【 Weak current College 】 understanding Gigabit Ethernet and routing switch technology

1, Gigabit Ethernet technology
1.1 early Ethernet technology

Ethernet: IEEE802.3 defines 10Mbps Ethernet standards, the use of carrier listens and collision detection (CSMA/CD) Protocol to half-duplex mode. Starting from the late 80 's Ethernet was a huge success. 10BaseT is running on a class 3 or a higher class of twisted-pair Ethernet 10Base2/5, is running in the Ethernet coaxial cable, optical fiber 10BaseFL is running in Ethernet. As a result of the conflict detection Protocol requires a 512-bit time grooves ensure no errors detected in the conflict, so Ethernet distance coverage was limited, the largest covers 10BaseFL distance 2km, 10BaseT network segments at a maximum distance covered as 100m.

Fast Ethernet: IEEE802.3u defines a 100Mbps Fast Ethernet standard, which you can use half-duplex mode run the CSMA/CD protocol, can also have full-duplex mode. Due to the fast Ethernet to Ethernet for backward compatibility, the dynasty of the 1990s, fast Ethernet becoming mainstream technology in LAN. 100BaseTX is run on 5 twisted-pair fast Ethernet, is running in fiber 100BaseFX fast Ethernet. For half-duplex operation of fast Ethernet, there are also restrictions on range of coverage, and due to the fast Ethernet to the 100Mbps speed run, time slot length is 512 bits, so it's the greatest distance covered is Ethernet 1/10, to 200m. But for full duplex fast Ethernet runs, in theory, you no longer have the distance limit, but actually limited by electrical or optical signal attenuation. If the actual run in single-mode fiber at 100BasFXSMF full duplex fast Ethernet maximum distance covered up to 20km above.

1.2 Gigabit Ethernet protocol

In June 1998 in Gigabit Ethernet Alliance of IEEE officially released by IEEE802.3 standard Gigabit Ethernet. The Ethernet speed increase to 1000MbPs. And in 1997, it has been a lot of manufacturers can't wait to launch a Gigabit Ethernet products, network world has brought a completely new solution.Up to now in 2000, we have been able to clearly see not only Ethernet and fast Ethernet to the desktop and workgroup network beat ATM, in man, Gigabit Ethernet is also due to their high level of compatibility and superior price/performance of the absolute. You can see into the future with prices, Gigabit Ethernet is the same as fast Ethernet.

1.2.1 half-duplex Gigabit Ethernet MAC layer protocol

For fast Ethernet, 512-bit time slot can transmit radio or optical 400m far, Gigabit Ethernet, 512-bit time tank radio or optical transmission distance only 40m far, uses a star topology of half-duplex Gigabit Ethernet coverage only 20m. Such distance coverage in reality cannot be massive. To solve this problem, IEEE on Ethernet MAC layer protocol for its first major modifications: carrier expands and frame burst.

(1) carrier extended

In order to enable Gigabit Ethernet distance coverage to achieve practical standards, half-duplex Gigabit Ethernet slot length extended to 4096 bit, half-duplex Gigabit Ethernet distance coverage expanded to 160m. In order to compatible Ethernet and fast Ethernet frame structure, half-duplex Gigabit Ethernet minimum frame length still need to remain as 64byte. But considering the time slot length 51byte, in order to be able to match the length of time worse, when a DTE sent less than 512byte frame, half-duplex Gigabit Ethernet MAC will be sent after the normal sending data to a carrier extended sequence until the end of a fine. For example: a DTE sends a 64byte frames, MAC will then join the 512-64 = 448byte carrier extended sequences. If you send a frame DTE has a length greater than 512byte, the MAC does not make any changes.

In the case of carrier extension solves half-duplex Gigabit Ethernet distance coverage of issues, but introduces a new problem: for length smaller Ethernet frames sent efficiency reduces. The frame for a 64byte, although sent faster than Fast Ethernet increased 10 times, but the send time increased by eight times. This efficiency has not increased over fast Ethernet, in order to solve half-duplex Gigabit Ethernet efficiency problem, IEEE also introduces frame burst of such technology.

(2) frame burst

Frame burst works as follows: for the first DTE sent is smaller than the frame, still 512byte use carrier to 512byte, but subsequently sent less than 512byte short frames are no longer using carrier expands, but joined the frame interval sequence 96bit continuous send a text frame, you can burst up to 65536. This approach can be established because a properly configured network environment, if a DTE start sending data, other DTE are available through the carrier monitoring protocols detected its signal and inhibit their data to launch. Use the frame burst of half-duplex Gigabit Ethernet efficiency has been improved, when a DTE continuous burst 64byte frame and burst continuous 65536, its efficiency is approximately 72%.

1.2.2 full-duplex Gigabit Ethernet MAC layer protocol

In full-duplex Gigabit Ethernet, since each Gigabit Ethernet DTE in a channel when communicating with exclusive, so there's no need to consider the issue of conflict over Ethernet. Naturally, full-duplex GigabitEthernet is also not affected by time slot length restrictions, and thus no distance limitations of coverage.

And half-duplex mode, full-duplex Gigabit Ethernet MAC layer differences mainly has following several points:

(1) accept the send activity in the frame will not be delayed

(2) full-duplex mode indication of conflict will be ignored

(3) no carrier extended, minimum frame length is 64 bytes

(4) no frame burst

In full-duplex switched Ethernet, if more than one person handling one output port at the mouth of the red, then the output data will be generated in the output port congestion, a number of input frames surge mouth sent will be discarded. If the Ethernet frame is hosted on the TCP/IP protocol ofthe packet, then the TCP transport mechanism automatically resend dropped packets, you can imagine each had a packet of input port that will be awarded, trigger a new round of congestion or packet loss, the result is causing network throughput rate dropped significantly. In order to avoid lost packets (drop-frame) and resend the occurrence of the phenomenon, the IEEE in Mac layer introduces 802.3x flow control protocol to avoid packet loss.

Flow control is when switch detects congestion of ports, you will send to the input port, notify pause frame its send traffic, and finally achieve the Elimination of congestion. Flow control does not improve the entire switch data throughput capacity, but avoids the packet switch.

1.2.3 Gigabit Ethernet physical layer protocol

IEEE defines several different physical medium of Gigabit Ethernet interfaces, 1000Base-CX, 1000Base-SX, 1000Base-LX 1000Base-T, of which, is used for 155 1000Base-CX Ω balanced coaxial cables on the interface, in reality there is no real product, 1000Base-T is available for 5 or more categories of twisted pair interface, it's standard is the standard IEEE802.3ab just in June 1999, now the market has just launched a commercial product.

1000Base-SX use 850nm wavelength laser interface, only applies to multimode fiber. 1000Base-LX use 1300nm wavelength laser interfaces for single mode and multi mode fiber. 1000Base SX is mainly used for a campus network and enterprise network backbone. A 1000Base LX mainly applies to man, man another application more is a long-range 1000Base LH Gigabit Ethernet optical interface, the general use of 1300nm or laser wavelength 1550nm, can reach more than even 100km 50km without relay transmission distance.

It's important to note that since the IEEE is the worst transport conditions of Gigabit Ethernet transmission distance, in practice, the manufacturers of the products of transmission distance far exceeds the standard, such as Alcatel-Lucent's PowerRail gigabits routing switch 1000Base-LX interface in practical tests can not relay transmission 22km.

1.3 Gigabit Ethernet efficiency

Half-duplex Ethernet efficiency problem has always been its weaknesses in a half duplex Ethernet in workstations (such as your computer) to a threshold value, although each workstation is sending data to 10Mbps rate, but because of conflict increases, each station had to wait for a long period of time it will be possible to send data, so each workstation has a steep decline in average bandwidth available. In full-duplex switched Ethernet, CSMA/CD protocol in conflict detection mechanism of CD is no longer needed, each workstation can get exclusive bandwidth. Therefore, full-duplex switched Ethernet efficiency is no longer depends on the number of workstations within your network, but by the length of Ethernet frames.

1.4 Gigabit Ethernet reliability

From the traditional sense, Ethernet is regarded as a local area network (LAN) technology, is a large number of applications to the enterprise network, the Ethernet switches and Ethernet's reliability has not been used as the key factors to consider. With switched full-duplex fast Ethernet and Gigabit Ethernet matures, more and more operators choose Gigabit Ethernet as the man MAN's preferred technology, Gigabit Ethernet, reliability becomes a carrier of the key factors to consider.

Use Gigabit Ethernet to achieve a reliable man, now there are two mature technology:

(1) Gigabit Ethernet port aggregation;

(2) Gigabit Ethernet, 1 + 1 backup.

If Gigabit Ethernet is used to host IP service, at the network layer IP level, or ring or mesh network topology, use IP routing protocols to ensure the network reliability.

Following these three techniques for a more detailed introduction. About the reliability of Gigabit router switch, see Chapter 2.

1.4.1 Gigabit Ethernet port aggregation (PortTrunking)

In Gigabit router switch, you can associate multiple Gigabit Ethernet link bundle for a virtual logical link to achieve increased bandwidth, reliability. This technique is called port aggregation (link bundling). More common is the four Gigabit Ethernet link bundle for a link, then the bandwidth up to 4Gbps two-way 8Gbps single direction.

In the port aggregation of multiple Gigabit Ethernet key road can load sharing, even if one fibre link fails, logical link will remain functional. Port party requires more fibre to composition, 2-port port aggregation requires 4 fibersThe three ports of gathering needs 6 fiber, 4-port port party needs 8 fiber. 1.4.2 Gigabit Ethernet with 1 + 1 back-up and a lot of ATM switch in the implementation of the ATM physical links like 1 + 1 backup, and Gigabit Ethernet can also implement 1 10 1 backup, i.e. a Gigabit router switch interface module, which corresponds to a Gigabit Ethernet key road, with two Gigabit Ethernet link to connect, a Gigabit Ethernet link as the main key road, another way is as an alternate key. When the main link of fibre failure, Gigabit router switch to 1ms time to switch to the standby key of the optical fiber transmission road. In that 1 + 1 backup mode, you need four 1GbPS fibre to complete the transmission bandwidth. Each Gigabit Ethernet link requires 1 episode 1 received two fibers.

1.4.3 use IP routing to ensure network reliability

If you are using Gigabit Ethernet to host IP service, you can use IP routing protocols to ensure the convergence of network reliability. Use IP routing, network topology can be more flexible, can be a star, ring, mesh networks, their blending. Here in passing that point: If you do not use IP routing, spanning tree Protocol (SPanningTree), Gigabit Ethernet on the road even when a physical key constitutes a ring or mesh networks, switch the actual Ethernet data exchange cannot constitute a ring and mesh network. In the ring and mesh network topology, even if a link or a network node failure, due to the existence of alternate routes, the entire network will not be paralyzed. Metropolitan area network in the most typical OSPF IP routing protocol is, running OSPF protocol router using periodic Hello messages, when passed the router status found adjacent node failures, the router will automatically recalculate routing, routing can be found to ensure that alternate network restoration work. This process is called routing of convergence. General OSPF protocol of convergence time is greater than 10 seconds. And Gigabit Ethernet, 1 + 1 backup mode in comparison with the OSPF protocol from a failure recovery time is much longer.

1.5 Gigabit Ethernet and other hosted IP man technology comparison

In its current form, and TV broadband metropolitan area network hosts are based on IP, hosting of business IP Platform mainly in ATM, Gigabit Ethernet, POS, DPT these four technologies. This article does not do ATM and Gigabit Ethernet IP-hosted, there will be another article devoted to this subject. POS was originally used for wide area network in SDH hosting IP technology, can also be used for exploration of the fibre used in the metropolitan area. DPT is CISCO company proprietary man technology.

2 Gigabit router switch

2.1 routing switch definition

Traditionally, only the second layer data forwarding device known as a switch, the switch only according to the purpose of the packet and the source MAC address to be processed and forwarded, without involving a third layer of packet contents. As for Ethernet, FDDI, token Cup of a switched local area network switches. The third layer of packets forwarded by routers to complete the IP Protocol, the router checks the third layer packet destination and source IP address, and then make the appropriate treatment or forwarded. In the mid 1990s, due to a hardware chip technology limitations, routers and switches are two separate network devices. The router's internal architecture resembling one dedicated computer that has a primary CPU, such as the 486 or MIPS, memory, CPU running software for packet forwarding and routing calculation and update. So the router performance relatively poor, often become a network bottleneck.

In order to resolve software-based router performance flaws in the new ASIC chip technology, switch to handle layer 2 packets of chips feature enhancements to enable third-layer packet processing, this switch with the routing function is called routing switch.

2.2 routing switch backplane and its implementation

Back plate is the central Exchange part of the switch to switch transfers data between the various ports. The back panel of the structure and capacity to decide a routing switch. Now the routing switch back panel there are three kinds of structure: cross matrix (CrossBar); shared memory; concurrent access to shared memory. The following were discussed in detail.

2.2.1 cross matrix (CrossBar)

This structure easy to design, extensibility, and, in its basic form, you can provide a low cost per port. However, it has several key limitations.

Static memory he and team head congestion problems common impact make it harder to a case-by-port forwarding on the basis of business based on priority. So cross matrix structure provides reliable QoS support of limited capacity, and the entire IP network to improve QoS capability requirements.

2.2.2 shared memory

Traditional shared memory structure is based on the bus. This kind of structure to overcome cross-matrix backplanes limitations, and they in the backplane capacity less than 10GbPS switches are very common. In a shared memory bus architecture, all the port access through a shared the President of the central memory. Use of arbitration mechanism to control access to shared port port. This eliminates cross matrix switch with port-based static memory allocation and team head congestion problems and to an efficient useof system memory. Shared memory problem is to construct a fast enough to deliver non-blocking performance in speeds exceeding 20Gbps arbitration institution now very difficult to do. For example: the chip, the data bus technology in General is a 64-bit, President of clock frequency (not the internal clock frequency-chip) to 100MHz, such system backplane performance up to 64 × 100MHz = 6.4GbPs, two-way calculation, the system backplane performance as 12.8GbPS. Therefore, subject to the current memory chips, promoting the CD mechanism for shared memory system scalability rather poor.

2.2.3 concurrent access to shared memory

Concurrent access to shared memory is a shared-memory architecture: all the ports share a central memory space. However, unlike traditional President shared memory structure, concurrent access to shared memory for each module on each port provides a dedicated memory can be written to the central agencies and to read the mechanism that no need bus quorum device. Concurrent access to shared memory to ensure that in all the ports at the same time achieving full wire-speed performance. Parallel shared memory address bus-based shared memory scalability problems, it's every module to the central memory access speed can reach over 10GbPS, and the entire Central memory can allow for more than 30 access road at the same time, such a system backplane capacity can be expanded to 300GbPs above. While concurrent access to shared memory does not lead people cross matrix backplanes caused the team's head congestion issues.

2.3 switch layer 3 packet forwarding mechanism (chest Chinese and distributed)

Each manufacturer's routing switch implementation mechanism, the realization of the routing functionality, there are centralized and distributed two mechanisms. Discussed in detail below.

2.3.1 centralized third layer packet forwarding

Centralized third layer packet forwarding is the switch has a dedicated hardware module (routing module) to the switch's third layer packets are forwarded. Switch of each interface modules such as Gigabit Ethernet module with Exchange, does not have a third layer of processing capabilities, you need to put the third layer of rice from the back of the packets sent to the routing module to query routing and forward. Strictly speaking, this kind of structure of switch more accurate name is a layer 3 switch, instead of routing switch. Centralized third layer packet forwarding is the early technology, its disadvantage is that the entire switch routing performance is limited by the capacity of its routing module. In addition, when an IP packet to be routed, it frequently from one Ethernet interface module through the backplane bus sent to routing module, the routing module processing, and President of the Board sent to the same Ethernet network interface module, a packet delivery wasted backplane of Nha Trang, Vietnam. And the routing module failure causes the entire switch within the routing features of the defunct. Practice many manufacturers switch routing module is integrated in the form of a card in the switch software routers. Therefore in the manufacturer's product, using centralized packet forwarding switch routing capability up to 15Mpps in General.

2.3.2 distributed third layer packet forwarding

With the ASIC chip technology, with routing function module is integrated into a chip, so manufacturers are routed to the chip design routing switch every interface module, this technology is referred to as distributed third layer packet forwarding. It does not require a dedicated module for the entire chassis do the forwarding of package services, the third layer packet forwarding can consist of a per-interface module routes on chip finish. Distributed third layer packet forwarding broke through the centralized third layer packet forwarding performance bottlenecks, but its routing control mechanism is centralized to complex it needs in each portreserved routing table information for fast packet forwarding. Although technically more complicated, since performance far beyond the centralized, distributed third layer packet forwarding technology already became the mainstream of routing switch now.

2.4 wire-speed packet forwarding

Now manufacturers are often direct call their routing switches of each port is a wire-speed, so how do you determine a routing switch all the port wire speed? wire-speed is measured to 64byte packets (second or third tier package) for that commonly used benchmarks are as follows:

◆ For Gigabit Ethernet port, a wire-speed packet forwarding rate 1.488Mpps.

◆ For fast Ethernet port, a wire-speed packet forwarding rate 148.8kpps.

◆ For OC-12 POS port, a wire-speed packet forwarding rate ports as 1.17Mpps.

◆ For OC-48 POS port, a wire-speed packet forwarding rate ports as 468MppS. For Gigabit Ethernet, calculated as follows: (64 + 8 + 12) byte× 1, 488, 095pps.×8bit = 1, 000, 000, 000bps Description: when the Ethernet frames to 64byte, consider 8byte. Frames and frames clearance 12byte fixed overhead. It is a wire-speed Gigabit Ethernet port in the forwarding of package 64byte packet forwarding rate 1.488Mpps. Fast Ethernet port speed of packet forwarding rate for Gigabit Ethernet, one-tenth of 148.8kpps. For POS port, calculated as follows: a OC-12 SDH in containers of effective rate of about 599MbPs, divide it by 64 × 8bit packet length, you can draw a line speed of OC-12POS port of packet forwarding rate 1.17Mpps. OC-48 container effective rate is four times the OC-12 a, OC-48POS port of wire-speed packet forwarding rate 1.17 × 4 = 4.68Mpps.

3. routing switch selection criteria

3.1 routing switch sculpt five major standards for radio and television sector in-building broadband IP man needs for Gigabit router switch selection of actual work, this article lists the five main selection criteria for the broadcasting sector, through theselection criteria, basically you can compare the various manufacturers of routing switch of the pros and cons of performance: * backplane capacity * non-blocking Gigabit port number * second-tier Pack transferredSend speed * layer 3 packet forwarding rate * routing number

3.1.1 backplane capacity

Measuring routing switch capacity size of the key indicators is switch backplane capacity, the unit is GbPs.

3.1.2 non-blocking Gigabit Ports quantity

A Gigabit router switch can exchange or routing multiple Gigabit Ethernet port, but the maximum supported number of Gigabit Ethernet port does not mean that it can fully non-blocking, wire-speed switching these Gigabit ports. Because some Gigabit router switch is designed for computer servers provide gigabit connections, and the existing computer Gigabit NIC communication speed is limited by computer bus, far 1000MbPS, typically 300 ~ 400MbPS, so for these installation a Gigabit Ethernet Server and there is no need to provide wire-speed Gigabit switching. However, some manufacturers of products often shy away from this design goal, publicity Gigabit router switch support up to Gigabit port number, but the true measure of Gigabit router switch is its ability to Exchange non-blocking Gigabit port number.

3.1.3 the second layer packet forwarding speed

For Gigabit router switch, layer 2 packet forwarding rate is the forward speed of Ethernet frames. To PPS (packets per second) to measure units.

3.1.4 the third layer packet forwarding speed

The third layer packet forwarding rate means a Gigabit router switch forwarding layer 3 Protocol package speed, such as forwarding IP or IPX packet. To PPS (packets per second) to measure units.

Here it should be noted that the second layer packet forwarding rate and third layer packet forwarding rate are two different concepts. Many manufacturers are often only the bag forward speed, with no clear distinction is the second-tier or Layer 3 packet forwarding rate. For a distributed routing Routing switches, in General, the second layer packet forwarding rate equal to the rate of three-layer packet forwarding. For centralized routing routing switch, the third layer packet forwarding rate often not equal to the second layer of packet forwarding rate because the centralized routing requires a separate routing module for layer 3 packet forwarding, while the second layer packet forwarding is in all Gigabit interface module.

3.1.5 routing number

Routing switch routing table support routing number, meaning it can support the larger the network topology, the typical metropolitan area network routing switch routing number is 64K/per port. Here is the number of routing manufacturers claim is each chassis or each port support, typically, the number of routing per chassis = port number × the number of routing per port. Current Internet backbone actual router running BGP-4 routing table is approximately 77K (source: www.telatra.net/ops/bgptable.html) and slow growth, growth rate slowing mainly due to the introduction of CIDR techniques. For a metropolitan area network routing switch, supports 64 k routing table capacity can meet the future a long time man needs.

3.2 judgment Gigabit router switch non-blocking standard

A Gigabit router switch should meet the following requirements before you can implement a genuine non-blocking.

(1) backplane is non-blocking architecture and common algorithms is if the backplane capacity ≥ port number × port rate × 2, then the routing switch on the back panel and is non-blocking.

(2) a second layer of wire-speed packet forwarding, the algorithm is if the chassis of the second layer packet forwarding rate = number of Gigabit ports, then speak 1.488Mpps × routing switch in layer 2 switching when possible line was built.

(3) a third layer of wire-speed packet forwarding, the algorithm is if the chassis of the third layer packetforwarding rate = number of Gigabit ports, so this 1.488Mpps × routing switch in the third Exchange can achieve wire-speed at.

There are many manufacturers of the product data meets the second and third floor wire-speed standard, but does not meet the back plate non-blocking, this indicates that the second and third floor wire-speed packet fails in the back right Exchange conditions obtain; if you have a large number of packets required by its back panel forward, then the routing switch will not be able to do marketing.

Another situation is back to meet the requirements of the non-blocking, but its second and third tiers of packet forwarding rate does not meet the standard, the lower speed that routing switch packet forwarding module bottlenecks.

Need to clarify that, in this article referred to blocking and congestion in Routing switches are two different concepts. Congestion is when more than one port to another port to send data, thanks to the receive port rate less than the rate of multiple ports and caused data packet loss or decline to send rates, it can use standard IEEE802.3X flow control protocol to be controlled. Obstruction is due to the shortcomings of the internal structure of the switch from a single port communication rate less than full-rate problems, the problem can be via streaming technology agreement.

4. concluding remarks

Gigabit Ethernet technology is maturing, apart from the original compatibility, broadband, cheap, good support for IP, is to enhance its reliability and scalability. In the current broadcasting broadband metropolitan area network primarily hosted IP services, Gigabit Ethernet is the preferred technology for broadband metropolitan area network.