What is the biggest, fastest LAN switch?This is actually a problem with many complex answers, but the answer depends on how you define "big" and "fast."
3Com, Extreme, Force10, Foundry, HP ProCurve and Cisco Ethernet switch vendor for whom the highest performance, maximum density and minimum delay and feuding.However, the current products offered by these vendors technologies have on average 3 years old.Meanwhile, Woven Systems, and Raptor Networks such as a pack of hungry new companies is toward how to build the "biggest" Ethernet switch to launch new challenges.They take a different way than a single large chassis, rather more like a cluster-type super-computing technology or InfiniBand network topology.
Performance is reflected in many ways
Ethernet 802.3ae standard by the current speed (speed of 10Gbps) limit, so you can guess, there is no faster than an independent Gigabit ports.Some other measure of performance of the switch is based on switch fabric bandwidth, and port density chassis supports.In addition, the port's own performance, delay and jitter performance are the elements of the switch.
Network Test president, "Network World" Test Alliance member David Newman said: "Everyone said wire-speed per port. So, the question now becomes in the event of packet loss in the number of months before you can achieve wire-speed ports. "According to the published equipment specifications, the largest enterprise core switches, including Force10's E1200, Foundry's RX series, Cisco Catalyst 6500 and Extreme's BlackDiamond.When comparing these specifications, Foundry's RX-16 switch is the highest-capacity switch that can run at full speed 64 Gigabit Ethernet ports, the excess allocation (that is, the total bandwidth of all ports on the switch capacity of more than) of a chassis up to 192 Gigabit ports total.Force10's E1200 TeraScale switch can run 56 Gigabit ports, or 224 excess configuration Gigabit ports.Extreme's BlackDiamond 10808 chassis can support 48 non-blocking Gigabit ports.The Cisco Catalyst 6513 can handle full-duplex operation of 32 Gigabit Ethernet connectivity.
Some people think that with the vendors for each slot and how to divide the total system bandwidth compared to how the switch is running full speed when variables such as jitter and packet loss rate is not very important.Newman said: "I think the more useful indicator than the throughput is latency. In this context, Cisco is clearly the best." He has been measured a Cisco Catalyst 4948 devices, the delay in the Gigabit speed is about3 microseconds.Newman said: "This is the equipment I have had the lowest measured latency. Force10's delay is usually high in comparison to Cisco Jiaohuan Ji hundreds of times, this means that thousands of data packets not addressed. But they're also solved the problem."
U.S. Department of Energy, based in the city of Berkeley, California research lab-Lawrence Berkeley National Laboratory (LBNL) in the data center and LAN core at the same time using the Force10 and Cisco switches.Laboratory LBLnet Services, said Mike Bennett, senior network engineer, specify which of the two products a "faster" or "better performance" very difficult, because they switch among different applications in.Bennett said: "I tested the configuration of two Gigabit ports 6500 and E1200 switch, they have no overload, and no packet loss. Thus, not simply a kind of switch faster than the other, theyshould be stated in the advertisement, the effective functioning of 才行. "
Architecture of next-generation
The other side of the Berkeley City Lawrence Livermore National Laboratory, the network is currently studying to achieve the storage area network Ethernet bandwidth and latency in the level of next-generation switch architecture.
Head of the laboratory network Dave Wiltzius said: "The next generation of equipment will enhance all the performance an order of magnitude, everything will be Gigabit. So we can provide more than 2000 to find a switch port or Gigabitswitch fabric. in fact we use the fat tree topology, and something like this fusion of the switch environment to create interest. "
Wiltzius envisaged fat tree topology involves a Mesh (mesh network) non-blocking switching architecture.This architecture a bit like the traditional public telephone network, which switches are connected together through multiple paths, and a number of ports with simple equipment.Important is that they can effectively use the bandwidth.The Ethernet can not do, or not very well done.
Wiltzius has been gathered in the use of port or two-story "hash" technology to achieve this effect the fat tree, that is, the number of Gigabit or Gigabit Ethernet links bundled in a larger virtual pipe.Will use the hash of the Ethernet switch connected to piping (or server connected to the switch), can lead to greater throughput virtual, but this connection is limited to 8 ports (that is, the hash of 8Gigabit link provides up to 80Gbps throughput).This method is used to connect the bundled transmission of data packets on the random algorithm.As the randomness of the algorithm using hashing technology, users will get a kind of uneven distribution, not necessarily provide the best bandwidth utilization.
Breakthrough data center switch
Arastra and Woven Systems is actively seeking to achieve a breakthrough switching technology emerging companies.Arastra is a privately held Palo Alto, a router company.Woven Systems is the Santa Clara City, a state of semi-secret development of Ethernet-based Mesh Network products.
Woven President and CEO Harry Quackenboss said: "We are doing the structure in the Gigabit Ethernet to provide the best characteristics of Fibre Channel or InfiniBand." Woven approach similar to the grid or distributed computing cluster.In this technique, symmetric multi-processor (SMP) server is connected through the network of single / dual-processor node replaced.
Quackenboss said: "The same thing will occur in the data center's local area network switching technologies. Major LAN switch is expensive, you can buy for data center applications the largest non-blocking 64-port switch only Foundry system."
Woven is developing a two-story center Gigabit and Gigabit Ethernet data switch, which switch to use special algorithms to simulate some way InfiniBand or Fibre Channel network.In this structure, the switch can be established between the multiple paths.As the transfer flow passage being closed (similar to spanning tree-based Ethernet), bandwidth can be more dynamically allocated to these paths.
Quackenboss said: "If you want to build a network consisting of two or more switches, you can use link aggregation or trunking technology will be bundled with the Ethernet segment. But if you want to put three or more switchesin the network, one of the switches will become a bottleneck. "exchange of technology and three-OSPF (Open Shortest Path First) and ECMP (equivalent to multi-path routing) and other protocols, can be used to create multi-path network, but these methodsalso increased costs.According to IDC's survey, the average cost of three-switch port is the port's five times the second floor.
In addition, the data center Ethernet connection into a two-story multi-path, can help managers to dynamically re-configure the application and the server without having to actually recall them.This idea somewhat similar to the carrier-class equipment manufacturers are developing the second floor of Metro Ethernet technology.
Large virtual switch using Mesh
Another company has launched the product has a three-year history of Raptor Networks, the company connected with the production structure of the composition of Mesh and Gigabit Ethernet fixed configuration Gigabit switches.Raptor does not focus on high-density data centers, but rather for the local backbone and aggregation of wiring between the transport stream.Company CEO Tom Wittenschlager said: "We created a special feature on the second floor of the ability of these features only three other companies to provide."
Raptor switch equipped with 24 Gigabit ports and six Gigabit Ethernet ports, with 160Gbps of total switching bandwidth.This single-chassis switches support the exchange of two to four, and uses a proprietary modified version of the Gigabit Ethernet technology.This technology allows Ethernet devices without the use of spanning tree protocol, you can connect together to form the two-story Multi-Path Mesh Networks.
These switches using Raptor Adaptive Switch Technology (RAST) link.RAST is a specific way to connect the switch protocol, this approach is similar to the chassis switches in the module connected to the backplane or switch structure.
Wittenschlager cited the company's test data: Raptor by 4 sets of the composition of the Mesh network switch will be a data packet input and output 8 Gigabit Ethernet ports, the results obtained for the 6.48 microseconds propagation delay.He said: "Every Raptor switch blade like a module as to enable transport streams with low latency between switches very fast delivery. To do this, Raptor will be routing information into the standard Ethernet frameunused header space, the heart rate and routing information makes the switch the data path between switches in the cluster were passed back and forth. "
In addition, Raptor has developed a physically separate switch back on in the public communications capabilities, as if they were all on the same chassis the same.This is actually a virtual switch, when the use of single-mode optical fiber on the long-distance Gigabit Ethernet connection, switch the distance between each other up to 80 kilometers.Wittenschlager explained that connect Gigabit ports on the Gigabit or non-Raptor switches, then it is seen as a major independent LAN switches, no additional configuration can be done as a simple Ethernet connection.
Recently, a switch from the 4 sets of Raptor Mesh network consisting of replacing the LA Care network in the company of two Catalyst 6509 switches form the core network.The Raptor switches are used to separate the company's flat layer LAN subnet as the VLAN, while the core network, Gigabit Ethernet, three Gigabit Ethernet channels connected Mesh network each device.The Catalyst switch was transferred to the LAN edge, used to connect the company's 350 end-users and other equipment.
LA Care, IT and security managers Rayne Johnson said: "In the addresses of the Cisco IOS-based routers and switches based on RAST of Raptor produced some of the spanning tree between the original circuit problem, the network has become both smooth and fast run." In addition,Raptor product installation costs at 18 million dollars, and Cisco offer to Johnson, 50 million, for the adoption of Gigabit Ethernet and VLAN capabilities to upgrade the core network.Johnson said, "I usually just try not to get involved in products developed by them. This in part is worth the price because you can not resist the temptation."