RU2233473C2 - Device and method for performing high-speed search for routes of internet protocol and controlling routing/transfer tables - Google Patents

Abstract

FIELD: Internet routing technologies.

SUBSTANCE: method for forming tables of transfer/routing for searching of Internet protocol address(IP-address) with use of access list includes steps of separation of IP-address prefix lengths range by previously given method, creation of header node having maximum level, on basis of several clusters picked in range of prefix lengths, while header node pints to each node in the access list, and creation of sub-nodes by means of picked clusters, while each of sub-nodes has a selected prefix lengths range as a key.

EFFECT: lower time needed for accessing memory and its effective use, simplified routes changing, effective composition of routing table and/or transfer table without sorting acceptable routes in accordance to length of routes prefix, inputted in process of tables forming.

7 cl, 2 tbl, 18 dwg

Description

Technical field

The present invention relates generally to routing methods for forwarding packets to a destination on the Internet, and in particular, to a device and method for performing a high-speed IP search (Internet Protocol) for a route and managing a routing table (or forwarding).

State of the art

Due to the increase in the number of Internet users, the expansion of the list of supported services and service areas, such as VoIP (Voice over Internet Protocol) application systems and stream-oriented systems, an exponential increase in Internet traffic is observed. A possible solution to this problem could be to forward packets to the next transit interface by finding the path to the destination with virtually no delay using high-speed routers with speeds of the order of gigabits / s or terabits / s. In order to find the path to the destination of packets forwarded through the physical input interfaces of the router, it is necessary to keep the routing table (or forwarding) in a compact form and minimize the search time.

As for the well-known router, before the advent of the recently commissioned high-speed routers, when the time required to process the packets and find their destinations is less than the time required for the transmission paths, the router as a relay node connecting subnets or other networks performed its functions well . Recently, an increase in the bandwidth of an optical network interface, such as POS OS-192 (10 Gb / s) or IP via DWDM (high-density wavelength multiplexing), has led to the fact that the router requiring a certain processing time turned out to be the narrowest place in high speed internet. Router descriptions can be found in: Keshav, S. and Sharma, R., Issues and Trends in Router Design, IEEE Communications Magazine, pages 144-151, May, 1998; Kumar, V. and Lakshman, T. and Stiliadis, D., Beyond Best Effort: Router Architectures for the Differentiated Services of Tomorrow's Internet, IEEE Communications Magazine, pages 152-164, May, 1998; Chan, H., Alnuweiri, H. and Leung, V., A Framework for Optimizing the Cost and Performance of Next Generation IP Routers, IEEE Journal of Selected Areas in Communications, Vol. 17, No. 6, pages 1013-1029, June 1999; Partridge, C. et al., A 50-Gb / s IP Router, IEEE / ACM Trans. on Networking, vol. 6, no. 3, pages 237-248.1998; and Metz, C., IP Routers: New Tool for Gigabit Networking, IEEE Internet Computing, pages 14-18, Nov.-Dec., 1998.

In early 1990, the IETF (Internet Engineering Task Force) working group introduced a new IP addressing scheme called CIDR (Classless Cross-Domain Routing) to efficiently use IP addresses in IPv4 (IP, version 4) (See RFC 1518, An Architecture for IP Address Allocation with CIDR, Sept., 1993; and RFC 1517, Applicability Statement for the Implementation of Classless Inter-Domain Routing (CIDR), Sept., 1993).

In this case, to capture the prefix in packets with prefixes of different lengths and find the IP address corresponding to the longest prefix, the LPM algorithm (matching the longest prefix) based on the so-called "trie structure" or Patricia data structure ( A rational algorithm for finding information encoded with alphanumeric characters) (See Doeringer, W., Karjoth, G. and Nassehi, M., Routing on Longest-Matching Prefixes, IEEE / ACM Trans. On Networking, vol. 4, no . 1, pages 86-97, Feb., 1996).

The LPM algorithm is described in detail below with reference to Table 1. Table 1 is a simple example of routing entries contained in a routing table (or forwarding). The asterisk indicates the possibility of filling the position with any number: "0" or "1", although in the case of using IPv4, the length of the prefix itself cannot exceed 32.

The route record or route is represented by the IP address:: = {<networkprefix>, <xost number>}. When the destination IP address is '10010101000 ... 0', the first bit of the destination IP address is compared with the network prefixes (simply prefixes) contained in the routing table. As a result, prefixes such as '0 *', '01010 *', '010101 *' and '0101011 *' are discarded because they have the first bit starting with '0'. When this destination address is compared with other prefixes, the prefixes '11 * ',' 1110 * 'and' 1001010101 * 'are excluded because they do not match the destination address starting from the second or tenth bit. Of the appropriate prefixes (for example, '1 *', '10 * 'and' 100101010 * '),' 100101010 * 'is selected as the longest (or most specific) prefix, and the IP packet with the destination IP address is sent to the neighboring router via the interface '10'.

In addition, a recently announced route search algorithm is stored in the cache memory of a particular processor (routing processor or forwarding processor) to reduce memory access time. Although the route-finding algorithm based on the routing table guarantees efficient searches, it cannot modify the routing table (or forwarders). Therefore, using the route search algorithm may cause a delay associated with the need to reflect changes in the routing table.

In addition, the routing table in the routing processor is copied through a DMA card (direct memory access), an IPC card (interprocess communication), or a switching structure, and then fed to the forwarding table in the forwarding processor. At this point, you need to build a new forwarding table, not just add or remove changed routes. This can lead to additional time delay, as well as to the fact that the internal memory of the router or the system bus will become a bottleneck due to the increase in memory bandwidth when requesting access to memory from the router.

In addition, in a number of algorithms, the initial routing table (or forwarding) must be constructed using the connection reachability information representing the current state of the router's connections. Therefore, in order to sequentially enter information on the reachability of connections in the routing table, an additional process of sorting each route in accordance with the prefix length is necessary in advance (See Degermark, M., Brodnik, A., Carlsson, S. and Pink, S., Small Forwarding Tables for Fast Routing Lookups, In Proceedings of ACM SIGCOMM '97, pages 3-14, Cannes, France, 1997; Srinivasan, V. and Varghese, G., Faster IP Lookups using Controlled Prefix Expansion, In Proceedings of ACM Sigmetries '98 Conf., Pages 1-11, 1998; Lampson, B., Srinivasan, V. and Varghese, G., IP Lookups using Multi-way and Multicolumn Search, In IEEE Infocom, pages 1248-1256, 1998; Tzeng, H. and Pryzygienda, T., On Fast Address-Lookup Algorithms, IEEE Journal on Selected Areas in Communications, Vol. 17, No. 6 , pages 1067-1082, June, 1999; Waldvogel, M., Varghese, G., Turner, J. and Plattner, B., Scalable High Speed IP Routing Lookups, In Proceedings of ACM SIGCOMM '97, Cannes, France, pages 25-37, 1997; and Waldvogel, M., Varghese, G., Turner, J. and Plattner, B., Scalable Best Matching Prefix Lookups, In Proceedings of PODC '98, Puerto Vallarta, page, 1998).

The use of conventional data structures in the routing table, such as the base tree or the Patricia trie structure, not only increases the number of memory calls required to find the path for the packet, but also requires a significant update time to reflect changes in routes that result from setting or removing routes from neighboring routers of the corresponding router. Degermark, M., Brodnik, A., Carlsson, S. and Pink, S. in "Small Forwarding Tables for Fast Routing Lookups", in Proceedings of ACM SIGCOMM'97, pages 3-14, Cannes, France, 1997 A compact forwarding table is proposed that can be stored in the forwarding processor cache, but it is difficult to reflect changes in routes in it. A method for controlled extension of a prefix proposed by Srinivasan, V. and Varghese, G., Faster IP Lookups using Controlled Prefix Expansion, in Proceedings of ACM Sigmetrics 98 Conf., Pages 1-11, 1998 (see U.S. Patent No. 6,011795, issued George Varghese and Srinivasan on “A Method and Device for Quick Hierarchical Address Search Using a Managed Prefix Extension”), and the Fast Address Search Algorithm proposed by Tzeng, N. and Pryzygienda, T., “On Fast Address-Lookup Algoritms, IEEE Journal on Selected Areas in Communications, Vol. 17, No.6, pages 1067-1082, June 1999, are also based on the use of a variable resolution trie structure, but when applying this algorithm, problems arise when adding and removing routes, since data structures are based on a trie structure (see US patent No. 6061712 for “Search method using the IP routing table” IP and US patent No. 6067574 for “High-speed routing using the tree compression procedure” in the name of Hong-Yi Tzeng). Connected Search Algorithm (Waldvogel, M., Varghese, G., Turner, J. and Plattner, B., Scalable High Speed IP Routing Lookups, In Proceedings of ACM SIGCOMM '97, Cannes, France, pages 25-37, 1997; and Waldvogel, M., Varghese, G., Turner, J. and Plattner, B., Scalable Best Matching Prefix Lookups, In Proceedings of PODC '98, Puerto Vallarta, page, 1998), based on the mapping of the trie structure to binary a tree using hash tables and a full prefix trie structure based on a variable resolution trie structure (Tzeng, H. and Pryzygienda, T., On Fast Address-Lookup Algorithms, IEEE Journal on Selected Areas in Communications, Vol. 17 , No. 6, pages 1067-1082, June, 1999), is ineffective due to the need to update records modified x routes, as these data structures are based on trie-structure (see. US Patent №6018524 issued Jonathan Turner, George Varghese and Marcel Waldvogel "Scalable High Speed IP Routing Lookups".

In addition, other variants of the trie structure have been proposed. For example, a scheme with a double trie structure is proposed (Kijkanjanarat, T. and Chao, H., Fast IP Lookups Using a Two-trie Data Structure, In Proceedings of Globecom '99, 1999), where two trie- are connected to reduce the search time structures, as well as the trie structure of LC (Nillson, S. and Karlsoson, G., IP-Addresses Lookup Using LC-Tries, IEEE Journal on Selected Areas in Communications, Vol.17, No.16, pages 1083-1092, 1999) to reduce the level length in the trie structure, and the trie structure of DP (Doeringer, W., Karjoth, G. and Nassehi, M., Routing on Longest-Matching Prefixes, IEEE / ACM Trans. On Networking, vol. 4 , no.1, pages 86-97, Feb., 1996). However, they also have problems reflecting changed routes in the routing table (or forwarding).

In addition, although the aforementioned schemes contribute to a reduction in search time, there remains a problem regarding updating routes. Circuits have also been proposed to reduce hardware-based search time. Gupta, P., Lin, S. and McKeown, N., Routing Lookups in Hardware at Memory Access Speeds, In Proceedings of IEEE INFOCOM '98 Conf., Pages 1240-1247, 1998, proposed a solution based on the use of multidimensional memory. At the same time, it is possible to reduce the search time compared to the options based on software, but this involves the use of a significant amount of memory and additional costs when switching to the IPv6 version. McAuley, A. and Francis, P., Fast Routing Table Lookup Using CAMs, In Proceedings of IEEE INFOCOM '93, Vol.3, pages 1382-1391, 1993, proposed a scheme using CAM (associative memory), but from Due to the high cost of CAM, this option is currently not being considered. Huang, N. and Zhao, S., in A Novel IP-Routing Lookup Scheme and Hardware Architecture for Multigigabit Switching Routers, IEEE Journal on Selected Areas in Communications, Vol. 17, No. 6, pages 1093-1104, June, 1999, an indirect search algorithm with access to pipelined memory is proposed to reduce the number of memory accesses, although this option is worse than switching to IPv6.

SUMMARY OF THE INVENTION

Thus, it is an object of the present invention to provide a method for performing high-speed IP route lookups and managing routing / forwarding tables capable of minimizing memory access time using a randomized algorithm, thereby ensuring cost-effective memory utilization.

It is also an object of the present invention to provide a method for high-speed IP route lookups and routing / forwarding table management that makes it easy to change routes and control the operations of adding and removing routes.

In addition, it is an object of the present invention to provide a method for performing high-speed searches of IP routes and managing routing / forwarding tables, which will allow efficiently building a routing table and / or forwarding table without sorting suitable routes according to the length of the route prefix entered during the table building process.

According to a first aspect of the present invention, there is provided a method for finding an IP address having a prefix length range as a key, using a skip list containing a header node that has a key with a predetermined maximum value, and a plurality of nodes (subnodes), each of which has a key fixed (or variable) range, predefined in descending order, and saves route entries corresponding to prefix lengths in hash tables associated with corresponding prefix lengths .

According to a second aspect of the present invention, a method for creating an IP routing table using a skip list includes: creating a header node having a maximum level for managing each node in the skip list; Insert a plurality of nodes having a selected prefix range into the skip list; and creating a hash table to save the routing entries specified in each node corresponding to the prefix range.

According to a third aspect of the present invention, there is provided a method for updating a routing table using a skip list where routing entries are stored in a hash table in accordance with the prefix length set in each node configured in accordance with the IP address prefix length range. The method includes: finding a node in which a prefix range is set corresponding to the prefix length of the updated route; search in a hash table having the same prefix length as the updated route in the found node; and updating the corresponding route in the hash table when the hash table is found.

According to a fourth aspect of the present invention, there is provided a method for finding a route from a routing table using a skip list, where the route entries are stored in a hash table in accordance with predefined prefix lengths at each node formed in accordance with the distribution of the IP address prefix range. The method includes: finding a neighboring node, starting from the first node of the skip list; comparing the destination address with the corresponding hash tables in the corresponding node; and accepting the matching prefix as the longest prefix when the hash table contains the destination address.

Brief Description of the Drawings

The present invention and its inherent advantages are explained in the following detailed description, illustrated by drawings, in which the same reference numerals indicate the same or similar components and which show the following:

1 is a diagram illustrating a distributed router architecture for which the present invention is applicable;

2 is a diagram illustrating a parallel architecture of a router for which the present invention is applicable;

3 is a diagram illustrating the structure of a forwarding mechanism for which the present invention is applicable;

4 is a diagram illustrating the structure of a routing processor;

5 is a diagram illustrating a longest prefix matching algorithm (LPM) according to an embodiment of the present invention;

6 is a diagram illustrating the structure of a skip list according to an embodiment of the present invention;

figures 7A and 7B are traffic illustrating the measured values of the distribution of the length of the prefix on the Internet;

8 is a diagram illustrating a structure of a header node according to an embodiment of the present invention;

9 is a diagram explaining an operation for constructing a skip list according to an embodiment of the present invention;

Figures 10A and 10B are diagrams illustrating an operation of constructing a routing table depending on prefix length values according to an embodiment of the present invention;

11 is a flowchart illustrating a procedure for creating a routing table according to an embodiment of the present invention;

12 is a flowchart illustrating a routing table update procedure according to an embodiment of the present invention;

Figures 13 and 14 are diagrams illustrating skip list structures for explaining search operations and updating routes according to an embodiment of the present invention;

15 is a flowchart illustrating a route search process according to an embodiment of the present invention; and

FIG. 16 is a diagram illustrating a structure of a skip list for explaining a complete search process according to an embodiment of the present invention.

Detailed Description of a Preferred Embodiment

Below, with reference to the drawings, a preferred embodiment of the present invention is described. In the following description, well-known functions and constructions are not described in detail so as not to obscure the invention with unnecessary details.

First, a router and a routing method are described to which the present invention is applicable. Chan et al. (A Framework for Optimizing the Cost and Performance of Next-Generation IP Routers, IEEE Journal of Selected Areas, in Communications, Vol. 17, No.6, pages 1013-1029, June, 1999) classify two The main architecture options for a high-speed router are: distributed architecture and parallel architecture, the main difference between them being determined by the location of the forwarding processor. The development trend of the structure of the high-speed backbone router is rather directed towards a distributed rather than a well-known centralized architecture.

Figure 1 shows a router with a distributed architecture to which the present invention is applicable. 1, a router includes: a line module of a board 110 provided with an FE forwarder through which packet input and output are performed; a routing processor 120 for constructing an initial RT routing table and managing a routing table; and a switching structure 130 used when switching packets to a specific port in a router.

Routing processor 120 comprises an RT routing table updated by reflecting the most recently changed routes therein. The routing table RT is formed on the basis of routing protocols such as RIP (routing information protocol), OSPF (first discovery of the shortest routes) or BGP-4 (border Internet protocol 4), but the list of protocols is not limited to those indicated here. From the routing table in routing processor 120, a compact table is called a FT forwarding table, which is designed to be efficiently searched. FT forwarding table is designed for efficient search, which is achieved by reducing the efficiency of route additions and deletions.

If a packet arriving from the line module of the board 110 cannot find the path to its destination from the FT forwarding table, then the corresponding packet must go through the switching structure 130 to the routing processor 120 to solve the problem of an inconsistent route. After finding the destination, the packet must again be routed to the switching structure 130 for forwarding to the output line module of the board 110. Otherwise, if the routing processor 120 cannot find the path to the destination even using the routing table RT, then such a packet is in the processor routing 120 is discarded.

Asthana, A., Delph, C., Jagadish, H., and Krzyzanowski, P., Towards a Gigabit IP Router, J. High Speed Network, vol. 1, no. 4, pages 281-288, 1992 and Partridge et al. (A 50-Gb / s IP Router, IEEE / ACM Trans. On Networking, vol. 6, no. 3, pages 237-248, 1998) a parallel router is disclosed architecture. FIG. 2 shows a parallel architecture router to which the present invention is applicable. In the parallel architecture router shown in FIG. 2, each forwarding processor 112 with the FT forwarding table is separate from the line module of the board 110. In the specified parallel architecture, a client-server model is used in each forwarding mechanism to find routes and parallel batch processing.

The routing table in routing processor 120 should be designed so that it is able to quickly reflect changes in routes and easily maintain them. In addition, the route to be added or removed should be reflected in the FT forwarding table to the forwarding processor 112 as fast as possible, and at minimal cost. When the route is not reflected, the incoming packet is sent to the routing processor 120 through the switching structure 130, which leads to an increase in transmission delay due to the passage along the additional route required when processing the corresponding packet.

Although the invention has been described with reference to a distributed router and a parallel router, it will be apparent to those skilled in the art that the invention can be used in a router having other types of architecture.

Figure 3 shows the structure of the forwarding processor 112 in a router with a distributed or parallel architecture. Figure 4 shows the structure of the routing processor 120.

3, the forwarding processor 112 includes n input interfaces, m output interfaces, n buffers for buffering packets input through the input interfaces, a packet classifier for classifying packets stored in the buffer according to classes or types, and a route search controller for transmitting packets classified using the forwarding table and packet classifier to the corresponding output interfaces by accessing the forwarding table.

According to FIG. 4, routing processor 120 includes n input / output interfaces for receiving and issuing packets from / to the switching structure 130 shown in FIG. 1 and 2, a switching interface for buffering packets input and output via input / output interfaces, and for ensuring interaction between the switching structure 130 and the controller for finding routes and services in the next step, the controller for searching routes and services forwarding packets from routing tables and the switching interface to the corresponding input / output interface through the switching interface by accessing the routing table.

According to the RFC standard, the routing (or forwarding) table is not limited to either structures, such as matrix, tree, and trie structures, which are mainly used for searching, or the algorithms used to process them. However, matching the longest prefix requires a minimum number of memory accesses, and changed routes should be easily updated in the routing table (or forwarders).

To this end, a randomized algorithm is used in this embodiment of the present invention. A randomized algorithm is an algorithm that implements random selection during its execution. The algorithm is focused on establishing the probability of the state of each input. That is, the processing of specific input data is not based on a specific hypothesis, but on the probability P (for example, based on a coin toss). The advantages of using a randomized algorithm are its simplicity and efficiency.

Using an algorithm according to the present invention based on a randomized algorithm guarantees a minimum number of memory accesses and simple routing table management. To manage the route records, only information about the change in the route is reconstructed, in contrast to the known method according to which the entire table is reconstructed. In addition, unlike a route search algorithm based on a common tree or binary trie structure, the algorithm according to the present invention searches for the longest prefix from the tail of the destination IP address, as shown in FIG.

In addition, a skip list for a routing table is used in the present invention (See Pugh, W, Skip Lists: A Probabilistic Alternatives to Balanced Trees, CACM 33 (6), pages 6689-676, 1990).

The skip list is considered as a linked list with n sorted nodes and is usually used in a data structure such as a balanced tree, such as a skew tree (See Sleator, D. and Tarjan, R., Self-Adjusting Binary Search Trees, JACM, Vol. 32, No. 3, July, 1985). This approach is simpler and more efficient than using a balanced tree in terms of input and delete operations. The present invention provides a variant of the skip list for matching by the longest prefix, as shown in FIG. 6.

6 shows an embodiment of a skip list according to an embodiment of the present invention. 6, each node has one pointer or multiple pointers, and the leftmost node is called the header node 210, since all operations according to this option of the skip list, such as searching, inserting, deleting and updating, must start from the header node 210.

The header node 210 as the key value contains + ∞ and a pointer (s) to point to the next node (s). Keys at nodes 220, 230, 240, 250, and 260 are sorted in descending order. The key in each node means “prefix length range” or “clustering of the prefix length value”, this key being entered in descending order. In this embodiment of the invention, the search begins from “leaves” to “root” (or “parent”), that is, in the reverse sequence of key values, which is fundamentally different from the method based on known trie-structures or structures such as a tree.

The key of the second node 220 has a 32-24 prefix length range, and any route having a prefix length belonging to this prefix length range is stored in a hash table corresponding to its prefix length. FIG. 6 shows a hash table 222 with a prefix length of 30 bits and a hash table 224 with a prefix length of 26 bits as an example of hash tables of a route belonging to the range of the second node 220. In addition, FIG. 6 shows a hash table 242 with a prefix length of 15 bits as an example of a hash table of a route belonging to a range of the fourth node 240 with a range of prefix lengths of 17-12.

Although Fig. 6 shows an example where the prefix length ranges of different nodes are tied to different lengths, that is, an example of a fixed-division prefix range, a group of nodes with a variable length can also be distinguished in the prefix length range, i.e., get a prefix range with a variable ( variable) separation. Srinivasan, V. and Varghese, G., "Faster IP Lookups using Controlled Prefix Expansion, In Proceedings of ACM sigmetrics '98 Conf., Pp. 1-11, 1998, discuss a dynamic programming method for splitting a prefix range for optimization In the algorithm according to the present invention, a group of prefix lengths can be distinguished using a similar method.However, in the embodiment of the present invention shown in Fig. 6, it is preferable to have a fixed range of prefix lengths, which avoids the additional use of computing resources in due to dynamic programming.

In the case of using IPv4, if the prefix length range is fixed and equal to 8, then the total number of input nodes can be 5 (1+ [32/8] (where 32 is the maximum number of bits in an IPv4 address)). As can be seen from the empirical results obtained from IPMA (Measurement and Analysis System for the Internet) (http://nic.merit.edu/ipma), the distribution of the actual base length of the prefix is asymmetric, that is, it assumes the existence of local features in the distribution of the base prefix lengths. Therefore, in the present invention, fewer nodes can be used. For example, FIGS. 7A and 7B show that prefix lengths of 17 to 27 are most often accessed. Accordingly, you can install one node with the key value of the prefix length range, including the corresponding prefix length, and set up many nodes corresponding to another prefix length range.

6, the variable 'MaxLevel' of the header node 210 represents the maximum level among all nodes belonging to the skip list, that is, the total number of pointers in the header node pointing to other nodes, with each node supporting the key value 'x → key' (or key (x)) and pointer (s) 'x → forward [L]' (or next (xL)) to indicate a pointer from node x to point to a neighboring node at level L of node x.

Each key in the node, except the header node 210, has a range or set of prefix lengths. 6, the node 220 has routing entries ranging from 32 to 24, with each prefix length value having its own hash table for storing routing entries matching the corresponding prefix length. If there are route entries corresponding to each prefix length value in the range from 32 to 24 inclusive, then the corresponding node will have nine (9) pointers pointing to nine (9) different hash tables. Each hash table only stores route entries that exactly match the length of the prefix. For this reason, this version of the skip list is called the skip list with a k-hashed node (s).

The skip list algorithm of the present invention is explained in the following description of a method for creating a header node and constructing a skip list.

8 is a diagram for explaining a method for creating a header node according to an embodiment of the present invention. According to Fig. 8, the header node has the key value + ∞, a MaxLevel = 5, (1+ [32/8]), where MaxLevel = 1 + [W / n], W is the IP address length in bits, for example, for IPv4 - 32, and for IPv6 - 128; an is the prefix range. Since the header node must cover all nodes exclusively within the skip list, if IPv4 is used, MaxLevel = 33 is suitable for data structures containing up to 2 ³² routes, if the probability p is 1/2 and the prefix range is 1. The header node has MaxLevel direct pointers indexed from 0 to MaxLevel-1. In this embodiment of the present invention, due to the use of clustering of the prefix range, MaxLevel = 5 is required under the assumption that the fixed prefix range is 8.

After creating the header node, a variant of the skip list is constructed, as shown in FIG. 9, where the operation of constructing the skip list according to an embodiment of the present invention is disclosed. If the top level pointer in the header node is zero, that is, if there are no neighboring nodes, or, in other words, subnodes connected to the pointers of the header node, then the level is reduced by 1 until the pointer points to the neighboring node. As shown in FIG. 9, a first node (a) is created with a prefix range value of 32-24, where a level 0 pointer in the header node points to the first node. After that, a second node (b) with a prefix range of 23-18 and a third node (c) with a prefix range of 17-12 are sequentially created. During this process, many nodes are created covering each range of prefix lengths.

The level of each created node is set randomly using the randomized algorithm shown below in Program 3 according to the present invention. First, the value of a random real number in the interval (0,1) is formed. If the generated value of the actual random number is less than the predefined reference probability value (here the reference probability value is set to p = 1/2 to create an ideal list of passes), and the level of the corresponding created node is also lower than MaxLevel, then the level of the corresponding node is increased. Otherwise, a node is created that has a value of level 0.

After building nodes with prefix ranges in the skip list, you need to build a routing table containing route entries. Next, with reference to figa describes a method of constructing a routing table in the case when the router has routes with a prefix length of 30.

10A, if the router has routes with a prefix length of 30, then a search is performed in the first node 220 with a range of prefix lengths of 32-24, and then the added route entries are entered as a hash table. A node in the skip list may have different hash tables depending on the range of prefix lengths. According to this procedure, routes with a prefix length of 15 are added, as shown in FIG. 10B, where a hash table with a prefix length of 15 is created under node 240 with a prefix range of 17-12.

The sequence of procedures for creating a routing table according to a variant implementation of the present invention is presented in Fig.11. In step 11a, a header node with MaxLevel is created to cover all nodes in the skip list. Then, in step 11b, nodes with a divided prefix range are inserted into the list. Although for simplicity, the invention contemplates the use of fixed separation of the prefix range, adaptive separation reflecting the actual distribution of the length of the prefix may also be allowed. After that, in step 11c, a hash table is created for each given route record under the node with the corresponding prefix range. The new scheme according to the present invention provides a search starting from routes having a maximum prefix length, as opposed to a search starting from the shortest routes, as is the case in known trie structures and tree structures.

A router often receives route changes from a router (s) connected to it. These changes should be reflected in the current routing table (or forwarding). The sequence of update steps for the new scheme is presented in Fig. 12. For this changed route (prefix) reflected in the routing table, in step 12a, the node corresponding to the given prefix length should be discarded. Next, in step 12b, a search is made in the hash table corresponding to the prefix length equal to the prefix length of the changed route. If the corresponding hash table is found in step 12c, then proceed to step 12d, where the dropped route in the corresponding hash table is updated or deleted. If it turns out at step 12c that the desired hash table does not exist, then the next step is performed at step 12e. If the route is to be added, then go to step 12f, where a hash table is created having a prefix length equal to the prefix length of the changed route. Then, in step 12d, the changed route is entered into the newly created hash table. Otherwise, the changed route in step 12h is discarded.

13 and 14 are diagrams illustrating routing and updating routes according to an embodiment of the present invention.

With reference to FIG. 13, a route search procedure with a prefix length of 15 is described below. It is necessary to find a node with the same prefix length (15). The search begins at the top level (i.e., level 4) of the header node 210 (sequence (1) in FIG. 13). The level 4 pointer points to the next node having level 4, that is, node 260 with the range (key) of the length of the prefix 8-1 (sequence (2) in FIG. 13). After that, the route prefix length is compared with the key of the node 200. The key value does not match, and the search returns to the same level (that is, level 4) of the header node 210 (sequence (3) in Fig. 13). The level of the header node 210 decreases by one (sequence (4) in FIG. 13), and the next level pointer (level 3) points to the next node having level 3, that is, node 240 with key 17-12 (sequence (5) on Fig.13). When a matching key value is detected in the node 240 (sequence (6) in FIG. 13), a hash table is searched corresponding to the prefix length of the changed route (sequence (7) in FIG. 13). If a hash table exists, then the next transition is performed (sequence (8) in FIG. 13). If the hash table does not exist, then to add a new route, it must be created.

On Fig shows several sequences where the route with the prefix '111010 ... 1' with the next transit interface '10' should be removed from the hash table, or replaced with '111011 ... 1' with the following transit '1 '. The search procedure for the corresponding prefix (111010 ... 1) is performed in the same manner as shown in Fig. 13.

With reference to FIGS. 15 and 16, a route search procedure for finding a route to a specific destination IP address based on a routing table is described below.

FIGS. 15 and 16 show a route search procedure defined as a match scheme for the longest prefix according to an embodiment of the present invention. A search based on the matching scheme for the longest prefix means that the route record should be compared by the longest prefix, for example, with the 32 prefix length in IPv4. 16 is a diagram explaining the entire search scheme according to an embodiment of the present invention, where the destination address has a prefix length of 15 '1110101 ... 1'.

In step 15a (FIG. 15), for a given destination address enclosed in the IP header, the first node in the skip list is scanned, thereby starting to view the route specified by the destination address (sequence (1) in FIG. 16) from level 4 of the header node 210 and the level in the header node 210 is reduced by one to level 0 (sequence (2) in FIG. 16), and level 0 has a pointer pointing to the first node 220 (sequence (3) in FIG. 16).

Then, in step 15b, a hash search is performed on the first node 220, starting with the top value of the prefix length range key, for example 32 (current prefix length). If the cache table does not exist, the procedure goes to step 15g. Otherwise, when the hash table in step 15c exists, it proceeds to step 15d. In step 15d, a search is performed among the prefixes in the found hash table of the prefix corresponding to the destination address. If it turns out that the found prefix matches the destination address (step 15e), then the matching prefix is returned as the longest prefix in step 15f. At the same time, if the prefix did not match at step 15e, the procedure continues at step 15g.

In step 15g, the following hash table is detected in the same prefix range. If the next hash table in step 15h exists, then the procedure returns to step 15b to repeat the above process. If the next hash table does not exist in step 15h, then proceed to step 15i, where they go to the neighboring node in the skip list, and then the procedure returns to step 15b to repeat the above process in the neighboring node. As described above, in contrast to the well-known search scheme starting from the root node, the new scheme is based on a search starting from the longest range of prefix with its hash table.

According to FIG. 16, a search is first performed in the skip list of the node, as described previously, and the search for the route specified by the destination address begins with the header node (sequence (1) in FIG. 16). The level in the header node 210 decreases to level 0 (sequence (2) in FIG. 16), where the pointer points to the node having the longest prefix range, and then proceed to the first node 220 indicated by the pointer of the header node 210 (sequence (3) in Fig.16). A search is then performed in the first node 220 to find a hash table containing 32 prefixes. If a matching table is found, then the destination address is compared with the prefixes in the hash table in order to find a match. If no hash table containing 32 prefixes was found in the first node 220, then, since each node in the skip list supports a pointer matrix that points to each hash table in the range of node prefix, then in the absence of a hash table containing prefixes length 32, the corresponding pointer will be zero.

In the example of FIG. 16, the first node 220 has only two hash tables 222 and 224 corresponding to the length of the prefix 30 and the length of the prefix 26, respectively. Therefore, when a search is performed in the first node 220 to find the hash table, each pointer in the prefix range is checked, and pointers with a zero value are skipped. Hash tables corresponding to pointers with a non-zero value are compared with the source destination address to find a match.

Thus, the first pointer with a non-zero value will point to the hash table 222, and the destination address is compared with the prefixes in the hash table 222 (sequence (4) in FIG. 16). If the destination address in the hash table 222 is not found, then the next hash table with a non-zero pointer is compared with the destination address, that is, the hash table 224 with the prefix length 26 (sequence (5) in FIG. 16). If the destination address is still not found and there are no more hash tables in the first node 220, then the search for the node 220 is unsuccessful.

If the search for node 220 is unsuccessful, then a pointer is searched pointing to the next node in the skip list, having the next longest prefix range. In this example, there is a pointer showing that the next node in the skip list with the second longest prefix range is node 230 (sequence (6) in FIG. 16). Accordingly, the node 230 is checked to find out if it has any hash tables, and since they do not exist, a pointer is searched in the node 230 pointing to the next node in the skip list.

In this case, a transition is made to the neighboring node 240 in the skip list (sequence (7) in Fig. 16), and a pointer is found pointing to the hash table 242, in accordance with the steps shown in Fig. 15. The destination address is compared with the prefixes in the hash table 242 having a length of 15 (sequence (8) in FIG. 16). If it matches, the route search operation completes successfully (sequence (9) in FIG. 16). If the destination address is not found even in the last hash table of the last node 260, then the search fails, and the packet with the corresponding address is sent along the route along with the default route record.

Skip list structures are presented in Programs 1, 2, and 3; procedures for reflecting changes in route entries in the pass list are shown in Programs 4, 5, 6 and 7; and the main part of the new route search algorithm is shown in Program 8.

Program 1: Build Operation

var px: bitstring init b'0 '; (*prefix*)

interface: integer init 0; (* next transit interface *)

W: integer init 32; (* # address bits, IPv4 *]

k: integer init 8; (* fixed prefix range, for example 8 or a variable prefix range from 0 to W "*)

hrange: integer init W; (* upper bound of the prefix range *)

Irange: integer-init W-k + 1; (* lower limit of the prefix range *)

MaxLevel: integer init [lg (W / k)]; (* under the assumption that p = 1/2 (where p is the probability that determines the possibility of creating a node with higher levels) *)

Buildup (list) (* building a skip list *)

begin

L: = [W / k] (* L is an integer equal to the limit value of W / k and the closest to the number of nodes created *);

while L≥0 do

Insert_Node [list, hrange, Irange);

L: = L-1

hrange: = - Irange-1; - -

Irange: == Irange-k + 1;

end while

end

Program 2: Insert a Node into the Skip List

Insert-Node (list, hrange, Irange) (* insert node *)

begin

update [0..MaxLevel];

x: = list≥ → header

for i: = list → level downto 0 do

while [hrange, Irange] ∈x → forward [i] → key do

x: = x → forward [i]

end • while

update [i]: = x;

end for

x: = x → forward [0];

if [hrange, Irange} ex → key then

x → value: = [hrange, Irange];

else

v: = RandomLevel ()

if v> list → level then

for i: == list → level + 1 to v do

update [i]: = list → header;

end for

list → level: = v,

end if

x: = MakeNode (y, [hrange, Irange});

for i: = 0 to level do

x → forward [i]: = update [i] → forward [i];

update [i] → forward [i]: = x;

end for

end if

end

Program 3: Level of Randomization

RandomLevel () (* node level randomization *)

begin

v: = 0;

r: = Random (); (* r∈ [0, 1) *)

while r <p and v <MaxLevel do

v. = v + l; - -

end while

return v;

end

Program 4: Search Operation

Search (list, px) (* search for a node with the px prefix in the skip list *)

begin

x: = list → header;

for i: = list → -level downto 0 do

while px∈x → forward [i] → key do

x: = x → forward [i];

end while

end for

x: = x → forward [0];

if x → key = px then

for j:: = hrange downto Irange-1 do

if hash-get (j, px);

return hash-get (j, px);

end if

end for

return error

end if

end

Program 5: Paste Operation

Insert (list, prefix, interface) (* insert route record *)

begin

if hash-insert (list, prefix, interface] then

else

return error;

end if

end

Program 6: Delete Operation

Delete [list, prefix, interface) (* delete

route recording *)

begin

if search (list, prefix) then - -

hash-delete (prefix, interface);

else

return error;

end if

end

Program 7: Update Route Recording

Update (list, prefix, interface) (* route update

records *)

begin

if search (list, prefix) then

hash-update (prefix, interface);

else

return error;

end if

end

Program 8: Search Operation

Lookup (list, destaddr) (* search for the route using the longest prefix *)

begin

x: = list → header;

x: = x → forward [0];

for i: = [W / k] downto 0 do

for j: = hrange downto lrange-1 do

if compare (destaddr, hash-get (j, destaddr)) then

return hash-get (j, destaddr);

end if

end for

x: = x → forward [i];

end for

return error;

end

Below with reference to the drawings, the effectiveness of the routing method according to the present invention is evaluated. On figa and 7B depicts graphs representing the distribution of the measured values of the length of the prefix on the Internet. The data represented by the graphs in FIGS. 7A and 7B are collected in the MAE-EAST NAP (Network Access Point) as part of the IPMA (Measurement and Analysis of Internet Performance) project.

Statistical data collected in the IPMA project is widely used as input for comparative analysis of search results performed by developers of route search algorithms. Currently, the router in the MAE-EAST NAP contains the maximum number of route entries and is mainly used as a model of a trunk router. PAIX NAP contains the minimum number of routing entries collected in the current IPMA project and is used primarily as an enterprise router model.

On figa and 7B depicts local diagrams showing that the most often repeated values of the length of the prefix, lying in the range 16-28. Thus, the observational data show that the number of nodes that should be in the modified skip list proposed in the invention can be significantly reduced. Another interesting point in IPMA (http://nic.merit.edu/ipma) is that the number of routing prefixes added or removed in accordance with the BGP protocol (Border Gateway Protocol) per day reaches 1899851 (as for November 1 1997). This means that 25 prefix updates per second are required, and the quick reflection of each changed route prefix in the routing table (or forwarding) becomes an important factor affecting the search time.

With reference to table 2 below is a comparison of the new algorithm with known algorithms. In the general case, when evaluating an algorithm for the worst case, the time required to process the corresponding algorithm or the required memory size is compared using the record "large O" (which determines the execution time of the algorithm). Table 2 gives an estimate of the complexity of the new algorithm according to the present invention, where there are N routing entries, W address bits are required to express each address, and k is a constant dependent on the algorithm.

In Table 2, to avoid confusion with the other base 10 logarithms, “lg” stands for base 2 logarithm.

In the case of IPv4, W corresponds to 32, and for IPv6 it corresponds to 128. In MAE-EAST and PAIX NAP, the number of N route entries is approximately 50,000 and 6,000, respectively. In this case, k in the skip list with k-hashed nodes is 4 if the prefix range is 8. In the case of IPv6, it will be 16.

The new algorithm is better than the well-known algorithm based on the search tree and trie structures (Patricia trie structure or LC trie structure) as applied to routing table, insert and delete operations. In addition, the new algorithm of the present invention is superior to the hash table based binary search algorithm (Waldvogel, M., Varghese, G., Turner, J. and Planner, B., Scalable High Speed IP Routing Lookups, In Proceedings of ACM SIGCOMM '97, Cannes, France, pages 25-37, 1997) and an algorithm based on the full prefix trie structure (Tzeng, H. and Pryzygienda, T., On Fast Address-Lookup Algorithms, IEEE Journal on Selected Areas in Communications, Vol . 17, No. 6, pages 1067-1082, June, 1999) in terms of construction time and insert or delete operation. Table 2 shows that the results of the complexity assessment show the superiority of the new algorithm over the algorithm based on a clean list of passes.

As described above, the algorithm according to the present invention creates nodes in accordance with the IP address prefix range using a skip list and a hash table, which constitutes a randomized algorithm. In addition, the present invention provides a routing method using a skip list in which route entries are stored in a hash table in accordance with the prefix length set at each node, thereby minimizing the route search time and providing efficient management of the corresponding table. In this way, you can perform high-speed route searches. This algorithm can also be applied to the IPv6 version that will be used in the next generation router.

From the results of the complexity assessment it follows that the new algorithm is superior to other known algorithms in any aspect, especially in terms of the operations of constructing a routing table (or forwarding), as well as insert and delete. In addition, given the current trend in the development of modern architectures, such as a distributed router architecture or a parallel router architecture, where the forwarding table in the forwarding processor is separated from the routing table in the routing processor, only updated routes, instead of all route entries, are sent to the router along with the corresponding paths You can unload the memory or resolve conflicts in the system buses. Reducing the search time and maintaining consistent routing information across all related routing tables can help improve other metrics, such as Quality of Service (QoS), Multicast, IPsec, and more.

Although the invention has been demonstrated and described with reference to a particular preferred embodiment, it will be apparent to those skilled in the art that various changes may be made in form and in detail within the spirit and scope of the invention as defined in the appended claims.

Claims (36)

1. A method of generating routing / forwarding tables to search for an Internet protocol address (IP address) using a skip list, including the steps of dividing the IP address prefix length range in a predetermined manner, creating a header node having a maximum level based on a number of clusters allocated in the range of prefix lengths, with the header node indicating each node in the skip list, and creating subnodes using a number of selected clusters, each of the subnodes having a selected range of lin prefix. 2. The method according to claim 1, characterized in that it further includes storing route entries corresponding to the corresponding prefix lengths in the corresponding range of prefix lengths in the hash tables provided in each subnode in accordance with the prefix lengths. 3. The method according to claim 2, characterized in that the route entries contain 32 bits or 128 bits. 4. The method according to claim 1, characterized in that the level of the header node of the sub-node is set randomly. 5. The method according to claim 1, characterized in that the range of prefix lengths is divided in a fixed or variable manner. 6. The method according to claim 1, characterized in that the range of prefix lengths is divided in such a way that each range of prefix lengths is covered by subnodes. 7. The method according to claim 1, characterized in that the routing / forwarding tables are contained in the routing processor and the forwarding processor, respectively. 8. A method of generating IP routing / forwarding tables using the skip list, including the steps of creating a header node for specifying each node to process each node in the skip list, creating a plurality of subnodes having as key a range of IP address prefixes, divided by a predetermined method, and creating hash tables in accordance with the prefix lengths for storing route entries in accordance with the prefix lengths in the corresponding range of prefix lengths in each subnode. 9. The method according to claim 8, characterized in that it further includes storing route entries matching the corresponding prefix in the hash tables. 10. The method according to claim 8, characterized in that the level of the header node of the sub-node is set randomly. 11. The method according to claim 8, characterized in that the range of prefix lengths is divided in a fixed or variable manner. 12. The method according to claim 8, characterized in that the range of prefix lengths is divided in such a way that each range of prefix lengths must be covered by a plurality of subnodes. 13. The method according to claim 8, characterized in that the route entries contain 32 or 128 bits. 14. The method of claim 8, wherein the header node has a key value + ∞ and a direct pointer (s) indexed from 0 to the maximum level value minus one. 15. The method according to claim 8, characterized in that the routing / forwarding tables are in the routing processor and the forwarding processor, respectively. 16. A method of finding a route in routing / forwarding tables using a skip list, in which the route entries are stored in a hash table in accordance with the prefix length set in each node created in accordance with the distribution of the IP address prefix range, the method including the steps of finding the node in which the prefix range corresponding to the prefix length of the desired route is set, finding the hash table with the same prefix length as the desired route in the found node, and finding the desired ma routes from the found hash table. 17. The method according to clause 16, wherein the step of finding the node includes finding a node to indicate the skip list header from a maximum node level, comparing the prefix length range of the found node with the prefix length of the desired route, and finding the node indicated at the next level of the header node when the prefix length of the desired route does not match the prefix length range of the found node. 18. The method according to clause 16, wherein the skip list includes a header node and a plurality of nodes, each of which has a key in a range predefined in descending order, and stores the route entries corresponding to the corresponding prefix lengths in hash tables, associated with the corresponding prefix lengths. 19. The method according to clause 16, wherein the routing / forwarding tables are in the routing processor and the forwarding processor, respectively. 20. The method according to clause 16, wherein the route entries contain 32 bits or 128 bits. 21. A method for updating routing / forwarding tables using a skip list, in which the route entries are stored as a hash table in accordance with the prefix length set in each node created in accordance with the IP address prefix length range, the method including the steps of finding the node in which the prefix range corresponding to the prefix length of the updated route is set, search in the hash table having the same prefix length as the updated route in the found node, and updates the corresponding ma the route in the hash table when the hash table is found. 22. The method according to item 21, wherein the step of finding the node includes finding a node to indicate the skip list header from a maximum node level, comparing the prefix length range of the found node with the prefix length of the updated route, and finding the node indicated at the next level of the header node when the prefix length of the updated route does not match the prefix length range of the found node. 23. The method according to item 21, wherein the step of updating the route includes adding, changing or deleting the updated route. 24. The method according to item 21, characterized in that it further includes the steps of creating a hash table having the same prefix as the updated route when the hash table is not found, and insert the updated route into the created hash table. 25. The method according to item 21, wherein the skip list includes a header node and a plurality of nodes, each of which has a key in a range predetermined in descending order, and stores the route entries corresponding to the corresponding prefix lengths in hash tables, associated with the corresponding prefix lengths. 26. The method according to item 21, wherein the routing / forwarding tables are in the routing processor and the forwarding processor, respectively. 27. The method according to item 21, wherein the route entries contain 32 bits or 128 bits. 28. A method for finding a route in routing / forwarding tables using a skip list in which route entries are stored in a hash table according to predefined prefix lengths at each node created in accordance with the distribution of the IP address prefix range, the method including the steps of finding a neighboring node, starting from the first node of the skip list, comparing the destination address with the corresponding hash tables in the corresponding node, and accepting the matching prefix as the longest The prefix, when the hash table contains the destination address. 29. The route search method according to claim 28, characterized in that it further includes finding the next node in the skip list when the hash table does not contain the destination IP address. 30. The route search method according to claim 28, wherein the skip list includes a header node and a plurality of nodes, each of which has a key in a range predetermined in descending order, and stores route entries corresponding to the corresponding prefix lengths in the hash tables associated with the corresponding prefix lengths. 31. The route search method according to claim 28, wherein the comparison step includes comparing the destination address with a hash table having the maximum prefix length among the hash tables of the corresponding node, and comparing the next hash table having the next largest prefix length, with the destination address when the destination address is not found in the hash table having the maximum prefix length. 32. The route search method according to claim 28, wherein the routing / forwarding lookup tables are in the routing processor and the forwarding processor, respectively. 33. The route search method according to claim 28, characterized in that the route entries contain 32 bits or 128 bits. 34. An Internet Protocol (IP) router containing at least one linear module of cards, each of which is equipped with a forwarding processor for inputting and outputting packets, the forwarding processor having a forwarding table having a skip list architecture that includes a header node and many nodes, each of which is created in accordance with the IP address prefix range, has a prefix length range as a key and stores route entries in the form of a hash table in accordance with the set prefix length, a switching structure for switching packets between internal ports, and a routing processor that includes a routing table having a skip list architecture that contains a header node and many nodes, each of which is created in accordance with the IP address prefix range, has a prefix length range as a key and stores the route entries in the form of a hash table in accordance with the set prefix length, and controls the entire routing operation of the linear modules of the boards and the switching structure. 35. An Internet Protocol (IP) router containing at least one linear module of cards, provided with a forwarding processor having a forwarding table with a skip list architecture that contains a header node and a plurality of nodes, each of which is created in accordance with the IP address prefix range , has a range of prefix lengths as a key and stores route entries in a hash table in accordance with the set prefix length, a switching structure for switching packets between internal ports, and a processor mar rutizatsii for controlling the overall operation of line modules routing boards and connection structure. 36. The IP router according to claim 35, wherein the routing processor comprises a routing table having a skip list architecture that contains a header node and a plurality of nodes, each of which is created in accordance with the IP address prefix range, has a prefix length range of as a key and stores route entries in the form of a hash table in accordance with the set prefix length.

Images