WO2007104204A1

WO2007104204A1 - A method for managing the router table using the ternary content addressable memory

Info

Publication number: WO2007104204A1
Application number: PCT/CN2006/003750
Authority: WO
Inventors: Junchuan Wang; Qinghua Shao; Wenlong Liu
Original assignee: Zte Corporation
Priority date: 2006-03-15
Filing date: 2006-12-30
Publication date: 2007-09-20
Also published as: CN101039252B; CN101039252A

Abstract

A method for managing the router table using the ternary content addressable memory divides the router table into a plurality of areas according to the prefix length of the routing, it is characterized in that: during updating the routing of the router table, the area space is adjusted dynamically between the area spaces within the router table when the actual area space is not enough or too big. The invention can solve the problem that the ternary content addressable memory implements the management of the router table of the IPv4/IPv6 and provide a simple and efficient method for implementing the update of the routing entry.

Description

FIELD OF THE INVENTION The present invention relates to the field of computer network technologies, and in particular, to a technology for implementing routing table management using TCAM (Ternary Content Addressable Memory), which is mainly applied to Network devices such as routers, switches, firewalls, and intrusion detection systems. BACKGROUND OF THE INVENTION In today's network devices, routing is the most basic function. At present, IPv4 (Internet Protocol version 4) is used mainly. The address is 32-bit. IPv6 (Internet Protocol version 6, Internet Protocol version 6) has also been slowly applied. The address is 128 bits. Implementing route lookup is critical to the performance of network devices. There are two main ways to do this. One is to use software, the most typical is the Trie table algorithm; the other is to use hardware implementation, the most typical is TCAM. Of course, hardware implementation is definitely much faster than software, especially in some high-speed networks, where hardware is used to accelerate.

TCAM is a dedicated triple-content addressable memory that enables fast, massively parallel searches. At the time of the search, all entries in the memory are simultaneously compared with the search key, and the search result is the physical address of the match. In the case of strip matching, each bit of the entry can be 0, 1, or X. If it is X, then the bit does not participate in the comparison. The default is success, otherwise the keyword is compared with the corresponding bit of the entry. The same is successful or otherwise fails. During the comparison process, if a bit does not match, the entry fails to match; if all bits match, then the entry matches successfully.

The inherent characteristics of TCAM hardware make TCAM very suitable for LPM (longest prefix match) lookup, while IPv4/IPv6 routing is searched using LPM, so in high-speed networks such as 10G networks, using TCAM for route lookup is a trend. The normal TCAM search process uses the principle of first matching, that is, if there are multiple matches, then the one that matches the first or the lowest address wins. Therefore, the LPM table must be sorted by the length of the prefix, which causes the existing entries to be moved when the delete entry is inserted, as shown in Figure 1. General TCAM has tens of K or even hundreds of K entries. If the first item is inserted/deleted, and moved in the most simple way, then the number of times the item moves is large and moves. The data associated with the entry, which simply does not meet the routing update requirements of high-speed networks. Paper "Fast The incremental updates on Ternary-CAMs for routing lookups and packet classification (http:〃 www.hoti.org/archive/hoti8papers/018.pdf) proposes two methods for managing TCAM routing tables: The first is PLO-OPT (prefix-length ordering), the other is CAO-OPT (chain-ancestor ordering). Although the PLO-OPT algorithm is simple, the average update efficiency is low, and the CAOJDPT algorithm has high update efficiency, but the implementation complexity is high. SUMMARY OF THE INVENTION The object of the present invention is to solve the problem of routing table management of IPv4/IPv6 implemented by TCAM, and to provide a simple and efficient method for updating routing entries. In order to achieve the above object, the present invention provides a management method for a triple-content addressable memory routing table, which divides a routing table into multiple regions according to a prefix length of a route, where when updating a route in the routing table, When the actual space is insufficient or too large, dynamic adjustment of the area size is performed between the respective area spaces in the routing table. The above-mentioned triple content addressable memory routing table management method, wherein when the route is inserted in the routing table, the method includes the following steps: Step 21: Determine whether the current area to which the route to be inserted belongs has a free space; Step 22 If yes, the route is directly inserted after the last route of the current area, and if not, the area above and below the current area is addressed to whether there is an area with free space; Step 23, if not If yes, the insertion fails. If yes, the current area is enlarged, the area with free space is reduced, and then the route is inserted in the current area. Step 24, incrementing the route count of the current area, and incrementing the route The total route count for the table. The above-mentioned triple content addressable memory routing table management method, wherein if the area having the free space is above the current area, the step 23 specifically includes the following steps: Step 31: From the free space The lower boundary of the area shrinks the space of one entry; step 32, moves the last item of each area between the area i or the next area of the free space to the previous area of the current area to the same Above the first entry of a zone; Step 33: Expand a space of an entry from an upper boundary of the current area, and insert a route at an upper boundary of the current area. The above-mentioned triple content addressable memory routing table management method, wherein, if the area with the free space is below the current area, the step 23 specifically includes the following steps: Step 41, the free space is provided The first entry of the region moves to the end of the last entry, and shrinks the space of an entry from the upper boundary of the free space; step 42, the previous region i of the region with free space or The first entry of each region between the next region of the current region moves below the last entry of the same region; Step 43, expanding the space of an entry from the lower boundary of the current region, and A route is inserted at a lower boundary of the current area. The above-mentioned triple content addressable memory routing table management method, wherein when the route is deleted in the routing table, the method includes the following steps: Step 51: Find a location of the route to be deleted in the TCAM, and calculate the to be deleted. Step 52, determining whether the route to be deleted is the last route of the current area; Step 53, if yes, directly deleting the route to be deleted, if not And moving the last item of the current area to the current deletion item position, and deleting the route to be deleted; Step 54: Decrement the prefix count of the current area, and decrement the total prefix count of the routing table. The above-mentioned triple content addressable memory routing table management method, further comprising the step of performing area shrinking, specifically comprising: Step 61: determining whether the size of the current area exceeds a desired value; Step 62, if not, not performing the area Shrinking, if yes, searching for an area with insufficient space above and below the current area; Step 63, if not, stopping the area contraction, and if so, shrinking the current area, expanding the area with insufficient space . The above-mentioned triple content addressable memory routing table management method, wherein when the area with insufficient space is above the current area, the method specifically includes the following steps: Step 71: Add the current area to the area with insufficient space The first entry of each region between the next region moves to the bottom of the last entry of the same region; Step 72, the space of one entry is reduced at the lower boundary of the current region; Step 73, The lower boundary of an area with insufficient space expands the space of an entry. The above-mentioned triple content addressable memory routing table management method, wherein when the area with insufficient space is below the current area, the method specifically includes the following steps: Step 81: Send the next area of the current area to the The last entry of each region between the insufficient space regions moves to the top of the same region i or the first entry; Step 82, the lower boundary of the current region is reduced by an entry space; Step 83; Expanding the space of an entry from the lower boundary of the insufficient space area. The above-mentioned triple content addressable memory routing table management method, further comprising a prefix distribution self-learning process, specifically comprising: Step 91: routing update to count; Step 92, determining whether the count value reaches the sampling period, and if not, ending the self The learning process, if yes, determines whether the current number of routing entries exceeds the threshold; Step 93, if not, ends the self-learning process, and if so, recalculates each prefix region, the formula is as follows: New expected value of the region = old expected value of the region X ( 1 - δ ) + The ideal area size X δ calculated according to the proportion of the current prefix of the region. δ is a coefficient used to adjust the proportion of the ideal size of the region. The above-mentioned triple content addressable memory routing table management method, wherein, when the routing table is initialized, an expected value size of each routing area is allocated according to a set prefix length distribution probability, and the expected value is initialized and allocated to each area. the same. By adopting the method of the invention, the routing table management problem of UDP AM to implement IPv4/IPv6 can be solved efficiently. Its advantages are as follows:

1. Simple implementation, no complicated algorithms and data structures;

2. The update efficiency is very high: Add the average number of moving entries in the route "0, delete the average number of moving routes -1;

3. With automatic adjustment and recovery function, it can maintain stable update efficiency;

4. With intelligent self-learning function, it can be automatically adjusted according to the actual route distribution;

5. It is possible to extend the management applied to other LPM tables. BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a schematic diagram showing a sequence of arrangement of an LPM table of the present invention in a TCAM; FIG. 2 is a diagram showing a prefix area distribution after initialization of the present invention; FIG. 3 is a flow chart of inserting a route of the present invention; 5 is a flowchart of deleting a route according to the present invention; FIG. 6 is an example of prefix region shrinking according to the present invention; and FIG. 7 is a flowchart of a prefix distribution learning of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described in detail below with reference to the accompanying drawings BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic view showing the arrangement of LPM tables of the present invention in TCAM. 2 is a prefix area i or a distribution map after initialization of the present invention. As shown in the figure: Assuming the route length is N, then the system has N prefix areas, each prefix area has a set of information, as follows: Upper_ boundary upper boundary lower_ boundary lower boundary preflx_num prefix number region— size area i or actual size expected—size area i or expected size global information g_region[N] information of all areas i or, as described above g__prefix_total routing table The total number of prefixes g—start The starting position of the routing table in the TCAM g—the capability routing table, the capacity presets the size of each area according to the previous statistical information, and initializes the area related information and the global information. The allocation ratio for each region is stored in the percentage [N] array. The global information is initialized as follows: g_prefix_total = 0 g— start = the starting position of the routing table in the TCAM g — capability = total capacity of the routing table. Area information The initial initial fc is as follows: g—region[N]. expected— size = g__region[ N]. region— size =

G-capability ^x percentage[N] g—region[N]. upper-boundary = g— start g—region[N]. lower— boundary = g—region[N]. upper— boundary + g_region[N]. region One size-1 G_region [N]. prefix— num = 0 The region I ( Ν > 1 > 0, I from Nl→l ) is initialized as follows: g—region[I]. expected— size = g—region[I]. Region— size =

G— capability xpercentage[I] g—region[I]. upper-boundary = g_region[I+ 1 ] .lower_boundary+ 1 g—region[I] .lower— boundary = g_region[I]. upper— boundary + g_region[I] Region one size - 1 g_region[I]. prefix - num = 0 Figure 3 is a flow chart of the insertion route of the present invention. 4 is an example of a prefix area extension of the present invention. As shown in the figure, the assumed route prefix length is I. The specific steps are as follows:

5301. Determine whether the prefix area where the inserted route is located I has an idle entry, and the judgment method is: g-region[I]. region_size> g-region [I] .prefix-num. ^ If there is free space in the mouth, insert a route at g_region[I].upper_boundary+g_region[I].prefix_num and jump to S311.

5302, the prefix area I has no free space, then you can consider expanding the area. First, determine whether the entire routing table has free space. The judgment method is as follows: g_ capability > g_prefix_total. If the entire routing table has no free space, then the insertion fails, 3 megabytes to S312.

5303, the routing table has free space, and the prefix area can be extended. The prefix area J with free space is searched in the upper and lower areas of the area I in turn, and the judgment conditions are as follows: g-region[J] .region_size> g-region[J], prefix-num The judgment process is an iterative process, in order to make The number of moves is the least. The judgment method is to judge J+l, J-1 first, and then judge J+2, J-2 until a prefix area with free space is found.

5304, if 】< 1, megabyte to step S308. See Figure 4 for an example of regional expansion.

5305, the free area J is above the area I. From the lower boundary of the region J shrinks 'j, the space of an entry: g-region[J] .region a size— G-region[J] .lower— boundary—

5306, each region 区域 between regions J-l to 1+1 (each region is full) is performed as follows (upward shift): g_region[x], upper_ boundary-

Move(g_region[x].lo er_boundary→g_region[x].upper_boundary) , move the last entry of the region to the top of the first entry in the same region. The move condition is g-region[x]. prefix— rmm > 0 g a region[x] .lower— boundary—

5307, expand the space of an entry from the upper boundary of the region i or I: g_region[I].region_size++ g_region[I]. upper-boundary—and insert a route at the g-region[I]. upper— boundary, then ji Mega to S311.

5308, free area: t or J is below area I.

Move(g_region[J].upper_boundary→g__region[J].upper_boundary+

G_region[J].prefix_num) , moves the first entry of the region to the end of the last entry in the region. The move condition is g_region[x]. prefix a num > 0 from the upper bound of region J, the space of an entry: g — region[J]. region— size— g—region[ J]. upper-boundary ++

5309, each region x (each region is full) between the regions J+l to 1-1 is sequentially processed as follows (g): g-region[x]. lower- boundary ++

Move (g a region [x], upper- boundary → g_region [x], lower- boundary), the first entry capture area moves following the last entry in the same region of a moving condition g_ _re gi ₀ n [ x]. Prefix_num > 0. G_region[x] . upper— boundary++

S310: Expand the space of an entry from the lower boundary of the region I: g_region[I].region_size++ g_region[I].lower_boundai*y++ and insert a route at the g-region[I]. lower-boundary.

5311, update prefix count information g_region[I] .preflx_num++ g_prefix__total++

5312, 4 Weng entered the end. Figure 5 is a flow chart of the present invention for deleting a route. Fig. 6 is an example of the prefix region contraction of the present invention. As shown in the figure, the length of the deleted route prefix is I. The specific steps are as follows:

5501, find the location of the route in the TC AM location, calculate the location of the route in the prefix area I rela_loca = location-g-region[I].upper-boundary.

5502. Determine whether the route is the last route in the area. The judgment method is as follows: rela_loca = g_region[I] .prefix_num - 1. If yes, delete the route directly at the location, otherwise move the last entry in the region to the current delete entry location (g_region[I].upper_boundary+g_region[I] .prefix— num- 1—location) „

5503, update the prefix count information: g_region[I] .prefix— num— g_prefix_total~

5504. Determine whether regional shrinkage is required. This is mainly to prevent the regional distribution from deteriorating due to frequent turbulence of the route, thereby reducing the average update efficiency. Proactively adjusting the '!· gray complex prefix area distribution during the update process can effectively ensure that at most one entry is moved when deleting, and the insertion does not need to move the entry. This can maintain a 4 艮 high update efficiency. The method for judging whether or not the region shrinks is required is as follows: g - region[I]. region_ size > g_region[I]. expected one size. : 3⁄4 mouth fruit does not need shell 3 trillion! j S513.

5505, need to shrink the area. In the upper and lower areas of the area I, the area J that satisfies the following conditions is searched sequentially: g-region[J].region_size<g-region[J].expected-size The judgment process is an iterative process, in order to minimize the number of movements. The judgment method is to judge J+l, J-1 first, and then judge J+2, J-2 until a prefix area with insufficient space is found.

5506, if J < I, mega to step S510. See Figure 6 for examples of area shrinkage.

5507, free area J is above area I. Each area X between the areas i or I to J-1 is performed as follows (downward): g-region[x] .lower a boundary ++

Move(g_region[x].upper_boundary—»g_region[x].upper_boundary

+g_region[x].prefix_num) , moves the first entry in the region to the end of the last entry in the region. The move condition is g_region[x]. prefix_num > 0 g_region[x] .upper— boundary ++

5508, shrinking from the lower boundary of the region I, j, the space of an entry: (note that the region I has moved down) g a region[I]. region— size— g_region[I].lo er_boundary~

5509, expand the space of an entry from the lower boundary of zone i or J: g_region[J].region_size++ g-region[J]. lower one boundary ++ then 3 trillion to S513

S510, the free area J is below the area I. Each of the areas X between the areas 1-1 to J is sequentially operated as follows (upward shift): G_region[x] .upper-boundar - -

Move(g_region[x].upper__boundary+g_region[x].prefix_num→g_region[x].u pper_boundai ), move the last entry of the region to the top of the first entry in the same region, and the move condition is g_region[x ]. prefix-num > 0 g-region[x] .lower-boundary- -

S511, shrinking a space from the lower boundary of the region I: g_region[I] .region_size_ g-region[I]. lower_ boundary- -

S512: Expand the space of an entry from the lower boundary of the region J: (note that the region J has been moved up) g_region[J].region__size++ g_region[J] . lower- boundary ++ S513 , the end of the deletion. When the route update is very frequent, then the original set area expected_size value may not be able to adapt to the current prefix distribution. In this case, a dynamic adjustment technique can be used to make the area expected-size update according to a certain period. It can be guaranteed that "^ positive area i or expected_ size reflects the distribution ratio of the actual prefix in the current routing table. This is actually an intelligent routing prefix distribution self-learning function f] because the prefix distribution is only caused when adding a deleted route. Change, so you only need to learn when the route is updated. The most accurate reflection of the current prefix distribution is to learn every time the route is updated, but this CPU performance will consume more than 4 ,, so it is not desirable, this A sampling period sample can be set, and the specific implementation method is as follows: Figure 7 is a flow chart of the prefix distribution learning of the present invention. As shown in the figure, the following steps are included:

S701, the count is updated every time the routing update is performed.

S702, the prefix distribution learning is performed when counter MOD sample = 0. S703, firstly, it is necessary to judge whether the current state is worth learning before learning. Only when g_prefix_total > threadshold, the routing entry exceeds a certain threshold P艮, the learning accuracy is ensured, because the sampling sample is larger, the calculation result is obtained. The more accurate. The method of learning each prefix distribution is as follows: Recalculate the expected_size of each prefix area I (1≤I≤N).

5704, First, assign 1 to I=1.

5705, judge whether I exceeds N, and if it exceeds, end learning.

5706, if I does not exceed N, then calculate according to the following formula: g-region[I]. expected_ size = (( g-region[I]. expected— size÷g_capability ) χ ( 1 — δ ) + ( G_region[I].prefix_num÷g__prefix_total ) χδ ) xg— capability

= g_region[I]. expected— sizex ( 1 — δ ) +

( g_region[I].prefix_num÷g_prefix_total ) χ g— capability χδ

( 0 < δ < 0.5 , recommended to take 0.1 ) ₀

5707, after the calculation is completed, assign 1 = 1+1, and then return to S705. The meaning of the above formula is as follows: The new expected value of the region I = the old expected value of the region I X ( 1 - δ ) + the ideal size χ δ of the region I calculated according to the proportion of the current prefix of the region I. Where δ is a coefficient used to adjust the proportion of the ideal size of the region I. Since the route has been constantly changing, the δ value cannot be large, otherwise the routing table management will always be in a turbulent state. The above is the basic principle of the intelligent prefix distribution learning function. For detailed flowchart, refer to Figure 7. The prefix distribution learning function can be implemented as long as it is called every time the routing is updated. The intelligent prefix distribution learning function is an option that can be enabled. If the routing update is infrequent in the actual application, the function may not be enabled. If the update is frequent, it is recommended to enable the function. The value of the sampling period sample is updated. The frequency is decremented, and the specific value can be determined according to the actual application. The invention may, of course, be embodied in various other embodiments without departing from the spirit and scope of the invention. The corresponding changes and modifications are intended to fall within the scope of the appended claims.

Claims

Claim

A method for managing a routing table of a triple-content addressable memory, which divides a routing table into a plurality of areas according to a prefix length of a route, wherein when the route in the routing table is updated, when the actual space of the area is insufficient or excessive When large, the dynamic adjustment of the area size is performed between the respective area spaces in the routing table.

The method for managing a triple content addressable memory routing table according to claim 1, wherein when the route is inserted in the routing table, the method includes the following steps:

Step 21: Determine whether there is free space in the current area to which the route to be inserted belongs; Step 22, if yes, insert the route directly after the current area i or the last route, and if not, in the current area The other areas above and below address whether there is an area with free space;

Step 23, if not, the insertion fails, if yes, the current area is enlarged, the area with the free space is reduced, and then the route is inserted in the current area;

Step 24: Increment the route count of the current area, and increment the total route count of the routing table.

The method of claim 2, wherein the step 23 includes the following steps:

Step 31: Shrink the space of one entry from the lower boundary of the area with the free space; Step 32, each area between the next area of the area with the free space to the previous area of the current area The last entry is moved to the top of the first entry in the same region;

Step 33: Expand a space of an entry from an upper boundary of the current area, and insert a route at an upper boundary of the current area.

The method for managing a triple content addressable memory routing table according to claim 2, wherein if the area having the free space is below the current area, the step t23 is specifically below the tongue step: Step 41: Move the first entry of the area with free space to the end of the last entry, and shrink the space of an entry from the upper boundary of the area with the free space; Step 42, the idle The first entry of each region between the previous region of the region of the space to the next region of the current region moves below the last entry of the same region;

Step 43: Expand a space of an entry from a lower boundary of the current area, and insert a route at a lower boundary of the current area.

The method for managing a triple content addressable memory routing table according to claim 1, wherein when the route is deleted in the routing table, the method includes the following steps:

Step 51: Find a location of the route to be deleted in the TCAM, and calculate a relative location of the route to be deleted in the current area to which it belongs;

Step 52: Determine whether the route to be deleted is the last route of the current area;

Step 53, if yes, directly deleting the route to be deleted, and if not, moving the last item in the current area to the current deletion item position, and deleting the route to be deleted;

Step 54 decrements the prefix count of the current region and decrements the total prefix count of the routing table.

The method for managing a triple-content addressable memory routing table according to claim 5, wherein the step of shrinking the region is further included:

Step 61: Determine whether the size of the current area exceeds a desired value; Step 62, if not, do not perform area contraction, and if yes, search for an area with insufficient space above and below the current area;

Step 63, if no, stop the area contraction, and if so, shrink the current area to expand the area i or where the space is insufficient.

The method for managing a triple-content addressable memory routing table according to claim 6, wherein when the area with insufficient space is above the current area, the method further includes the following steps:

Step 71, the current area is to the next area of the space-deficient area The first entry of each region between the two moves to the bottom of the last entry of the same region; Step 72, shrinking the space of one entry at the lower boundary of the current region; Step 73, the region with insufficient space The lower boundary expands the space of an entry.

The method for managing a triple-content addressable memory routing table according to claim 6, wherein when the area with insufficient space is below the current area, the method further includes the following steps:

Step 81: Move the last item of each area between the next area of the current area to the area of the insufficient space to the first item of the same area; Step 82, ^! Finding a lower boundary of the current region to reduce the space of one entry; Step 83, expanding an entry space from a lower boundary of the insufficient space.

The method for managing a triple-content addressable memory routing table according to claim 1, further comprising: a prefix distribution self-learning process, specifically comprising:

Step 91, the route is updated into 4 counts;

Step 92: Determine whether the count value reaches the sampling period, and if not, end the self-learning process, and if yes, determine whether the current number of routing entries exceeds a threshold;

Step 93, if no, end the self-learning process, and if so, recalculate each prefix area, the formula is as follows:

The new expected value of the region = the old expected value of the region X ( 1 - δ ) + the ideal size of the region calculated by the proportion of the current prefix of the region χδ ,

δ is a coefficient used to adjust the proportion of the ideal size of the area.

10. The triple content addressable memory routing table management method according to claim 9, wherein, when the routing table is initialized, an expected value size of each routing area is allocated according to a set prefix length distribution probability, and the expected value is initialized. The time is the same as the size assigned to each area.