The chip speed doubles every 18 months, while Internet traffic doubles every six months. As the hub of the Internet, routers are moving in three directions: faster, better quality of service, and easier integration management. The function of the router Before we analyze the development trend of the router, let's briefly introduce the function of the router. Traditionally, the router works on the third layer of the so-called network 7-layer protocol. Its main task is to receive data packets from a network interface and decide which next destination to forward to (based on the destination address). It may be the ultimate destination) and decide which network interface to forward from. This is the most basic function of the router - packet forwarding. In order to maintain and use the router, the router also needs to have configuration or control functions.
According to the TCP/IP protocol, the specific process of packet forwarding of a router is:
The network interface receives the data packet; this step is responsible for the network physical layer processing, that is, the encoded data signal is restored to data. Different physical network media determine different network interfaces. For example, corresponding to 10Base-T Ethernet, the router has 10Base-T Ethernet interface, corresponding to SDH, the router has SDH interface, corresponding to DDN, and the router has V.35 interface.
According to the network physical interface, the router invokes the corresponding link layer (the second layer in the network layer 7 protocol) function module to interpret the link layer protocol header for processing this data packet. This step is relatively simple, mainly for the verification of data integrity, such as CRC check, frame length check. In recent years, the trend of IP over something is very obvious. The rapid development of fiber network technology and the establishment of IP as a de facto standard make IP on the DWDM (dense wavelength division multiplexing) fiber (at the network layer - network layer 7). The third layer in the protocol) is skipped over the link layer and loaded directly onto the physical layer.
After the integrity verification of the data frame is completed at the link layer, the router begins processing the IP layer of this data frame. This process is at the heart of the router's capabilities. According to the destination IP address of the IP header in the data frame, the router searches for the IP address of the next hop in the routing table. The TTL (Time to Live) field of the IP packet header starts to be decremented, and a new checksum is calculated. If the network interface type of the received data frame is different from the network interface type of the forwarded data frame, the IP data packet may also be segmented or reassembled due to the specification of the maximum frame length.
According to the next hop IP address found in the routing table, the IP data packet is sent to the corresponding output link layer, encapsulated with the corresponding link layer header, and finally sent out through the output network physical interface.
The following describes the development trend of the router.
faster
Traditionally, routers are often considered to be bottlenecks in network speed. When the LAN speed has reached hundreds of megahertz, the processing speed of the router is only up to tens of megabits. In recent years, along with the explosive growth of the Internet, everyone's research on routers has also focused on improving the processing speed of routers. In 1996 and 1997, a number of innovative small companies such as Nexabit, Juniper, and Avici emerged in the United States, which increased the processing speed of routers to the peak of the peak, and launched the Gigabit routers in a very short period of time. Even Cisco can only look back on speed. Since these high-speed routers have introduced the switching structure without exception, these routers are also called GSR-Gigabit Switch Routers and TSRs. The optical interface speed of these routers also quickly jumped from OC-12 (622Mbps) to OC-48 (2.5Gbps) to OC-192 (10Gbps), which has already left the ATM switch far behind. Since then, the irreplaceable position of ATM in the core network has completely shaken. The prolonged IP-TM technology battle has finally ended with an overwhelming advantage of IP. However, from the following analysis, we can also see that the speed of IP routers is directly benefited from the concept and technology of ATM. Many new concepts and technologies proposed in the IP field are also directly or indirectly derived. At ATM, two excellent technologies are gradually beginning to merge. In fact, many of the companies that are engaged in the development of high-speed IP routers are the researchers who have studied ATM technology in the past. Specifically, the dramatic increase in the speed of IP routers comes from the following four technical advances.
Architecture. The hardware architecture of the router has undergone roughly six changes (discussed in "Router Architecture"), from the earliest single-bus, single-CPU architecture to single-bus, multi-CPU, and multi-bus multi-CPU. Up to now, high-speed IP routers have borrowed more ATM methods and implemented cross-switching to achieve line-rate non-blocking interconnections between ports. The technology of high-speed crossbars is very mature, and has been widely used in ATMs and high-speed parallel computers. The speed of high-speed crossbars that can be purchased directly on the market is as high as 50 Gbps. Along with the introduction of high-speed cross-switches, some corresponding technical problems have also been introduced, especially for IP multicast, broadcast and quality of service (QoS), using mature scheduling strategies and algorithms, and these problems have been well solved. .
ASIC technology. Over the years, ASICs have become more widely used for cost and performance reasons, and they are almost certainly called ASICs. To greatly improve the speed in the router, the first thing that comes to mind is the ASIC. Some use ASIC for packet forwarding, some use ASIC to check routing, and ASIC chips for IPV4 routing are already on the market. In the trend of ASIC booming and mass application, there is a trend worth paying attention to. This is the emergence of so-called programmable ASIC, which is probably the result of the rapid changes in the network itself. Due to the considerable investment in the design and production of ASICs, AISC is generally only used in processes that have been fully standardized, and the structure and protocols of the network have changed quite rapidly, so correspondingly in the field of network equipment, there has been a strange " Programmable ASIC". At present, there are two types of so-called "programmable ASIC", one is represented by 3COM company's FIRE (Flexible Intelligent Routing Engine) chip. This ASIC chip has a CPU embedded in it, so it has a certain degree of flexibility. The other is represented by Vertex Networks' HISC dedicated chip. This chip is a CPU specially designed for communication protocol. The CPU architecture is designed to be specially adapted to the protocol processing. By rewriting the microcode, this dedicated chip can be made. The ability to handle different protocols to accommodate similar changes from IPV4 to IPV6.
Three-tier exchange. This is a revolutionary breakthrough in the protocol process and a source of GSR and TSR names. Since the mysterious Ipsilon company introduced the "One Route, Then Exchange" IPSwitch technology in 1994, major companies have introduced their own proprietary Layer 3 switching technology. Such as Cisco's Tag Switch, 3Com's Label Switch, and so on. Combining the advantages of these proprietary technologies, the IETF finally introduced the superior performance of Multi-Protocol Label Switching (MPLS) in 1998. Compared with the original idea of "one route, then exchange", MPLS considers the three-layer switching technology from a higher level of network structure, and tries to solve the problem of traffic management of the three-layer switching network in one fell swoop. Unlike the original Ipswitch technology, the MPLS protocol changes the IP protocol packet. At the edge of the network, the MPLS router labels each incoming IP packet. In the subsequent transmission, the core routing switching device will Determining the forwarding path based only on this label is very similar to the virtual circuit concept in the ATM world. At present, research in this area is still in progress. The main technical difficulty lies in how to determine the label allocation scheme on the network edge router in the network autonomous system, and how to dynamically adjust this scheme according to the network load and fault condition.
IP over SDH, IP over DWDM. The technological advances in this area are entirely due to the advancement of fiber-optic communication technology. As the core position of IP is gradually recognized, the way of IP over ATM and then ATM over SDH is replaced by IP directly over SDH. SDH uses time division multiplexing to carry multiple channels of data. Therefore, a large number of multiplexer cross connectors are required in the core network. DWDM (Dense Wavelength Division Multiplexing) allows multiple signals to be transmitted at different wavelengths on a single fiber. Generally, four wavelengths running on one fiber at a time can be called DWDM. Since the introduction of 16 wavelength DWDM fiber-optic communication products in 1996, the 40-wavelength DWDM technology has been put into practical use. 80 or even 96-wavelength DWDM products will be launched in 2000, and China has already developed 8 wavelengths. DWDM technology. Due to the use of wavelength division multiplexing, the transmission of data on the fiber becomes quite simple. Advances in optical communication technology allow optical signals to be transmitted directly over a distance of 800 kilometers without the need for any optoelectronic or optical regenerative amplifiers. IP packets are directly modulated at a certain wavelength without being reused and demultiplexed. Even in the core network, wavelength information is directly used as path information of the IP data stream.
Better service quality
The speed increase of the router described above is still only to accommodate the sharp increase in data traffic. The more fundamental and deeper changes in the development trend of routers are: IP-based packet-switched data will quickly replace the circuit-switched communication methods that have been developed for nearly a hundred years in the next few years, becoming the mainstream of the communication business model. This means that IP routers not only need to provide faster speeds to accommodate the rapidly growing traditional computer data traffic, but IP routers will also gradually provide the services offered by the original telecommunications network. However, traditional IP routers do not care about the service type of IP packets. Generally, they only forward data packets according to the principle of advanced and first-out, voice telephone data, real-time video data, Internet browsing data, and other data types. Treated indiscriminately. It can be seen that IP routers are the key to improving the quality of service (QOS) in order to provide all services including telecommunication broadcasting. This is also the direction that the major network equipment manufacturers (including Cisco, 3Com, Nortel, etc.) are trying to advance. The high-, medium-, and low-end routers introduced by major vendors support QoS to varying degrees. For example, Cisco's highest-end 12000 series has strong support for QoS in both hardware and software protocols, and its new low-cost. The end product 2600 series also supports new business applications such as voice telephony. In fact, QoS is not only a development trend of routers, but the entire IP network with router as the core is developing in this direction. The concept of "three networks in one" is the product of this direction. However, the network with the traditional IP router as the core has been unable to adapt to the trend of “three networks in one”. All countries led by the United States are advancing the development of network technologies that can provide better and faster service quality. Among them, the research and development of routers is the key, and the company has become the main driving force for this technology.
Support for QoS comes from both software and hardware. From a hardware perspective, faster forwarding speeds and wider bandwidth are fundamental prerequisites. In terms of software agreements, recent efforts have been shown in the following results:
PV4 header service type field. There is a 3-bit area in the IPV4 header to identify the priority of this IP packet. According to this priority, the IP router can determine the forwarding priority of different IP packets. It can be said that since the date of the IP protocol, the mechanism for providing better QoS in the future has been guaranteed. But because IP networks are not focused on QoS in the early days of booming. Therefore, this person's 3-digit area is generally not used. However, as we can see from the analysis below, it is absolutely not enough to define the service type in the IP packet. Through signaling, the required quality of service must be guaranteed in all aspects of the entire network.
RSVP (Resource Reservation Protocol) and the corresponding series of protocols. This is a profound step forward for IP routers to move forward with better quality of service. Traditionally, IP routers are only responsible for packet forwarding, and the routing protocol knows the address of the neighboring router. RSVP, like the signaling protocol of a circuit-switched system, notifies each node (IP router) through which a data stream passes, and negotiates with the endpoint to provide quality assurance for this data stream. As soon as the RSVP protocol appeared, it was widely recognized and basically solved the problem of resource reservation. However, with the passage of time, the explosive growth of the network, the problems exposed by RSVP are more and more, mainly reflected in the following aspects:
The most fundamental thing is that RSVP is a negotiation service object for each data stream. In the case of explosive growth of network traffic, the number of data streams forwarded by the router increases sharply. To improve the forwarding speed, a lot of special designs have been made in the router. It is simply impossible to make complex resource reservation protocols for each data stream.
Secondly, when the route is modified due to busy lines or router failures, it is necessary to perform a relatively time-consuming RSVP process again.
For the above two reasons, the IETF has introduced another QoS policy, DiffServ (Differentiated Service). At present, the framework of DiffServ has been basically determined, and the Internet2 of the United States also chooses DiffServ as its QoS policy. Compared with DiffServ, RSVP is an Integrated Service with centralized control strategy, while DiffServ is a decentralized control strategy whose essence is to control only the behavior of each hop in the path. The terminal application device negotiates with the edge router through the SLA (Service Level Agreement) to obtain the service level that the application data stream can be guaranteed. According to this service level, the edge router marks each received packet with a level, and the core router only determines the mobility behavior when forwarding according to the mark of the service level of each packet. Since the customer only negotiates with the edge router and obtains the service level guarantee, the actual service quality of the same level of service level provided by different edge routers is different in an interconnected large network due to uneven network traffic. It is necessary to exchange traffic information between different network areas that provide QoS service levels through SLAs to avoid or reduce the occurrence of the above situation.
Multi-protocol labeling
According to the TCP/IP protocol, the specific process of packet forwarding of a router is:
The network interface receives the data packet; this step is responsible for the network physical layer processing, that is, the encoded data signal is restored to data. Different physical network media determine different network interfaces. For example, corresponding to 10Base-T Ethernet, the router has 10Base-T Ethernet interface, corresponding to SDH, the router has SDH interface, corresponding to DDN, and the router has V.35 interface.
According to the network physical interface, the router invokes the corresponding link layer (the second layer in the network layer 7 protocol) function module to interpret the link layer protocol header for processing this data packet. This step is relatively simple, mainly for the verification of data integrity, such as CRC check, frame length check. In recent years, the trend of IP over something is very obvious. The rapid development of fiber network technology and the establishment of IP as a de facto standard make IP on the DWDM (dense wavelength division multiplexing) fiber (at the network layer - network layer 7). The third layer in the protocol) is skipped over the link layer and loaded directly onto the physical layer.
After the integrity verification of the data frame is completed at the link layer, the router begins processing the IP layer of this data frame. This process is at the heart of the router's capabilities. According to the destination IP address of the IP header in the data frame, the router searches for the IP address of the next hop in the routing table. The TTL (Time to Live) field of the IP packet header starts to be decremented, and a new checksum is calculated. If the network interface type of the received data frame is different from the network interface type of the forwarded data frame, the IP data packet may also be segmented or reassembled due to the specification of the maximum frame length.
According to the next hop IP address found in the routing table, the IP data packet is sent to the corresponding output link layer, encapsulated with the corresponding link layer header, and finally sent out through the output network physical interface.
The following describes the development trend of the router.
faster
Traditionally, routers are often considered to be bottlenecks in network speed. When the LAN speed has reached hundreds of megahertz, the processing speed of the router is only up to tens of megabits. In recent years, along with the explosive growth of the Internet, everyone's research on routers has also focused on improving the processing speed of routers. In 1996 and 1997, a number of innovative small companies such as Nexabit, Juniper, and Avici emerged in the United States, which increased the processing speed of routers to the peak of the peak, and launched the Gigabit routers in a very short period of time. Even Cisco can only look back on speed. Since these high-speed routers have introduced the switching structure without exception, these routers are also called GSR-Gigabit Switch Routers and TSRs. The optical interface speed of these routers also quickly jumped from OC-12 (622Mbps) to OC-48 (2.5Gbps) to OC-192 (10Gbps), which has already left the ATM switch far behind. Since then, the irreplaceable position of ATM in the core network has completely shaken. The prolonged IP-TM technology battle has finally ended with an overwhelming advantage of IP. However, from the following analysis, we can also see that the speed of IP routers is directly benefited from the concept and technology of ATM. Many new concepts and technologies proposed in the IP field are also directly or indirectly derived. At ATM, two excellent technologies are gradually beginning to merge. In fact, many of the companies that are engaged in the development of high-speed IP routers are the researchers who have studied ATM technology in the past. Specifically, the dramatic increase in the speed of IP routers comes from the following four technical advances.
Architecture. The hardware architecture of the router has undergone roughly six changes (discussed in "Router Architecture"), from the earliest single-bus, single-CPU architecture to single-bus, multi-CPU, and multi-bus multi-CPU. Up to now, high-speed IP routers have borrowed more ATM methods and implemented cross-switching to achieve line-rate non-blocking interconnections between ports. The technology of high-speed crossbars is very mature, and has been widely used in ATMs and high-speed parallel computers. The speed of high-speed crossbars that can be purchased directly on the market is as high as 50 Gbps. Along with the introduction of high-speed cross-switches, some corresponding technical problems have also been introduced, especially for IP multicast, broadcast and quality of service (QoS), using mature scheduling strategies and algorithms, and these problems have been well solved. .
ASIC technology. Over the years, ASICs have become more widely used for cost and performance reasons, and they are almost certainly called ASICs. To greatly improve the speed in the router, the first thing that comes to mind is the ASIC. Some use ASIC for packet forwarding, some use ASIC to check routing, and ASIC chips for IPV4 routing are already on the market. In the trend of ASIC booming and mass application, there is a trend worth paying attention to. This is the emergence of so-called programmable ASIC, which is probably the result of the rapid changes in the network itself. Due to the considerable investment in the design and production of ASICs, AISC is generally only used in processes that have been fully standardized, and the structure and protocols of the network have changed quite rapidly, so correspondingly in the field of network equipment, there has been a strange " Programmable ASIC". At present, there are two types of so-called "programmable ASIC", one is represented by 3COM company's FIRE (Flexible Intelligent Routing Engine) chip. This ASIC chip has a CPU embedded in it, so it has a certain degree of flexibility. The other is represented by Vertex Networks' HISC dedicated chip. This chip is a CPU specially designed for communication protocol. The CPU architecture is designed to be specially adapted to the protocol processing. By rewriting the microcode, this dedicated chip can be made. The ability to handle different protocols to accommodate similar changes from IPV4 to IPV6.
Three-tier exchange. This is a revolutionary breakthrough in the protocol process and a source of GSR and TSR names. Since the mysterious Ipsilon company introduced the "One Route, Then Exchange" IPSwitch technology in 1994, major companies have introduced their own proprietary Layer 3 switching technology. Such as Cisco's Tag Switch, 3Com's Label Switch, and so on. Combining the advantages of these proprietary technologies, the IETF finally introduced the superior performance of Multi-Protocol Label Switching (MPLS) in 1998. Compared with the original idea of "one route, then exchange", MPLS considers the three-layer switching technology from a higher level of network structure, and tries to solve the problem of traffic management of the three-layer switching network in one fell swoop. Unlike the original Ipswitch technology, the MPLS protocol changes the IP protocol packet. At the edge of the network, the MPLS router labels each incoming IP packet. In the subsequent transmission, the core routing switching device will Determining the forwarding path based only on this label is very similar to the virtual circuit concept in the ATM world. At present, research in this area is still in progress. The main technical difficulty lies in how to determine the label allocation scheme on the network edge router in the network autonomous system, and how to dynamically adjust this scheme according to the network load and fault condition.
IP over SDH, IP over DWDM. The technological advances in this area are entirely due to the advancement of fiber-optic communication technology. As the core position of IP is gradually recognized, the way of IP over ATM and then ATM over SDH is replaced by IP directly over SDH. SDH uses time division multiplexing to carry multiple channels of data. Therefore, a large number of multiplexer cross connectors are required in the core network. DWDM (Dense Wavelength Division Multiplexing) allows multiple signals to be transmitted at different wavelengths on a single fiber. Generally, four wavelengths running on one fiber at a time can be called DWDM. Since the introduction of 16 wavelength DWDM fiber-optic communication products in 1996, the 40-wavelength DWDM technology has been put into practical use. 80 or even 96-wavelength DWDM products will be launched in 2000, and China has already developed 8 wavelengths. DWDM technology. Due to the use of wavelength division multiplexing, the transmission of data on the fiber becomes quite simple. Advances in optical communication technology allow optical signals to be transmitted directly over a distance of 800 kilometers without the need for any optoelectronic or optical regenerative amplifiers. IP packets are directly modulated at a certain wavelength without being reused and demultiplexed. Even in the core network, wavelength information is directly used as path information of the IP data stream.
Better service quality
The speed increase of the router described above is still only to accommodate the sharp increase in data traffic. The more fundamental and deeper changes in the development trend of routers are: IP-based packet-switched data will quickly replace the circuit-switched communication methods that have been developed for nearly a hundred years in the next few years, becoming the mainstream of the communication business model. This means that IP routers not only need to provide faster speeds to accommodate the rapidly growing traditional computer data traffic, but IP routers will also gradually provide the services offered by the original telecommunications network. However, traditional IP routers do not care about the service type of IP packets. Generally, they only forward data packets according to the principle of advanced and first-out, voice telephone data, real-time video data, Internet browsing data, and other data types. Treated indiscriminately. It can be seen that IP routers are the key to improving the quality of service (QOS) in order to provide all services including telecommunication broadcasting. This is also the direction that the major network equipment manufacturers (including Cisco, 3Com, Nortel, etc.) are trying to advance. The high-, medium-, and low-end routers introduced by major vendors support QoS to varying degrees. For example, Cisco's highest-end 12000 series has strong support for QoS in both hardware and software protocols, and its new low-cost. The end product 2600 series also supports new business applications such as voice telephony. In fact, QoS is not only a development trend of routers, but the entire IP network with router as the core is developing in this direction. The concept of "three networks in one" is the product of this direction. However, the network with the traditional IP router as the core has been unable to adapt to the trend of “three networks in one”. All countries led by the United States are advancing the development of network technologies that can provide better and faster service quality. Among them, the research and development of routers is the key, and the company has become the main driving force for this technology.
Support for QoS comes from both software and hardware. From a hardware perspective, faster forwarding speeds and wider bandwidth are fundamental prerequisites. In terms of software agreements, recent efforts have been shown in the following results:
PV4 header service type field. There is a 3-bit area in the IPV4 header to identify the priority of this IP packet. According to this priority, the IP router can determine the forwarding priority of different IP packets. It can be said that since the date of the IP protocol, the mechanism for providing better QoS in the future has been guaranteed. But because IP networks are not focused on QoS in the early days of booming. Therefore, this person's 3-digit area is generally not used. However, as we can see from the analysis below, it is absolutely not enough to define the service type in the IP packet. Through signaling, the required quality of service must be guaranteed in all aspects of the entire network.
RSVP (Resource Reservation Protocol) and the corresponding series of protocols. This is a profound step forward for IP routers to move forward with better quality of service. Traditionally, IP routers are only responsible for packet forwarding, and the routing protocol knows the address of the neighboring router. RSVP, like the signaling protocol of a circuit-switched system, notifies each node (IP router) through which a data stream passes, and negotiates with the endpoint to provide quality assurance for this data stream. As soon as the RSVP protocol appeared, it was widely recognized and basically solved the problem of resource reservation. However, with the passage of time, the explosive growth of the network, the problems exposed by RSVP are more and more, mainly reflected in the following aspects:
The most fundamental thing is that RSVP is a negotiation service object for each data stream. In the case of explosive growth of network traffic, the number of data streams forwarded by the router increases sharply. To improve the forwarding speed, a lot of special designs have been made in the router. It is simply impossible to make complex resource reservation protocols for each data stream.
Secondly, when the route is modified due to busy lines or router failures, it is necessary to perform a relatively time-consuming RSVP process again.
For the above two reasons, the IETF has introduced another QoS policy, DiffServ (Differentiated Service). At present, the framework of DiffServ has been basically determined, and the Internet2 of the United States also chooses DiffServ as its QoS policy. Compared with DiffServ, RSVP is an Integrated Service with centralized control strategy, while DiffServ is a decentralized control strategy whose essence is to control only the behavior of each hop in the path. The terminal application device negotiates with the edge router through the SLA (Service Level Agreement) to obtain the service level that the application data stream can be guaranteed. According to this service level, the edge router marks each received packet with a level, and the core router only determines the mobility behavior when forwarding according to the mark of the service level of each packet. Since the customer only negotiates with the edge router and obtains the service level guarantee, the actual service quality of the same level of service level provided by different edge routers is different in an interconnected large network due to uneven network traffic. It is necessary to exchange traffic information between different network areas that provide QoS service levels through SLAs to avoid or reduce the occurrence of the above situation.
Multi-protocol labeling
