Mathematical analysis on forwarding information base compression
 131 Downloads
Abstract
With the fast development of Internet, the size of routing table in the backbone router continues to grow rapidly. forwarding information base (FIB), which is derived from routing table, is stored in linecard to conduct routing lookup. Since the linecard’s memory is limited, it would be worthwhile to compress the FIB for consuming less storage. Therefore, various FIB compression algorithms have been proposed. However, there is no wellpresented mathematical support for the feasibility of the FIB compression solution, nor any mathematical derivation to prove the correctness of these algorithms. To address these problems, we propose a universal mathematical method based on the Group theory. By defining a Group representing the longest prefix matching rule, the bound of the worst case of FIB compression solution can be figured out. Furthermore, in order to guarantee the ultimate correctness of FIB compression algorithms, routing table equation test is proposed and implemented to verify the equivalence of the two routing tables before and after compression by traversing the 32bit IP address space.
Keywords
FIB compression Group Trietransformation LPM RTET1 Introduction
The backbone routing table has been growing at an exponential rate, driven mainly by multihoming and the rapid development of mobile communication (Meng et al. 2005). The fast increasing routing table incurs fast increasing FIB. The linecard that holds the FIB usually adopts fast memory, which is expensive and difficult to scale, and it would be worthwhile to improve the memory efficiency by compressing the FIB. Besides, for the routing lookup schemes based on software (Waldvogel et al. 1997; Degermaerk et al. 1997; Nilsson et al. 1998), FIB compression can be used to reduce their memory requirements; for the routing lookup algorithms based on TCAM (Zheng et al. 2006; Lin et al. 2007; Yang et al. 2012), FIB compression can be used to reduce the hardware cost and power consumption. Furthermore, FIB compression is applicable to any longest matching prefix (LPM) database. Therefore, a variety of FIB compression algorithms are proposed (Draves et al. 1999; Cain 2002; Zhao et al. 2010; Li et al. 2011, 2013; Liu et al. 2010; Yu 2010). These algorithms compress the routing table by transforming the binary trie^{1} structure.
In addition, as is known to all, routing table lookup adopts the longest prefix matching (LPM) rule because of the introduction of classless interdomain routing (CIDR). Due to CIDR, the routing tables’ prefixes are overlapped, which means that some prefixes are a part of others. This brings many negative effects on the performance of routing lookup and incremental update (Yang et al. 2012). There are mainly two overlap elimination algorithms: Leafpushing (Srinivasan and Varghese 1999) and ONRTC (Yang et al. 2012) algorithm. They can totally eliminate the overlap also by transforming the binary trie.^{2}
 (a)
Feasibility and effectiveness According to the information theory, it is definite that the compressed routing table holds the information equivalent to the original one. Therefore, if and only if there is redundancy in the original routing table, the FIB compression solution is feasible. Then is there redundancy in the routing table? What’s the premise of the existence of redundancy? After a profound data mining of the routing tables, we find that although the routing table is rapidly growing (some backbone routers have more than 400 K FIB entries today), the port number of a router is extremely limited (ranging from 3 to 80) and almost static. This observation intuitionally gives a positive answer to the existence of redundancy. Fortunately, the redundancy caused by the almighty gap between the prefix number and port number in the routing table can be quantized by Pigeonhole Principle. Based on this observation, we also deduce the bound of the worst case of the FIB compression solution in this paper.
 (b)
Correctness After an indepth study, we reveal that the LPM rule can be well expressed by the regular expression syntax. We also find that the LPM rule can be well expressed by the Group^{3} theory. Based on these two advancements, two basic equivalent atomic models are induced—election model and representative model. We insist that all the trietransformation algorithms can be proven by these two fundamental atomic models.
Actually, FIB compression algorithm is a tough and errorprone task during the algorithm design and implementation. In order to guarantee the ultimate correctness of FIB compression algorithms, we propose routing table equation test (RTET) to verify the equivalence of the two routing tables before and after compression by traversing the 32bit IP address space.

We propose a universal mathematical method based on a new defined Group, and apply this method to four classical FIB compression algorithms.

We compute the bound of the worst case of various FIB compression solutions.

We propose and implement routing table equation test (RTET) for the first time, to verify the equivalence of the two tries before and after binary trie transformation by traversing the 32bit IP address space. At the end, we implement and verify four classical algorithms by RTET.
The remaining parts of the paper are organized as follows. Section 2 surveys the related work. Section 3 elaborates on a novel mathematical method which can prove the correctness of trietransformation algorithms. The bound of the worst case of FIB compression solution is analyzed in Sect. 4. Section 5 applies this mathematical proof to four classical FIB compression algorithms. The ultimate correctness of FIB compression algorithm is guaranteed by RTET, which is illustrated in Sect. 6. Finally, we conclude this paper in Sect. 7.
2 Related work
IRTF RRG (2014) and IETF GROUP (2015) have been working on the issues about routing scalability for years. Forwarding Information Base^{4} (FIB) compression is a local solution and needs no change to the existing routing protocols, and the representative papers are (Draves et al. 1999; Cain 2002; Zhao et al. 2010; Li et al. 2011, 2013).
Richard Draves et al. (1999) proposed the famous ORTC algorithm, which constructs optimal routing tables. However, ORTC algorithm has not been applied to real routers, for the reason that it is so complicated that it consumes a lot of time and memory, which is not conducive to incremental updates. The core operations of ORTC are ‘UNION’ and ‘AND’, and thus the correctness of these two operations is proven in this paper.
In Cain (2002), a patent technology proposed a compression algorithm, which is simple and fast, but its compression ratio is not high. There is no mathematical proof in this patent. For convenience, it is called patent algorithm in this paper.
Xin Zhao et al. (2010) proposed a 4level algorithm. The first two levels are plain, and cannot achieve a high compression ratio. Although the third and fourth levels can achieve high compression ratio, they can only deal with nonroutable space cases. The nonroutable packets mean those that should be dropped, because no nexthop could be found, but they are forwarded anyway (weak correctness). It is called roaming garbage in this paper. Our real traffic trace shows that the amount of traffic caused by roaming garbage could be up to 0.31% of the total traffic and it covers 0.38% of the whole IP address space. There is no mathematical proof in that paper, either. The 4level algorithm consists of four transformation models. The first two models with strong correctness as well as the other two with weak correctness are proved to be right in this paper.
Li et al. (2011, 2013) proposed the NSFIB compression algorithm, which can achieve a much better compression ratio than ORTC and 4level by taking advantage of multiple next hops, the overhead is sometimes choosing the suboptimal routing path. The author also presented several practical choices to build the sets of alternative next hops for the prefixes, and developed an optimal online algorithm with constant running time. NSFIB algorithm is potentially applied to multicast network (Li et al. 2011).
As a forementioned, there are mainly two algorithms to eliminate overlap: leafpushing (Srinivasan and Varghese 1999) and ONRTC (Yang et al. 2012). In Srinivasan and Varghese (1999), leafpushing algorithm is proposed to eliminate overlap. This algorithm is simple, which just pushes the internal nodes to leaf nodes, inevitably causing the expansion of routing table size. The correctness of this algorithm isn’t proven theoretically, either. In order to reduce the routing table size of leafpushing, ONRTC algorithm in Yang et al. (2012) is proposed to construct optimal nonoverlap routing tables, achieving 71% compression ratio according to the experimental results.
There is another kind of FIB compression algorithm, and we call it entropy compression in this paper. Rétvári et al. (2013) and IRTF Routing Research Group (2014 applied the theory of information entropy to IP prefixes for the first time, and there are two successors (Rottenstreich et al. 2013; Korosi et al. 2014). This kind of compression algorithm transforms the FIB into new format other than prefixes, and thus can hardly work with TCAM or SRAM pipeline. The goal of this kind of algorithm is to approach the theory bound of information entropy. The overhead of entropy compression is difficult to perform incremental update.
All these trietransformation algorithms do not emphasize the mathematical analysis and proof. This motivates our work reported in this paper.
3 Mathematic proof
In this section, a novel mathematical method, which can be used to prove the correctness of trietransformation algorithms, is proposed. Firstly, a new Group is defined, and its four conditions are proven. Secondly, two fundamental atomic models are proven to be right based on this new defined Group. We insist that all the trietransformation algorithms can be proven by these two atomic models, and their applications are highlighted in Sect. 5. We insist that for all FIB compression algorithms, only those compressing the routing table by transforming the binary trie structure can use this proposed theory, and others cannot.
3.1 Group definition

A is a node in the trie, while (A) represents node A’s prefix. Solid nodes have nexthop, while hollow nodes have not.

(AB) represents the bit string of the path between node A and B, while no solid nodes appear in the path.

If A is an ancestor of B, then \( {\text{A}} \subset {\text{B}}. \)

L(A) represents the prefix length of node A.

P represents a trie, and (A) represents a prefix, then P(A) means the nexthop of prefix (A) in trie P.

\( (\widetilde{\text{A}} ) \) represents a prefix with the same length of (A), but it is different from (A). \( {\text{P}}(\widetilde{\text{A}}) \)) means the next hop of prefix (\( \widetilde{\text{A}} \)) in trie P.

(A*) is a 32bit prefix, and (A) is a part of (A*). P(A*) means the next hop of prefix (A*).
In mathematics, a Group (Vvedensky 2005) is a set of elements together with a binary composition law, which must satisfy four conditions: closure, associativity, identity, and invertibility.
Definition 1
LPM Group.
Condition 1
Closure
Proof
Therefore, LPM Group satisfies Closure. \( \square \)
Condition 2
Associativity
Proof
 (1)
If x = 0, \( \left( {x \oplus y} \right) \oplus z = y \oplus z,x \oplus \left( {y \oplus z} \right) = y \oplus z. \)
Therefore,$$ \left( {x \oplus y} \right) \oplus z = x \oplus \left( {y \oplus z} \right). $$Similarly, if y = 0 or z = 0, \( (x \oplus y) \oplus z = x \oplus (y \oplus z). \)
 (2)
\( {\text{x}} \ne 0\;{\text{and}}\;{\text{y}} \ne 0\;{\text{and}}\;{\text{z}} \ne 0. \).
 (2.1)If \( x + y = 0 \), in order to make \( \left( {x \oplus y} \right) \oplus z \) and \( x \oplus (y \oplus z) \) meaningful, \( y + z \) must be zero. Therefore,$$ \begin{aligned} \left( {x \oplus y} \right) \oplus z & = 0 \oplus z = z, \\ x \oplus \left( {y \oplus z} \right) & = x \oplus 0 = x = z, \\ \therefore\, \left( {x \oplus y} \right) \oplus z & = x \oplus \left( {y \oplus z} \right). \\ \end{aligned} $$
 (2.2)If \( x > 0, \;and\;y > 0 \)$$ \begin{aligned} \left( {x \oplus y} \right) \oplus z & = y \oplus z = z, \\ x \oplus \left( {y \oplus z} \right) & = x \oplus z = z. \\ \therefore\, \left( {x \oplus y} \right) \oplus z & = x \oplus \left( {y \oplus z} \right). \\ \end{aligned} $$
Therefore, LPM Group satisfies Associativity.
Condition 3
Identity
Proof
Therefore, LPM Group satisfies Identity. \( \square \)
Condition 4
Invertibility
Proof
Therefore, − x is the inverse of x.
According to the above four conditions, it can be concluded that G is a Group.
LPM Group is used to describe the matching process and results of prefixes in this paper, and thus we define the nexthop and induce Theorem 1 in the following. \( \square \)
Definition 2
P(R)
\( \forall {\text{IP}} \) address R, R = [0,1]{32}, the match result of each bit is S_{i} for IPv4, \( {\text{i}} = 1,2, \ldots ,32 \); for IPv6, \( {\text{i}} = 1,2, \ldots ,128 \); According to the longest prefix matching rule, the nexthop of R is \( P(R) = S_{1} \oplus S_{2} \oplus S_{3} \oplus \ldots S_{32} = \oplus_{i = 1}^{32} S_{i} \).
Theorem 1
If the match results of every section of two prefixes are same, then the nexthops of the two prefixes are same.
Proof
Therefore, \( P1(R) = P2(R) \).
This theorem can be used to prove the equivalence of the nexthop of two tries section by section with regard to one IP address.\( \square \)
Theorem 2
Decision Theorem
The necessary and sufficient condition that two tries are equivalent is that the nexthops are equal in the two tries for any IP addresses by LPM rule.
Obviously, this Decision Theorem naturally holds. Combining Theorem 1 and Theorem 2, we can prove the equivalence of two tries (or two models) section by section. Above all, two basic models which are used to prove other models in this paper are illustrated below.
3.2 An example of election and representative model
In essence, all the trietransformation algorithms follow a process that is similar to the election process of the democratic society. Each node has a nexthop, while each candidate has a vote (Dai et al. 2016a, b, 2018; Zhu et al. 2017). Actually, any candidate’s nexthop can be selected as representative, resulting in different compression ratios. All the nodes which own the same nexthop with the representative can be deleted.
To make a clearer picture of election and representative model, an intuitive example is given in Fig. 2. Figure 2a is the original trie, Fig. 2b is the trie after election, and Fig. 2c is the trie after representative.
Election Nodes A, B, C, and D are four ‘candidates’ nodes, participating in ‘election’. Obviously, hop 2 (rectangle B and D) should be elected as ‘representative’, and thus the nexthop of node E is set to 2.
Representative Node E executes its right of representative: keeping its voters silent, i.e., deleting its voters (node B and D).
3.3 Election and representative models
We insist that all the trietransformation algorithms can be proven by two basic transformation models: election model and representative model.
3.3.1 Election model
Election model Two or more nodes elect their common ancestor node, and no solid node appears in the path from the candidate nodes to the common ancestor node. Any candidates can be elected as a representative, resulting in different compression ratios.

Node X_{i} is the elected representative, and it can own subtree.

\( {\text{A}} \subset {\text{X}}_{\text{i}} ,\;{\text{i}} = 1, 2, 3, \ldots ,{{n}} . \)

There is no solid node in the path between Xi and A.

There is no missing node.
Election result If such t exists: \( \forall {\text{j}}! = {\text{t}},\;{\text{C}}_{\text{j}} \le {\text{C}}_{\text{t}} \) holds, then X_{t} is the elected representative. If such t does not exist, election fails. Then the common ancestor’s nexthop is set to NULL, and participates in the next round election. In this way, an optimal compression ratio can be achieved.
Proof
\( \forall {\text{IP}} \) address R, obviously, \( L\left( R \right) = K, R = \left[ {0,n} \right]\left\{ K \right\}. \) Suppose \( R = [0,n]\{ L(A)\} [0,n][0,n]\{ K  L(A)  1\} \).
According to Theorems 1 and 2, \( P1 \Leftrightarrow P2 \).
If P2 is the election model of P1, we say \( {\text{P}}2 = {\text{Ele}}({\text{P}}1) \). Actually, any node can be elected as a representative, resulting in different compression ratios, and the proof method is similar.
3.3.2 Representative model
Proof
\( \forall {\text{IP}} \) address R, obviously \( L\left( R \right) = 32,\;R = \left[ {0,1} \right]\left\{ {32} \right\}. \) Suppose \( R = [0,1]\{ L(A)\} [0,1]\{ L(AB)\} [0,1]\{ 32  L(B)\} \),
According to Steps 1, 2 and 3, Theorems 1 and 2, we can get \( {\text{P}}1\left( {\text{R}} \right) = {\text{P}}2\left( {\text{R}} \right) \). Therefore, \( \forall R,\;P1\left( R \right) = P2\left( R \right) \) holds according to Theorem 2. \( \therefore\, {\text{P}}1 \Leftrightarrow {\text{P}}2. \)\( \square \)
If P2 is the representative model of P1, we say \( {\text{P}}2 = {\text{Rep}}({\text{P}}1) \). We insist that all models can be proven by the combination of election model and representative model.
4 The worst case of FIB compression solution
In this section, the bound of the worst case of FIB compression solution is computed, so as to prove the feasibility and effectiveness of FIB compression algorithms.
4.1 Bound of the worst case for full IP address space
4.1.1 Pigeonhole principle
In mathematics, the Pigeonhole Principle states that if n + 1 objects are distributed into n boxes, then at least one box contains two or more of the objects (Brualdi 2009). This is a simple but very useful principle. For example, if there are five people from four countries, there are at least two people from the same country.
4.1.2 The worst case for full IP address space^{5}
In this case, the number of compressed routing table by optimal algorithm is R in Eq. (1).
4.2 The worst case for complete binary trie
From above, we can compute the bound of the worst case of FIB compression ratio of Full IP Address Space. However, this is not enough to evaluate the FIB compression ratio in general cases. A more aggressive and general conclusion needs to be drawn.
In order to follow the above method, the trie must be a complete binary trie. Therefore, the actual binary trie used to store routing table should be equally transformed into a complete trie firstly. Fortunately, there are already two algorithms can construct an equivalent complete trie: Leafpushing (Srinivasan and Varghese 1999) and ONRTC (Yang et al. 2012) algorithm. These two algorithms are originally to eliminate overlap. Fortunately, the trie after overlap elimination is a complete trie which is equivalent to the original trie for packet forwarding.
Theorem 3
As long as the port number n is smaller than M (the prefix number of routing table after compressed by ONRTC), FIB compression solution is feasible, regardless of the distribution of nodes or ports.
Proof
According to the precondition, n < M, as well as the Pigeonhole Principle, at least two nodes in the trie after overlap elimination own the same nexthop. Therefore, at least two nodes can be compressed into one. In other words, FIB compression solution works. In conclusion, FIB compression solution is feasible, as long as n < M.
Theorem 4
This equation suggests that given the number of routing table size N (or \( {\text{M}} \approx 0.71{\text{N}} \)) and the port number, FIB compression solution can achieve a result of R. The mathematical derivation process is similar with Eq. (1).
This result suggests that FIB algorithm can achieve a compressed routing table size of 248,582 for RRC 07 regardless of the distribution of nodes and nexthop.
5 Application to FIB compression algorithms
We insist that our group theory can be used to prove the correctness of most FIB compression algorithms. To verify this claim, we apply our group theory to four classic FIB compression algorithms: Leafpushing algorithm, ORTC algorithm, patent algorithm, and 4level algorithm in this section as follows.
5.1 Application to leafpushing algorithm
5.2 Application to ORTC algorithm
Specifically, given two next hop set A and B, if A and B have no next section, then A#B is the union of A and B; otherwise, A#B is the intersection of A and B.
5.2.1 Union \( ( \cup ) \)
Proof
5.2.2 And (‘\( \cap \)’)
Proof
5.3 Application to patent algorithm
Proof
5.4 Application to 4level Algorithm
5.4.1 Level 1
Proof
5.4.2 Level 2
The model of Level 2 is exactly the same as the patent model, and thus the mathematical proof is omitted.
5.4.3 Level 3
Proof
This proves that level 3 compression satisfies the weak correctness.
5.4.4 Level 4
Proof
This proves that level 4 compression satisfies the weak correctness. For 4level algorithm, if the root node is not NULL, i.e., the nexthop of the root node is nonzero, then the Level 3 and Level 4 models achieve no compression. The root node prefix stands for default routing, and thus if and only if there is no default routing in a routing table, Levels 3 and 4 work.
6 Routing table equation test
The mathematical proof method has been elaborated above, but there might be flaws in the process of mathematical derivation and coding. How to guarantee the ultimate correctness of these algorithms? The ultimate correctness refers to that for any IP address, the compressed routing table tells the same nexthop with the original table. Therefore, we propose routing table equation test (RTET) to judge the equivalence of the two routing tables. RTET firstly builds two tries, then traverses 32bit IP address space, and compares the nexthop of two tries by using the same IP address. If and only if all are equal, the two routing tables are equivalent. Otherwise, RTET stops and tells the prefix and the different nexthop of the two tries. One comparison of two routing tables by using RTET takes about 16 min. The algorithms (Draves et al. 1999; Cain 2002; Zhao et al. 2010; Li et al. 2011) are all implemented and verified by RTET, using the routing tables downloaded from RIPE Network Coordination Centre.
7 Conclusions

Leafpushing algorithm can totally eliminate overlap, but inevitably causing the increase of routing table size.

Patent algorithm only compresses the two sibling nodes which own the same nexthop, and thus the compression efficiency is less than ORTC and 4level algorithm.

4level algorithm can achieve a good compression, if and only if the root node is NULL.

If the NULL root node is considered, ORTC algorithm can achieve a better compression.
Footnotes
 1.
Trie is a treebased data structure allowing the organization of prefixes on a digital basis by using the bits of prefixes to direct the branching (RuizSánchez et al. 2001).
 2.
Both FIB compression and overlap elimination algorithms transform the binary trie, thus they are called trietransformation algorithms in this paper.
 3.
Group (mathematics) (Vvedensky 2005) is a set together with a binary operation satisfying certain algebraic conditions.
 4.
FIB is also known as forwarding table, which is stored in linecards to forward data packets. Each entry of FIB stores a prefix and the corresponding nexthop, such as 200.45.65.0/24:40. It suggests that if an incoming IP address matched the prefix 200.45.65.0/24 by LPM, this packet should be forwarded to the interface 40 (40 is also related to a corresponding nexthop IP address).
 5.
In this paper, Full IP Address Space refers to the binary trie whose internal nodes are all pushed to level 32.
Notes
Acknowledgements
We would like to thank the anonymous reviewers for their thoughtful suggestions. This work is partially supported by Primary Research & Development Plan of China (2018YFB1004403), National Basic Research Program of China (973 Program, 2014CB340405), and NSFC (61672061).
Compliance with ethical standards
Conflict of interest
On behalf of all authors, the corresponding author states that there is no conflict of interest.
References
 Brualdi, R.A.: Introductory Combinatorics, Chapter 3, 5th edn, pp. 69–70. Machine Press China, Beijing (2009)Google Scholar
 Cain, B.: Auto aggregation method for IP prefix/length pairs. http://www.patentgenius.com/patent/6401130.html (2002)
 Dai, H., Zhong, Y., Liu, A.X., Wang, W., Li, M.: Noisy bloom filters for multiset membership testing. In: Proc. ACM SIGMETRICS, pp 139–151 (2016)Google Scholar
 Dai, H., Shahzad, M., Liu, A.X., Zhong, Y.: Finding persistent items in data streams. Proc. VLDB Endow. 10(4), 289–300 (2016)CrossRefGoogle Scholar
 Dai, H., Meng, L., Liu, A.X.: Finding persistent items in distributed, datasets. In: Proc. IEEE INFOCOM (2018)Google Scholar
 Degermaerk, M., Brodnik, A., Carlsson, S., and Pink, S: Small forwarding tables for fast routing lookups. In: Proc. SIGCOMM, NY (1997)Google Scholar
 Draves, R., King, C., Venkatachary, S., Zill, B.D.: Constructing optimal IP routing tables. In: Proc. IEEE INFOCOM, pp. 88–97 (1999)Google Scholar
 IETF Global Routing Operations (GROW). http://datatracker.ietf.org/wg/grow/charter/ (2015)
 IRTF Routing Research Group. https://irtf.org/concluded/rrg (2014)
 Korosi, A., Tapolcai, J., Mihálka, B., Mészáros, G., Rétvári, G.: Compressing IP forwarding tables: realizing informationtheoretical space bounds and fast lookups simultaneously. In: Proc. IEEE ICNP, IEEE, pp. 332–343 (2014)Google Scholar
 Li, D., Cui, H., Hu, Y., et al.: Scalable data center multicast using multiclass bloom filter, network protocols (ICNP). In: 19th IEEE international conference on, pp. 266–275 (2011)Google Scholar
 Li, Q., Wang, D., Xu, M., Yang, J.: On the scalability of router forwarding tables: nexthopselectable FIB aggregation. In: Proc. IEEE INFOCOM (2011)Google Scholar
 Li, Q., Xu, M., Chen, M.: NSFIB construction and aggregation with next hop of strict partial order. INFOCOM, 2013, pp. 550–554. In: Proceedings IEEE. IEEE (2013)Google Scholar
 Li, F.Z., Xiao, S.W., Pei, T., Li, Jie: Achievable rate maximization for cognitive hybrid satelliteterrestrial networks with afrelays. IEEE J. Sel. Areas Commun. 36(2), 304–313 (2018)MathSciNetCrossRefGoogle Scholar
 Li, Z., Chang, B., Wang, S., Liu, A., Zeng, F, Luo, G.: Dynamic compressive wideband spectrum sensing based on channel energy reconstruction in cognitive internet of things. IEEE Trans. Ind. Inf. (2018)Google Scholar
 Lin, D., Zhang, Y., Hu, C., Liu, B., Zhang, X., Pao, D.: Route table partitioning and load balancing for parallel searching with TCAMs. In: Proc. IPDPS (2007)Google Scholar
 Liu, Y., Zhao, X., Nam, K., Wang, L., Zhang, B.: Incremental forwarding table aggregation. In: Proc. IEEEE GLOBECOM (2010)Google Scholar
 Meng, X., Xu, Z., Zhang, B., Huston, G., Lu, S., Zhang, L.: IPv4 address allocation and the BGP routing table. ACM SIGCOMM Comput Commun Rev 35, 71–80 (2005)CrossRefGoogle Scholar
 Nilsson, S., Karlsson, G.: Fast address lookup for internet routers. In: Proceedings of IEEE broadband communications (1998)Google Scholar
 Rétvári, G., Tapolcai, J., Kőrösi, A., Majdán, A., Heszberger, Z.: Compressing IP forwarding tables: towards entropy bounds and beyond. ACM SIGCOMM Comput. Commun. Rev. 43, 111–122 (2013)CrossRefGoogle Scholar
 RIPE Network Coordination Centre. http://www.ripe.net/datatools/stats/ris/risrawdata
 Rottenstreich, O., Radan, M., Cassuto, Y., Keslassy, I., Arad, C., Mizrahi, T., Revah, Y., Hassidim, A.: Compressing forwarding tables. In: Proc. IEEE INFOCOM, IEEE, pp. 1231–1239 (2013)Google Scholar
 RuizSánchez, M.Á., Biersack, E.W., Dabbous, W.: Survey and taxonomy of IP address lookup algorithms. Netw. IEEE 15, 8–23 (2001)CrossRefGoogle Scholar
 Srinivasan, V., Varghese, G.: Fast IP lookups using controlled prefix expansion. ACM TOCS 17, 1–40 (1999)CrossRefGoogle Scholar
 Vvedensky, D.D.: Group Theory, pp. 14–15. World Scientific Pub., Singapore (2005)Google Scholar
 Waldvogel, M., Varghese, G., Turner, J., Plattner, B.: Scalable high speed ip routing lookups. ACM. 27(4), 25–36 (1997)Google Scholar
 Xiao, F., Wang, Z., Ye, N., Wang, R., Li, X.Y.: One more tag enables finegrained RFID localization and tracking. IEEE/ACM Trans. Netw. 26(1), 161–174 (2018a)CrossRefGoogle Scholar
 Xiao, F., Chen, L., Sha, C., Sun, L., Wang, R., Liu, A.X., Ahmed, F.: Noise tolerant localization for sensor networks. IEEE/ACM Trans. Netw. 26(4), 1701–1704 (2018b)CrossRefGoogle Scholar
 Yang, T., Duan, R., Lu, J., Zhang, S., Dai, H., Liu, B.: CLUE: achieving fast update over compressed table for parallel lookup with reduced dynamic redundancy. In: Proc. IEEE ICDCS (2012)Google Scholar
 Yang, T., Zhang, T., Zhang, S., Liu, B.: Constructing optimal nonoverlap routing tables. Accepted by Proc. IEEE ICC (2012)Google Scholar
 Yu, H.: A memory and timeefficient onchip TCAM minimizer for IP lookup. DATE ‘10. In: Proceedings of the conference on design, automation and test in Europe (2010)Google Scholar
 Zhao, X., Liu, Y., Wang, L., Zhang, B.: On the aggregatability of router forwarding tables. In: Proc. IEEE INFOCOM (2010)Google Scholar
 Zheng, K., Hu, C., Lu, H., Liu, B.: A TCAMbased distributed parallel IP lookup scheme and performance analysis. IEEE/ACM Trans. Netw. 14, 863–875 (2006)CrossRefGoogle Scholar
 Zhu, H., Xiao, F., Lijuan, S., Wang, R., Yang, P.: RTTWD: robust devicefree throughthewall detection of moving Human with WiFi. IEEE J. Sel. Areas Commun. 35(5), 1090–1103 (2017)CrossRefGoogle Scholar