tests, interoperating with Cisco during the accelerated spanning tree test and with both of Enterasys' switches in the VRRP tests. The only vendor gamesmanship occurred when Alcatel challenged Cisco to test interoperability between Cisco's Catalyst running Cisco's proprietary Hot Standby Routing Protocol (HSRP) and Alcatel's reverse-engineered version of the same protocol. Cisco did not want to run this test, declining to explain its reasoning. Cisco escaped on a technicality -- our test plan specified industry-standard VRRP, not HSRP, as the test architecture. The Tolly Group's test methodology could easily have been used with a test of HSRP had Cisco elected to proceed. The obvious conclusion is that Cisco doesn't want to help Alcatel prove it can interoperate with Cisco's proprietary protocol. Score another point for marketing prevailing over engineering. Still, it is unlikely that Alcatel would have made the challenge if it was not able to back it up. Overall, the growing level of support and participation is a good sign that vendors are committed to supporting key standards. Consumers are getting peace of mind that the choices they make today will not adversely affect their networks in the future. Network managers can buy with confidence, knowing vendors can deliver on their interoperability claims. Check out the results.
Methodology: The methodology describes configuration for static routes; however, a pair of switches that demonstrates IP routing interoperability via RIP Version 1.0 or 2.0 has already demonstrated IP routing interoperability. Because all of the vendors successfully transferred TCP/IP traffic during tests of RIP 1.0 and 2.0, there was no further testing was necessary to verify IP routing.
Degree of complexity: Low.
Participants: All Layer 3 switches.
Results: All switches were interoperable.
Tolly analysis: This type of interoperability lets you span static entries from one subnet to another when you are trying to connect a multisubnet environment. This gives the same effect as RIP 1.0 and 2.0 but is more labor-intensive because it involves actual manual route entries.
Test 7: IPX routing (optional)
Description: Each switch pair must properly route IPX traffic by static route entries or RIP.
Methodology: The participants configured their switches to connect one interface to a common
network and one interface to a private network. Switches were configured for static IPX routes or IPX RIP, as required, and supported by each pair of switches. Two NetIQ Chariot endstations were configured to run a one-minute test. Note: All vendors that were tested were configured for IPX RIP.
Degree of complexity: Low.
Participants: All Layer 3 switches.
Results: All switches were interoperable.
All vendors were asked to configure their switches for IP routing, and all vendors chose to configure their switches dynamically. IPX RIP was originally removed from this year's lineup of optional tests because of a rapidly dwindling interest in IPX as a protocol of choice. Regardless, vendors chose IPX RIP configurations demonstrating their further support for this protocol.
Tolly analysis: IPX routing has the same benefits as RIP 1.0 and 2.0, but instead of using IP, it uses IPX protocol. IPX routing traditionally comes into play in networks running Novell NetWare operating systems that require this type of interoperability.
Test 8: VRRP (optional)
Description: Each IP router pair must demonstrate that a back-up router will take over if a primary device fails, in accordance with the protocol.
Methodology: Participants alternately configured one port as a master on one network and a slave on another network, and connected to a hub on each subnet. Endstations were attached to each of the two networks. The Tolly Group engineers performed a continuous series of pings between the endstations. Then engineers powered off the master router/ switch and recorded the number of pings lost until the network reconverged around the failure and across the slave router/switch. Participants reversed the roles of master and slave and The Tolly Group repeated the test.
Degree of complexity: Moderate.
Participants: Foundry's BigIron 4000 and Enterasys' Matrix E7 and SSR 2000.
Results: All switches tested were interoperable.
If a switch successfully reconverged, a pass was recorded. Unsuccessful reconvergence would have resulted in a failure. Cisco did not participate in the VRRP test because the Catalyst 6500 supports the company's proprietary HSRP. Alcatel also claimed to support HSRP and offered to test its OmniCore 5022 with the Cisco Catalyst 6500; however, Cisco chose not to test with Alcatel, as we explained above.
Tolly analysis: VRRP lets managers configure two devices to work in concert to back up one another in the event of a failure. VRRP differs from accelerated spanning tree because it has two active paths, both forwarding data, and if one fails, the other switch accepts the additional data load. In accelerated spanning tree, one path is inactive until device failure and then comes alive via the BPDUs.
LAN switches: living in an heterogeneous World
Round 2 of the Tolly Group/Network World
 
|
|||
 | |||
|
Go with the flow
Subscribe to the High-Speed LANs newsletter
This time, we grounded the study in basically the same test suite to broaden the base of vendors and increase participation in optional tests. The final lineup included switches from six vendors, with newcomers Alcatel, Extreme Networks and NBase-Xyplex joining returning vendors Cisco, Enterasys and Foundry Networks. The returning vendors brought either upgraded gear or entirely different products. Last year, almost half of the dozen products tested were Layer 2-only devices. This year, the devices were Layer 2 and Layer 3 with the exception of the NBase-Xyplex switch. Four of last year's participants did not return. IBM has gotten out of the switching business. Hewlett-Packard, Avaya and Nortel Networks declined, saying interoperability issues with their offerings have changed little since Round 1. Among the prominent switch vendors that have missed the first two rounds are 3Com, Alteon, Intel, Marconi and Network Peripherals. Intel planned to participate in this round but withdrew just as the test began, citing lack of resources to support the test. The bottom line on this round: Thumbs up virtually across the board, because all vendors proved interoperability in our mandatory tests. Additionally, participation in the more advanced "optional" tests increased dramatically -- a good sign for network managers looking for deeper levels of interoperability from their network gear. Mandatory tests are those considered absolutely essential in enterprise networks: autonegotiation, IP routing, Routing Information Protocol (RIP) 1.0 and RIP 2.0 interoperability. When disparate switches can work together on these levels, network managers are afforded more choices when designing and upgrading their infrastructures. The optional tests stress important functions indeed, but ones that might not be necessary on every network. These tests included link aggregation, accelerated spanning tree reconvergence, Gigabit Ethernet uplink support, IPX routing and IP Virtual Router Redundancy Protocol (VRRP). Only one test conducted last year, Layer 2 flow control, was dropped from this year's test suite. The suite of tools used to carry out these tests comprised Netcom Systems' Advanced Multiport Performance Tester/Analyzer/Simulator, SmartBits' SMB-200 test tool, Agilent's Internet Advisor, NetIQ's (formerly Ganymede Software) Chariot, and Wavetek Wandel Goltermann's Domino Fast Ethernet Internetworking Analyzer DA-350. While the majority of last year's participants were reluctant to venture beyond the mandatory tests, all of the Round 2 participants took on these more challenging tests. Some even ventured into nonstandardized areas where their support for accelerated convergence for spanning tree and VRRP were scrutinized. Foundry stepped up to the plate for both tests, interoperating with Cisco during the accelerated spanning tree test and with both of Enterasys' switches in the VRRP tests. The only vendor gamesmanship occurred when Alcatel challenged Cisco to test interoperability between Cisco's Catalyst running Cisco's proprietary Hot Standby Routing Protocol (HSRP) and Alcatel's reverse-engineered version of the same protocol. Cisco did not want to run this test, declining to explain its reasoning. Cisco escaped on a technicality -- our test plan specified industry-standard VRRP, not HSRP, as the test architecture. The Tolly Group's test methodology could easily have been used with a test of HSRP had Cisco elected to proceed. The obvious conclusion is that Cisco doesn't want to help Alcatel prove it can interoperate with Cisco's proprietary protocol. Score another point for marketing prevailing over engineering. Still, it is unlikely that Alcatel would have made the challenge if it was not able to back it up. Overall, the growing level of support and participation is a good sign that vendors are committed to supporting key standards. Consumers are getting peace of mind that the choices they make today will not adversely affect their networks in the future. Network managers can buy with confidence, knowing vendors can deliver on their interoperability claims. Check out the results.
Layer 2
Test 1: Port autonegotiation (mandatory) Description: Each pair of switches must properly negotiate the highest Ethernet speed (10/100) and appropriate duplex (half/full) settings common to both devices. Methodology: Ports were configured to autonegotiate speed and duplex support. The two switches were connected via a crossover cable. If both reported configurations were identical, Tolly engineers generated a stream of 1,518-byte frames for one minute at a rate appropriate for the reported configuration. Note: If the switches reported 10M bit/sec full duplex, engineers generated more than 10M bit/sec in each direction to confirm the speed setting. Degree of complexity: Low. Participants: Alcatel's OmniCore 5022; Cisco's Catalyst 6500; Enterasys' Matrix E7 and SSR 2000; Extreme Networks' Alpine 3808; Foundry's BigIron 4000; and NBase-Xyplex's OptiSwitch 800. Results: All switches were interoperable. To pass the test, a switch had to indicate that it passed traffic at the highest possible link speed and duplex level, and the switch pair had to forward traffic at more than the next-lower speed/duplex setting. For instance, to pass at full-duplex Fast Ethernet, the switch pair would have to forward more than 100M bit/sec of traffic total across the link. A fail would be recorded if a switch indicated any link speed and duplex other than the highest possible supported or if the switch pair failed to forward the throughput indicated above. Tolly analysis: This test was a winner for all. Autonegotiation lets two devices decide their maximum speed and duplex without having a network manager hard-code each switch's adapters and ports. This is important because improper link speed creates a no-link situation and improper duplex settings can severely affect throughput. If one switch is configured for full duplex and another switch is configured for half duplex, the devices cannot transmit and receive data without errors. The advantage with autonegotiation is, if it works properly, it reduces the need for a network manager to negotiate link speed and duplex settings on each device in the network. Test 2: Fast Ethernet link aggregation (optional) Description: Link aggregation allows for two (or more) physical links to function as a single logical link. Each switch pair must communicate over an aggregated link consisting of two full-duplex, Fast Ethernet links. Switches were tested to comply with the IEEE 802.3ad standard for link aggregation. Methodology: The two switches were configured for link aggregation across two Fast Ethernet full-duplex ports. We generated bidirectional traffic streams consisting of 1,518-byte packets at a load exceeding the bandwidth of a single port, and validated successful receipt of traffic greater than a single, full-duplex link could forward. Degree of complexity: Moderate. Participants: Alcatel's OmniCore 5022; Cisco's Catalyst 6500; Enterasys' Matrix E7 and SSR 2000; Extreme Networks' Alpine 3808; Foundry's BigIron 4000; and NBase-Xyplex's OptiSwitch 800. Results: All switches were interoperable. A switch passed if its aggregated links successfully transmitted traffic greater than one full-duplex link. A switch would have failed if the aggregated links did not transmit traffic greater than a single full-duplex link. Tolly analysis: This optional test was also popular in last year's study, which all vendors also passed. Link aggregation gives network managers a cost-effective way to leverage their existing network infrastructure to allow for fractional bandwidth between Fast Ethernet and Gigabit Ethernet. Test 3: Gigabit Ethernet uplink (optional) Description: Each pair of switches must pass traffic across a single, full-duplex Gigabit Ethernet link between the devices. Methodology: The participants configured a Gigabit Ethernet (1000Base-SX) link between their switches for Layer 2 forwarding. Data was generated between disjointed Fast Ethernet LANs across the Gigabit Ethernet uplink to validate connectivity. Degree of complexity: Low. Participants: Alcatel's OmniCore 5022; Cisco's Catalyst 6500; Enterasys' Matrix E7 and SSR 2000; Extreme Networks' Alpine 3808; Foundry's BigIron 4000; and NBase-Xyplex's OptiSwitch 800. Results: All switches interoperated. There were no problems linking participating switches via a single Gigabit Ethernet uplink and passing a stream of 1,518-byte packets from one device to the next. Tolly analysis: As workgroup switched networks converge with backbone campus networks, it is not uncommon to need this type of Gigabit link between switches from other vendors. Test 4: Accelerated convergence for spanning tree (optional) Description: Each switch pair must demonstrate the ability to reconverge its spanning tree configuration and begin passing traffic less than 10 seconds after a device failure using Bridge Protocol Data Units (BPDU). Methodology: We initiated a sequence of pings at one-second intervals and validated that the request and response crossed the same path. We powered off the switch on the path and monitored the number of sequential pings lost before the network reconverged, and the pings achieved success. The number of pings lost represents the time required for the network to reconverge using BPDUs. Degree of complexity: High. Participants: Cisco's Catalyst 6500 and Foundry's BigIron 4000. Results: The two participants were interoperable. Switches that reconverged the network in less than 10 seconds passed the test. A fail result would have registered when a reconvergence of the network took longer than 10 seconds. Tolly analysis: When designing a network that has built-in redundancy, you need to ensure that if a device fails, the redundant link can recover and reconverge using BDPUs within the specified time frame. Most of the vendors that did not participate in this part of the test don't yet support this function. Others, such as Enterasys, say they are awaiting the IEEE 802.1w standard for rapid reconfiguration of spanning tree to be finalized. Extreme and Alcatel say their switches will support accelerated convergence for spanning tree in upcoming releases.Layer 3
Test 5: IP RIP Version 1.0 and 2.0 (mandatory) Description: Each switch pair must exchange IP routing table information in compliance with the Internet Engineering Task Force IP RIP for Versions 1.0 (RFC-1058) and 2.0 (RFC-1723). RIP 1.0 Methodology: Two switches were configured for RIP 1.0 and no static routes. Engineers validated that no static route entries appeared, and that routing tables were empty at the beginning of the test. After interconnecting, we validated successful forwarding of traffic and confirmed that each switch showed appropriate routing information in its IP routing table. RIP 2.0 Methodology: Same as RIP 1.0. Degree of complexity: Low. Participants: All Layer 3 switches. Results: All switches were interoperable. Tolly analysis: Having a multivendor environment that can successfully exchange RIP 1.0 and 2.0 routing information removes the burden of a network manager having to hard-code static route table entries into each switch. Test 6: IP routing (mandatory) Description: Each switch pair must properly route TCP/IP traffic from static route table entries.
|
The future . . .
Most participants voiced an interest in multivendor quality-of-service (QoS) tests for the next round, such as those that take a closer look at how vendors implement standards such as 802.1p, 802.1q, Differentiated Services, and type of service. Testing QoS will be interesting, considering that the standards discuss how data is tagged but do not outline how they are implemented. It will be interesting to see if the differing implementation methods can work together. Alcatel representatives suggest routing functions such as Open Shortest Path First (OSPF), Border Gateway Protocol 4 and multicasting. Mike Wong, technical marketing manager for Extreme, agrees. "For the next round, we would like to see OSPF and Multicast PIM-DM and PIM-SM or Protocol Independent Multicast (Dense Mode and Sparse Mode)," he says. Multicasting is becoming more important, and testing this feature with multiple devices will be interesting. OSPF is found in more complex networks and it is also important to find out if this leading-edge feature can interoperate with a variety of vendors. As networking becomes increasingly pervasive, "green field" sites may cease to exist -- there will always be legacy gear in the picture. In such a world, multivendor interoperability will become mandatory. As we move toward a "data utility" concept that will converge data, voice and video in the corporation, more sophisticated feature sets -- some that are even beyond what the standards committees control -- will appear on LAN switches. High levels of interoperability will need to be maintained for all vendors to have a fighting chance in the competitive enterprise marketplace. Related linksKevin Tolly is president and CEO, Bradley is managing editor and Brian Tolly is an engineer at The Tolly Group, a strategic consulting and independent testing firm in Manasquan, N.J. Tolly was the project manager for this study. They can be reached at ktolly@tolly.com, hbradley@tolly.com and btolly@tolly.com, respectively.
The Tolly Group's SwitchCentral
