*** empty log message ***

2008-03-23 03:45:01 +00:00 · 2008-03-23 03:45:01 +00:00 · 32afd2075f
parent f3ab04d6ef
commit 32afd2075f
6 changed files with 61 additions and 345 deletions
--- a/BDZ.t2t
+++ b/BDZ.t2t
@ -1,401 +1,112 @@
-BMZ Algorithm
+BDZ Algorithm
 %!includeconf: CONFIG.t2t
 ----------------------------------------
-==History==
+==Introduction==
 At the end of 2003, professor [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] was
 finishing the second edition of his [book http://www.dcc.ufmg.br/algoritmos/].
 During the [book http://www.dcc.ufmg.br/algoritmos/] writing, 
 professor [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] studied the problem of generating 
 [minimal perfect hash functions concepts.html]
 (if you are not familiarized with this problem, see [[1 #papers]][[2 #papers]]). 
 Professor [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] coded a modified version of 
 the [CHM algorithm chm.html], which was proposed by
 Czech, Havas and Majewski, and put it in his [book http://www.dcc.ufmg.br/algoritmos/].
 The [CHM algorithm chm.html] is based on acyclic random graphs to generate 
 [order preserving minimal perfect hash functions concepts.html] in linear time. 
 Professor [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] 
 argued himself, why must the random graph 
 be acyclic? In the modified version availalbe in his [book http://www.dcc.ufmg.br/algoritmos/] he got rid of this restriction.
 The modification presented a problem, it was impossible to generate minimal perfect hash functions
 for sets with more than 1000 keys.
 At the same time, [Fabiano C. Botelho http://www.dcc.ufmg.br/~fbotelho],
 a master degree student at [Departament of Computer Science http://www.dcc.ufmg.br] in 
 [Federal University of Minas Gerais http://www.ufmg.br],
 started to be advised by [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] who presented the problem 
 to [Fabiano http://www.dcc.ufmg.br/~fbotelho].
 During the master, [Fabiano http://www.dcc.ufmg.br/~fbotelho] and 
 [Nivio Ziviani http://www.dcc.ufmg.br/~nivio] faced lots of problems.
 In april of 2004, [Fabiano http://www.dcc.ufmg.br/~fbotelho] was talking with a 
 friend of him (David Menoti) about the problems
 and many ideas appeared.
 The ideas were implemented and a very fast algorithm to generate
 minimal perfect hash functions had been designed.
 We refer the algorithm to as **BMZ**, because it was conceived by Fabiano C. **B**otelho,
 David **M**enoti and Nivio **Z**iviani. The algorithm is described in [[1 #papers]].
 To analyse BMZ algorithm we needed some results from the random graph theory, so 
 we invited professor [Yoshiharu Kohayakawa http://www.ime.usp.br/~yoshi] to help us.
 The final description and analysis of BMZ algorithm is presented in [[2 #papers]].
 ----------------------------------------
 ==The Algorithm==
 The BMZ algorithm shares several features with the [CHM algorithm chm.html].   
 In particular, BMZ algorithm is also
 based on the generation of random graphs [figs/img27.png], where [figs/img28.png] is in 
 one-to-one correspondence with the key set [figs/img20.png] for which we wish to 
 generate a [minimal perfect hash function concepts.html].
 The two main differences between BMZ algorithm and CHM algorithm
 are as follows: (//i//)  BMZ algorithm generates random 
 graphs [figs/img27.png] with [figs/img29.png] and [figs/img30.png], where [figs/img31.png], 
 and hence [figs/img32.png] necessarily contains cycles, 
 while CHM algorithm generates //acyclic// random 
 graphs [figs/img27.png] with [figs/img29.png] and [figs/img30.png],
 with a greater number of vertices: [figs/img33.png];
 (//ii//) CHM algorithm generates [order preserving minimal perfect hash functions concepts.html]
 while BMZ algorithm does not preserve order.  Thus, BMZ algorithm improves
 the space requirement at the expense of generating functions that are not
 order preserving. 
 Suppose [figs/img14.png] is a universe of //keys//.
 Let [figs/img17.png] be a set of [figs/img8.png] keys from [figs/img14.png].
 Let us show how the BMZ algorithm constructs a minimal perfect hash function [figs/img7.png].
 We make use of two auxiliary random functions [figs/img41.png] and [figs/img55.png], 
 where [figs/img56.png] for some suitably chosen integer [figs/img57.png], 
 where [figs/img58.png].We build a random graph [figs/img59.png] on [figs/img60.png],
 whose edge set is [figs/img61.png]. There is an edge in [figs/img32.png] for each 
 key in the set of keys [figs/img20.png].
 In what follows, we shall be interested in the //2-core// of
 the random graph [figs/img32.png], that is, the maximal subgraph 
 of [figs/img32.png] with minimal degree at 
 least 2 (see [[2 #papers]] for details).
 Because of its importance in our context, we call the 2-core the
 //critical// subgraph of [figs/img32.png] and denote it by [figs/img63.png].
 The vertices and edges in [figs/img63.png] are said to be //critical//.
 We let [figs/img64.png] and [figs/img65.png].
 Moreover, we let [figs/img66.png] be the set of //non-critical//
 vertices in [figs/img32.png].
 We also let [figs/img67.png] be the set of all critical
 vertices that have at least one non-critical vertex as a neighbour.
 Let [figs/img68.png] be the set of //non-critical// edges in [figs/img32.png].
 Finally, we let [figs/img69.png] be the //non-critical// subgraph 
 of [figs/img32.png].
 The non-critical subgraph [figs/img70.png] corresponds to the //acyclic part//
 of [figs/img32.png].
 We have [figs/img71.png].
 We then construct a suitable labelling [figs/img72.png] of the vertices
 of [figs/img32.png]: we choose [figs/img73.png] for each [figs/img74.png] in such
 a way that [figs/img75.png] ([figs/img18.png]) is a
 minimal perfect hash function for [figs/img20.png].
 This labelling [figs/img37.png] can be found in linear time
 if the number of edges in [figs/img63.png] is at most [figs/img76.png] (see [[2 #papers]] 
 for details).
 Figure 1 presents a pseudo code for the BMZ algorithm.
 The procedure BMZ ([figs/img20.png], [figs/img37.png]) receives as input the set of
 keys [figs/img20.png] and produces the labelling [figs/img37.png].
 The method uses a mapping, ordering and searching approach.
 We now describe each step.
 | procedure BMZ ([figs/img20.png], [figs/img37.png])                              
 | &nbsp;&nbsp;&nbsp;&nbsp;Mapping ([figs/img20.png], [figs/img32.png]);                
 | &nbsp;&nbsp;&nbsp;&nbsp;Ordering ([figs/img32.png], [figs/img63.png], [figs/img70.png]); 
 | &nbsp;&nbsp;&nbsp;&nbsp;Searching ([figs/img32.png], [figs/img63.png], [figs/img70.png], [figs/img37.png]);
 | **Figure 1**: Main steps of BMZ algorithm for constructing a minimal perfect hash function 
 ----------------------------------------
 ===Mapping Step===
 The procedure Mapping ([figs/img20.png], [figs/img32.png]) receives as input the set 
 of keys [figs/img20.png] and generates the random graph [figs/img59.png], by generating 
 two auxiliary functions [figs/img41.png], [figs/img78.png].
 The functions [figs/img41.png] and [figs/img42.png] are constructed as follows.
 We impose some upper bound [figs/img79.png] on the lengths of the keys in [figs/img20.png].
 To define [figs/img80.png] ([figs/img81.png], [figs/img62.png]), we generate 
 an [figs/img82.png] table of random integers [figs/img83.png].
 For a key [figs/img18.png] of length [figs/img84.png] and [figs/img85.png], we let 
 | [figs/img86.png]
 The random graph [figs/img59.png] has vertex set [figs/img56.png] and 
 edge set [figs/img61.png].  We need [figs/img32.png] to be 
 simple, i.e., [figs/img32.png] should have neither loops nor multiple edges.
 A loop occurs when [figs/img87.png] for some [figs/img18.png].
 We solve this in an ad hoc manner: we simply let [figs/img88.png] in this case. 
 If we still find a loop after this, we generate another pair [figs/img89.png].
 When a multiple edge occurs we abort and generate a new pair [figs/img89.png]. 
 Although the function above causes [collisions concepts.html] with probability //1/t//,
 in [cmph library index.html] we use faster hash 
 functions ([DJB2 hash http://www.cs.yorku.ca/~oz/hash.html], [FNV hash http://www.isthe.com/chongo/tech/comp/fnv/],
 [Jenkins hash http://burtleburtle.net/bob/hash/doobs.html] and [SDBM hash http://www.cs.yorku.ca/~oz/hash.html]) 
 in which we do not need to impose any upper bound [figs/img79.png] on the lengths of the keys in [figs/img20.png].
 As mentioned before, for us to find  the labelling [figs/img72.png] of the 
 vertices of [figs/img59.png] in linear time,
 we require that [figs/img108.png]. 
 The crucial step now is to determine the value 
 of [figs/img1.png] (in [figs/img57.png]) to obtain a random 
 graph [figs/img71.png] with [figs/img109.png].
 Botelho, Menoti an Ziviani determinded emprically in [[1 #papers]] that 
 the value of [figs/img1.png] is //1.15//. This value is remarkably
 close to the theoretical value determined in [[2 #papers]], 
 which is around [figs/img112.png].
 ----------------------------------------
-===Ordering Step===
+===Assigning Step===
 The procedure Ordering ([figs/img32.png], [figs/img63.png], [figs/img70.png]) receives 
 as input the graph [figs/img32.png] and partitions [figs/img32.png] into the two 
 subgraphs [figs/img63.png] and [figs/img70.png], so that [figs/img71.png].
 Figure 2 presents a sample graph with 9 vertices
 and 8 edges, where the degree of a vertex is shown besides each vertex.
 Initially, all vertices with degree 1 are added to a queue [figs/img136.png].
 For the example shown in Figure 2(a), [figs/img137.png] after the initialization step.
 | [figs/img138.png]
 | **Figure 2:** Ordering step for a graph with 9 vertices and 8 edges.
 Next, we remove one vertex [figs/img139.png] from the queue, decrement its degree and
 the degree of the vertices with degree greater than 0 in the adjacent
 list of [figs/img139.png], as depicted in Figure 2(b) for [figs/img140.png].
 At this point, the adjacencies of [figs/img139.png] with degree 1 are
 inserted into the queue, such as vertex 1.
 This process is repeated until the queue becomes empty.
 All vertices with degree 0 are non-critical vertices and the others are
 critical vertices, as depicted in Figure 2(c).
 Finally, to determine the vertices in [figs/img141.png] we collect all 
 vertices [figs/img142.png] with at least one vertex [figs/img143.png] that 
 is in Adj[figs/img144.png] and in [figs/img145.png], as the vertex 8 in Figure 2(c).
 ----------------------------------------
-===Searching Step===
+===Ranking Step===
 In the searching step, the key part is
 the //perfect assignment problem//: find [figs/img153.png] such that
 the function [figs/img154.png] defined by
 | [figs/img155.png]
 is a bijection from [figs/img156.png] to [figs/img157.png] (recall [figs/img158.png]).
 We are interested in a labelling [figs/img72.png] of
 the vertices of the graph [figs/img59.png] with
 the property that if [figs/img11.png] and [figs/img22.png] are keys 
 in [figs/img20.png], then [figs/img159.png]; that is, if we associate
 to each edge the sum of the labels on its endpoints, then these values
 should be all distinct.
 Moreover, we require that all the sums [figs/img160.png] ([figs/img18.png])
 fall between [figs/img115.png] and [figs/img161.png], and thus we have a bijection
 between [figs/img20.png] and [figs/img157.png].
 The procedure Searching ([figs/img32.png], [figs/img63.png], [figs/img70.png], [figs/img37.png])
 receives as input [figs/img32.png], [figs/img63.png], [figs/img70.png] and finds a 
 suitable [figs/img162.png] bit value for each vertex [figs/img74.png], stored in the
 array [figs/img37.png].
 This step is first performed for the vertices in the
 critical subgraph [figs/img63.png] of [figs/img32.png] (the 2-core of [figs/img32.png]) 
 and then it is performed for the vertices in [figs/img70.png] (the non-critical subgraph
 of [figs/img32.png] that contains the "acyclic part" of [figs/img32.png]).
 The reason the assignment of the [figs/img37.png] values is first
 performed on the vertices in [figs/img63.png] is to resolve reassignments
 as early as possible (such reassignments are consequences of the cycles
 in [figs/img63.png] and are depicted hereinafter).
 ----------------------------------------
 ====Assignment of Values to Critical Vertices====
 The labels [figs/img73.png] ([figs/img142.png])
 are assigned in increasing order following a greedy
 strategy where the critical vertices [figs/img139.png] are considered one at a time,
 according to a breadth-first search on [figs/img63.png].
 If a candidate value [figs/img11.png] for [figs/img73.png] is forbidden
 because setting [figs/img163.png] would create two edges with the same sum,
 we try [figs/img164.png] for [figs/img73.png]. This fact is referred to 
 as a //reassignment//.
 Let [figs/img165.png] be the set of addresses assigned to edges in [figs/img166.png].
 Initially [figs/img167.png].
 Let [figs/img11.png] be a candidate value for [figs/img73.png].
 Initially [figs/img168.png].
 Considering the subgraph [figs/img63.png] in Figure 2(c),
 a step by step example of the assignment of values to vertices in [figs/img63.png] is 
 presented in Figure 3.
 Initially, a vertex [figs/img139.png] is chosen, the assignment [figs/img163.png] is made
 and [figs/img11.png] is set to [figs/img164.png].
 For example, suppose that vertex [figs/img169.png] in Figure 3(a) is 
 chosen, the assignment [figs/img170.png] is made and [figs/img11.png] is set to [figs/img96.png].
 | [figs/img171.png]
 | **Figure 3:** Example of the assignment of values to critical vertices.
 In Figure 3(b), following the adjacent list of vertex [figs/img169.png],
 the unassigned vertex [figs/img115.png] is reached.
 At this point, we collect in the temporary variable [figs/img172.png] all adjacencies 
 of vertex [figs/img115.png] that have been assigned an [figs/img11.png] value, 
 and [figs/img173.png].
 Next, for all [figs/img174.png], we check if [figs/img175.png].
 Since [figs/img176.png], then [figs/img177.png] is set 
 to [figs/img96.png], [figs/img11.png] is incremented 
 by 1 (now [figs/img178.png]) and [figs/img179.png].
 Next, vertex [figs/img180.png] is reached, [figs/img181.png] is set 
 to [figs/img62.png], [figs/img11.png] is set to [figs/img180.png] and [figs/img182.png].
 Next, vertex [figs/img183.png] is reached and [figs/img184.png].
 Since [figs/img185.png] and [figs/img186.png], then [figs/img187.png] is 
 set to [figs/img180.png], [figs/img11.png] is set to [figs/img183.png] and [figs/img188.png].
 Finally, vertex [figs/img189.png] is reached and [figs/img190.png].
 Since [figs/img191.png], [figs/img11.png] is incremented by 1 and set to 5, as depicted in 
 Figure 3(c).
 Since [figs/img192.png], [figs/img11.png] is again incremented by 1 and set to 6, 
 as depicted in Figure 3(d).
 These two reassignments are indicated by the arrows in Figure 3.
 Since [figs/img193.png] and [figs/img194.png], then [figs/img195.png] is set 
 to [figs/img196.png] and [figs/img197.png]. This finishes the algorithm.
 ----------------------------------------
 ====Assignment of Values to Non-Critical Vertices====
 As [figs/img70.png] is acyclic, we can impose the order in which addresses are
 associated with edges in [figs/img70.png], making this step simple to solve
 by a standard depth first search algorithm.
 Therefore, in the assignment of values to vertices in [figs/img70.png] we
 benefit from the unused addresses in the gaps left by the assignment of values
 to vertices in [figs/img63.png].
 For that, we start the depth-first search from the vertices in [figs/img141.png] because 
 the [figs/img37.png] values for these critical vertices were already assigned
 and cannot be changed.
 Considering the subgraph [figs/img70.png] in Figure 2(c),
 a step by step example of the assignment of values to vertices in [figs/img70.png] is 
 presented in Figure 4.
 Figure 4(a) presents the initial state of the algorithm.
 The critical vertex 8 is the only one that has non-critical vertices as
 adjacent.
 In the example presented in Figure 3, the addresses [figs/img198.png] were not used.
 So, taking the first unused address [figs/img115.png] and the vertex [figs/img96.png], 
 which is reached from the vertex [figs/img169.png], [figs/img199.png] is set 
 to [figs/img200.png], as shown in Figure 4(b).
 The only vertex that is reached from vertex [figs/img96.png] is vertex [figs/img62.png], so
 taking the unused address [figs/img183.png] we set [figs/img201.png] to [figs/img202.png],
 as shown in Figure 4(c).
 This process is repeated until the UnAssignedAddresses list becomes empty.
 | [figs/img203.png]
 | **Figure 4:** Example of the assignment of values to non-critical vertices.
 ----------------------------------------
 ==The Heuristic==[heuristic]
 We now present an heuristic for BMZ algorithm that 
 reduces the value of [figs/img1.png] to any given value between //1.15// and //0.93//.
 This reduces the space requirement to store the resulting function 
 to any given value between [figs/img12.png] words and [figs/img13.png] words.
 The heuristic reuses, when possible, the set 
 of [figs/img11.png] values that caused reassignments, just before 
 trying [figs/img164.png].
 Decreasing the value of [figs/img1.png] leads to an increase in the number of 
 iterations to generate [figs/img32.png].
 For example, for [figs/img244.png] and [figs/img6.png], the analytical expected number 
 of iterations are [figs/img245.png] and [figs/img246.png], respectively (see [[2 #papers]] 
 for details),
 while for [figs/img128.png] the same value is around //2.13//.
 ----------------------------------------
 ==Memory Consumption==
 Now we detail the memory consumption to generate and to store minimal perfect hash functions
-using the BMZ algorithm. The structures responsible for memory consumption are in the 
+using the BDZ algorithm. The structures responsible for memory consumption are in the 
 following:
- Graph:
+- 3-graph:
  + **first**: is a vector that stores //cn// integer numbers, each one representing 
    the first edge (index in the vector edges) in the list of 
-    edges of each vertex. 
+    incident edges of each vertex. The integer numbers are 4 bytes long. Therefore,
    The integer numbers are 4 bytes long. Therefore,
    the vector first is stored in //4cn// bytes.
  + **edges**: is a vector to represent the edges of the graph. As each edge
-    is compounded by a pair of vertices, each entry stores two integer numbers 
+    is compounded by three vertices, each entry stores three integer numbers 
    of 4 bytes that represent the vertices. As there are //n// edges, the 
-    vector edges is stored in //8n// bytes. 
+    vector edges is stored in //12n// bytes. 
  + **next**: given a vertex [figs/img139.png], we can discover the edges that 
-    contain [figs/img139.png] following its list of edges, 
+    contain [figs/img139.png] following its list of incident edges, 
    which starts on first[[figs/img139.png]] and the next
    edges are given by next[...first[[figs/img139.png]]...]. Therefore, the vectors first and next represent 
-    the linked lists of edges of each vertex. As there are two vertices for each edge,
+    the linked lists of edges of each vertex. As there are three vertices for each edge,
-    when an edge is iserted in the graph, it must be inserted in the two linked lists 
+    when an edge is iserted in the 3-graph, it must be inserted in the three linked lists 
-    of the vertices in its composition. Therefore, there are //2n// entries of integer
+    of the vertices in its composition. Therefore, there are //3n// entries of integer
-    numbers in the vector next, so it is stored in //4*2n = 8n// bytes.
+    numbers in the vector next, so it is stored in //4*3n = 12n// bytes.
-  + **critical vertices(critical_nodes vector)**: is a vector of //cn// bits, 
+  + **Vertices degree (vert_degree vector)**: is a vector of //cn// bytes
-    where each bit indicates if a vertex is critical (1) or non-critical (0). 
+    that represents the degree of each vertex. We can use just one byte for each
-    Therefore, the critical and non-critical vertices are represented in //cn/8// bytes.
+    vertex because the 3-graph is sparse, once it has more vertices than edges. 
-    
+    Therefore, the vertices degree is represented in //cn// bytes.
  + **critical edges (used_edges vector)**: is a vector of //n// bits, where each 
    bit indicates if an edge is critical (1) or non-critical (0). Therefore, the 
    critical and non-critical edges are represented in //n/8// bytes. 
 - Other auxiliary structures 
  + **queue**: is a queue of integer numbers used in the breadth-first search of the
    assignment of values to critical vertices. There is an entry in the queue for 
    each two critical vertices. Let [figs/img110.png] be the expected number of critical 
    vertices. Therefore, the queue is stored in //4*0.5*[figs/img110.png]=2[figs/img110.png]//.
  + **visited**: is a vector of //cn// bits, where each bit indicates if the g value of 
    a given vertex was already defined. Therefore, the vector visited is stored
    in //cn/8// bytes.
  + **function //g//**: is represented by a vector of //cn// integer numbers.
    As each integer number is 4 bytes long, the function //g// is stored in
    //4cn// bytes. 
 - Acyclicity test:    
  + **List of deleted edges obtained when we test whether the 3-graph is a forest (queue vector)**: 
    is a vector of //n// integer numbers containing indexes of vector edges. Therefore, it 
    requires //4n// bytes in internal memory. 
-Thus, the total memory consumption of BMZ algorithm for generating a minimal 
+  + **Marked edges in the acyclicity test (marked_edges vector)**: 
-perfect hash function (MPHF) is: //(8.25c + 16.125)n +2[figs/img110.png] + O(1)// bytes.
+    is a bit vector of //n// bits to indicate the edges that have already been deleted during 
-As the value of constant //c// may be 1.15 and 0.93 we have:
+    the acyclicity test. Therefore, it requires //n/8// bytes in internal memory. 
- || //c// |  [figs/img110.png] | Memory consumption to generate a MPHF |
+
-  | 0.93  |  //0.497n//  |         //24.80n + O(1)//             |
+- MPHF description     
-  | 1.15  |  //0.401n//  |         //26.42n + O(1)//             |
+  + **function //g//**: is represented by a vector of //2cn// bits. Therefore, it is 
-  
+    stored in //0.25cn// bytes
-  | **Table 1:** Memory consumption to generate a MPHF using the BMZ algorithm.
+  + **ranktable**: is a lookup table used to store some precomputed ranking information.
-  
+    It has //(cn)/(2^b)// entries of 4-byte integer numbers. Therefore it is stored in 
-The values of [figs/img110.png] were calculated using Eq.(1) presented in [[2 #papers]].
+    //(4cn)/(2^b)// bytes. The larger is b, the more compact is the resulting MPHFs and
    the slower are the functions. So b imposes a trade-of between space and time.
  + **Total**: 0.25cn + (4cn)/(2^b) bytes
 Thus, the total memory consumption of BDZ algorithm for generating a minimal 
 perfect hash function (MPHF) is: //(28.125 + 5c)n + 0.25cn + (4cn)/(2^b) + O(1)// bytes.
 As the value of constant //c// may be larger than or equal to 1.23 we have:
 || //c// |  //b//  | Memory consumption to generate a MPHF  (in bytes) |
  | 1.23  |  //7//  |         //34.62n + O(1)//                         |
  | 1.23  |  //8//  |         //34.60n + O(1)//                         |
  | **Table 1:** Memory consumption to generate a MPHF using the BDZ algorithm.
 Now we present the memory consumption to store the resulting function.
-We only need to store the //g// function. Thus, we need //4cn// bytes.
+So we have:
-Again we have:
+ || //c// |  //b//  | Memory consumption to store a MPHF (in bits) |
- || //c// | Memory consumption to store a MPHF |
+  | 1.23  |  //7//  |         //2.77n + O(1)//                     |
-  | 0.93  |            //3.72n//               |
+  | 1.23  |  //8//  |         //2.61n + O(1)//                     |
  | 1.15  |            //4.60n//               |
-  | **Table 2:** Memory consumption to store a MPHF generated by the BMZ algorithm.  
+  | **Table 2:** Memory consumption to store a MPHF generated by the BDZ algorithm.  
 ----------------------------------------
 ==Experimental Results==
-[CHM x BMZ comparison.html]
+Experimental results to compare the BDZ algorithm with the other ones in the CMPH
-
+library are presented in Botelho, Pagh and Ziviani [[1 #papers],[2 #papers]].
 ----------------------------------------
 ==Papers==[papers]
-+ [F. C. Botelho http://www.dcc.ufmg.br/~fbotelho], D. Menoti, [N. Ziviani http://www.dcc.ufmg.br/~nivio]. [A New algorithm for constructing minimal perfect hash functions papers/bmz_tr004_04.ps], Technical Report TR004/04, Department of Computer Science, Federal University of Minas Gerais, 2004.
+ [F. C. Botelho http://www.dcc.ufmg.br/~fbotelho], R. Pagh, [N. Ziviani http://www.dcc.ufmg.br/~nivio]. [Simple and space-efficient minimal perfect hash functions papers/wads07.pdf]. //10th International Workshop on Algorithms and Data Structures (WADs'07),// Springer-Verlag Lecture Notes in Computer Science, vol. 4619, Halifax, Canada, August 2007, 139-150.
-+ [F. C. Botelho http://www.dcc.ufmg.br/~fbotelho], Y. Kohayakawa, and [N. Ziviani http://www.dcc.ufmg.br/~nivio]. [A Practical Minimal Perfect Hashing Method papers/wea05.pdf]. //4th International Workshop on efficient and Experimental Algorithms (WEA05),// Springer-Verlag Lecture Notes in Computer Science, vol. 3505, Santorini Island, Greece, May 2005, 488-500.
+ [F. C. Botelho http://www.dcc.ufmg.br/~fbotelho]. [Near Space-Optimal Perfect Hashing Algorithms papers/thesis.pdf]. //Thesis Proposal//, //Department of Computer Science//, //Federal University of Minas Gerais//, July 2007.
 %!include: ALGORITHMS.t2t
--- a/README.t2t
+++ b/README.t2t
@ -48,7 +48,8 @@ The CMPH Library encapsulates the newest and more efficient algorithms in an eas
  generalization of a graph where each edge connects 3 vertices instead of only 2. The 
  resulting functions are not order preserving and can be stored in only //(2 + x)cn// 
  bits, where //c// should be larger than or equal to //1.23// and //x// is a constant 
-  larger than //0// (actually, x = 1/b and b is a parameter that should be larger 2). 
+  larger than //0// (actually, x = 1/b and b is a parameter that should be larger than 2). 
  For //c = 1.23// and //b = 8//, the resulting functions are stored in approximately 2.6 bits per key.
 %html% - [BMZ Algorithm bmz.html].
 %txt% - BMZ Algorithm.
  A very fast algorithm based on cyclic random graphs to construct minimal
@ -81,15 +82,17 @@ The CMPH Library encapsulates the newest and more efficient algorithms in an eas
 ==News for version 0.8==
- [An algorithm to generate MPHFs that require around 2.62 bits per key to be stored bdz.html], which is referred to as BDZ algorithm. The algorithm is the fastest one available in the literature for sets that can be treated in internal memory.
+- [An algorithm to generate MPHFs that require around 2.6 bits per key to be stored bdz.html], which is referred to as BDZ algorithm. The algorithm is the fastest one available in the literature for sets that can be treated in internal memory.
 - The hash functions djb2, fnv and sdbm were removed because they do not use random seeds and therefore are not useful for MPHFs algorithms.
 - All reported bugs and suggestions have been corrected and included as well.
 ==News for version 0.7==
 - Added man pages and a pkgconfig file.
 [News log newslog.html]
 ----------------------------------------
 ==Examples==
@ -234,7 +237,7 @@ Use the project page at sourceforge: http://sf.net/projects/cmph
 ==License Stuff==
-Code is under the LGPL. 
+Code is under the LGPL and the MPL 1.1. 
 ----------------------------------------
 %!include: FOOTER.t2t
--- a/2
+++ b/2
@ -1,6 +1,7 @@
 #!/bin/sh
 txt2tags -t html --mask-email -i README.t2t -o index.html
 txt2tags -t html -i BDZ.t2t -o bdz.html
 txt2tags -t html -i BMZ.t2t -o bmz.html
 txt2tags -t html -i BRZ.t2t -o brz.html
 txt2tags -t html -i CHM.t2t -o chm.html
@ -12,6 +13,7 @@ txt2tags -t html -i CONCEPTS.t2t -o concepts.html
 txt2tags -t html -i NEWSLOG.t2t -o newslog.html
 txt2tags -t txt --mask-email -i README.t2t -o README
 txt2tags -t txt -i BDZ.t2t -o BDZ
 txt2tags -t txt -i BMZ.t2t -o BMZ
 txt2tags -t txt -i BRZ.t2t -o BRZ
 txt2tags -t txt -i CHM.t2t -o CHM
--- a/papers/fch92.pdf
+++ b/papers/fch92.pdf
--- a/papers/thesis.pdf
+++ b/papers/thesis.pdf
--- a/papers/wads07.pdf
+++ b/papers/wads07.pdf