U.S. Pat. No. 9,201,944, filed Jun. 12, 2013 and titled “Techniques For Maintaining Column Vectors Of Relational Data Within Volatile Memory,” is incorporated herein by reference.
This disclosure relates to in-memory graph formats. Presented herein are techniques for encoding heterogeneous graphs for efficient use.
Property graphs are becoming an increasingly important tool for data analytics, due to their explicit and fine-grained representation of relationships between entities. A property graph is composed of vertices, which represent entities, and relationships, which connect the vertices and may be either directed (i.e. edges) or not (i.e. arcs). Vertices, edges, and arcs may have properties or attributes that describe them. In some techniques, with a property graph, vertices and edges are untyped and have a same universal set of properties, which may be NULL or have a default and/or invalid value for some vertices or edges.
In that way, a property graph is well suited to handle completely unstructured data. However, in real-world graphs, vertices and edges often have a natural concept of a type, i.e., a specific set of properties that is defined for each of them (e.g. a vertex type representing a person, with properties ‘name’, ‘date of birth’, and ‘address’, and another vertex type representing a book, with properties ‘title’, ‘author’, and ‘ISBN’). Such graphs are heterogeneous that contain more than one kind of vertex and/or edge.
The problem we tackle with this invention is how to represent heterogeneous graphs in-memory in an efficient way that is amenable to support graph analysis algorithms and queries. Current graph analytics systems generically implement a property graph model, i.e., they treat vertices, edges, and arcs as untyped. However, that approach results in two inefficiencies when dealing with heterogeneous graphs. First is inefficient memory utilization, because all properties that are not part of the vertex or edge type will still have a default or NULL value for that element, thus consuming extra memory.
Second is inefficient memory access when performing computations that only access a subset of the types, because vertices (or edges) of a whole graph and regardless of their semantic differences, are consolidated into a same collection of all vertices (or edges). Additionally, commercial systems typically do not support mixed graphs that have both directed edges and undirected arcs.
In the drawings:
In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, that the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form in order to avoid unnecessarily obscuring the present invention.
Embodiments are described herein according to the following outline:
1.0 General Overview
2.0 Example Computer
3.0 Example Vertex Loading Process
4.0 Example Query Process
5.0 Globally Unique Identifiers
6.0 Topology Encoding
7.0 Example Topology Encoding Process
8.0 Edge Property Storage
9.0 Populating Edge Properties
10.0 Bidirectional Traversal
11.0 Example Reverse Encoding Process
12.0 Undirected Arcs
13.0 Bidirectionally Encoding Arcs Having Homogeneous Endpoints
14.0 Hardware Overview
15.0 Software Overview
16.0 Cloud Computing
1.0 General Overview
Techniques are provided herein for efficient representation of heterogeneous graphs in memory. In an embodiment, vertices and edges of the graph are segregated by type. Each property of a type of vertex or edge has values stored in a respective vector. Directed or undirected edges of a same type are stored in compressed sparse row (CSR) format. The CSR format is more or less repeated for edge traversal in either forward or reverse direction. An edge map translates edge offsets obtained from traversal in the reverse direction for use with data structures that expect edge offsets in the forward direction. Subsequent filtration and/or traversal by type or property of vertex or edge entails minimal data access and maximal data locality, thereby increasing efficient use of the graph.
In an embodiment, a computer associates each vertex of a graph, which resides in memory, with a unique offset. For each relationship of a plurality of relationships in a data source, the following occurs. The relationship becomes associated with the following objects that reside in the memory: a distinct edge type in the graph for connecting a source vertex of a particular source vertex type with a destination vertex of a particular destination vertex type, a distinct source vector that is sorted, and a distinct destination vector that is sorted. For each instance of the relationship, a directed edge, which represents the instance of the relationship, is created, in the graph, as an instance of the distinct edge type of the relationship. The directed edge comprises a source vertex of the graph and a destination vertex of the graph. For each vertex of the particular source vertex type the following occurs. For each directed edge, of the distinct edge type, that has said vertex as its source vertex, the computer stores, into the distinct destination vector, the unique offset of the destination vertex of the directed edge. Into the distinct source vector, at the unique offset of said vertex, a last position, within the distinct destination vector, that contains a unique offset is stored. Subsequently, a reading of the graph is accelerated by accessing the distinct source vector of a relationship of a distinct edge type and/or the distinct destination vector of the relationship.
2.0 Example Computer
Computer 100 may load and process graph 160, such as for querying, mutation, or analytics. Graph 160 is a logical graph of interconnected vertices, such as vertices 181-182. Vertices 181-182 may have properties, such 186-187, and graph 160 may be a property graph.
2.1 Data Source
Graph 160 is a logical view of underlying data that is provided by data source(s) such as 110. Data source 110 may be a durable data store, such as a relational database, a columnar database, a property store, a resource description framework (RDF) triplestore, a spreadsheet, a Hadoop file system (HDFS) file, or a data file. Data source 110 may spontaneously generate data, such as when data source 110 is a computer program, a web service, or a data feed such as a stream.
Data source 110 contains records, such as 141-142, that may be tuples, rows, or other structural units. Records 141-142 may aggregate data fields, such as 131-132, that store more or less atomic values. Data source 110 may have records, such as 141-142, of a same type, such as 122, or of different types such as 121-122. Record type 122 may have data fields 131-132, while record type 121 may have other data fields. In an embodiment, data source 110 is a relational database; record types 121-122 are tables in the relational database; records 141-142 are rows in a database table; and data fields 131-132 are columns in that database table.
As shown, graph 160 and data source 110 reside in a same computer 100. However, data source 110 may be remote (i.e. external to computer 100), such as residing in a cross mounted file system, network attached storage (NAS), a storage area network (SAN), or a remote service such as a relational database management system (RDBMS) accessed through open database connectivity (ODBC) or a web service accessed through hypertext transfer protocol (HTTP) according to simple object access protocol (SOAP), common gateway interface (CGI), and/or representational state transfer (REST). Data source 110 may serialize records 141-142 for consumption by computer 100 according to an encoding format such as extensible markup language (XML), JavaScript object notation (JSON), comma separated values (CSV), Java object serialization, or a proprietary/native binary format.
2.2 Heterogeneous Graph
Graph 160 is heterogeneous, which means that vertices, such as 181-182, may have a same type, such as 172, or different types such as 171-172. Computer 100 may have a mapping, such as marshalling or data binding logic, that causes subsets of vertex types 171-172 to be generated from subsets of record types 121-122. For example, one record type, such as 122, may map to one vertex type, 172. Other mappings are possible, such as one record type to many vertex types, many record types to one vertex type, or many to many.
2.3 Random Access Memory
Fast access to graph 160 may require low latency storage, such as random access memory (RAM), which may be volatile, such as 105, such as cheap and dense dynamic RAM (DRAM) or fast static RAM (SRAM). In an embodiment not shown, memory 105 is non-volatile (NVRAM), such as flash. Because RAM 105 typically has less capacity than disk storage due to capital cost, computer 100 uses a dense (i.e. space efficient) storage format for vertices 181-182.
2.4 Vertex Properties
The vertex storage format is columnar for vertex properties, such as 186-187, which are shredded (i.e. separated by property), such that each property is stored into a distinct property vector, such as 191-192. As shown, fields 131-132 respectively correspond to properties 186-187. Each of vertices 181-182 may have their own values for properties 186-187. For example, vertex 181 may have been generated from record 141 whose field 131 value is red. Thus, the value of color property 187 may be red for vertex 181. Likewise, vertex 182 may have a color 187 value of blue because record 142 has a value of blue for field 131. Thus, records of a same type have a same set of fields, and vertices of a same type have a same set of properties.
All of the values (e.g. red, blue) of color property 187 for vertices 181-182 of vertex type 172 are stored in property vector 191. Likewise, all of the values of property 186 are stored in vector 192. Some embodiments may have various ways of compressing vector 191, such as bit or byte packing (i.e. no padding for alignment), dictionary encoding, or run length encoding. Property vector storage may also accelerate processing, such as by reducing virtual memory swapping or by facilitating burst mode processing such as with single instruction multiple data (SIMD) and/or slicing register(s) (i.e. packing multiple values from vector 191 into a same central processing unit, CPU, register). Techniques for property vectors (a.k.a. column vectors) are detailed in U.S. Pat. No. 9,201,944, titled “Techniques For Maintaining Column Vectors Of Relational Data Within Volatile Memory,” which is incorporated herein by reference.
2.5 Vertex Encoding
Each vertex of vertex type 172 may have a (e.g. zero based) distinct serial number. A sequence of serial numbers may be reused for different vertex types, so long as each serial number is unique within a given vertex type. For example, both of vertex types 171-172 may have its own vertex with a same serial number, such as zero. Although not shown, each of vertex types 171-172 may have an array of vertices of that type. The offset of a given vertex within that array may be used as the (e.g. implied) serial number of that vertex. For example, vertices 181-182 both are of vertex type 172 and may reside in a same vertex array at different offsets, such as zero and one respectively.
For demonstrative purposes, property vector 191 is shown as having two columns, offset and value. However, so long as property values are stored in property vectors 191-192 for vertices 181-182 in a same ordering as vertices 181-182 are stored in the vertex array (not shown) of vertex type 172, then the offset (e.g. serial number) of vertex 182 in the vertex array may be used for random access into property vectors 191-192 for vertex 182. Thus, the shown offset column of vector 191 may be an implied column that is not actually stored. Thus in an embodiment, the only data that vector 191 contains are values red and blue.
3.0 Example Vertex Loading Process
The depicted process is performed by computer 100 in two phases, loading and use. During loading, the graph is created in memory based on data from data source(s). Loading occurs during steps 201-205, which are repeated for each record type provided by the data source(s) for the graph. Loading may be parallelized per data source or per record type.
Loading occurs in two phases, preparation and encoding. Preparation occurs during steps 201-202, during which empty data structures are created and linked within memory. Step 201 associates each record type with a distinct vertex type in a graph being constructed in memory.
In an embodiment, a static data binding mapping is consulted that indicates that record types 121-122 are respectively mapped to vertex types 171-172. In an embodiment without a static data binding, data source 110 may be interrogated and record types 121-122 may be discovered. From metadata about record types 121-122, vertex types 171-172 may be generated.
In an embodiment, a mapping of record types to vertex types need not be one-to-one. For example rows may be segregated, such as when record type 122 maps to both of vertex types 171-172 such that red records, such as 141, map to vertex type 171 and blue records, such as 142, map to vertex type 172. In another example fields may be separated, such as when field 131, color 187, and vertex type 172 are bound as shown, and although shown otherwise, field 132 and property 186 may be bound to vertex type 171. In another example, both of record types 121-122 map to vertex type 172.
Step 202 associates each data field in the record type with a vertex property vector that resides in memory. For example, a binding or inference may determine that data fields 131-132 map to vertex properties 186-187 as shown. Based on that information, empty vertex property vectors 191-192 are created.
Vertex encoding occurs during steps 203-205, which are repeated for each record of the record type. A vertex or subset of vertices may be a unit of parallelism during encoding. Each repetition of steps 203-205 populates one vertex. In an embodiment, empty graph 160 is created in step 201 or 202, and vertices may be added to graph 160 as each vertex is initialized. In an embodiment, empty graph 160 is not created until later, but not later than immediately after step 205, and no vertices are added to graph 160 until all vertices are initialized.
Step 203 creates a vertex, having a vertex type, from a record having a record type that corresponds to the vertex type. For example, vertex 181 is created from record 141.
Step 204 assigns a unique offset integer to the vertex. For example, vertex 181 may be stored into a vertex array (not shown) in memory at a distinct offset within the array. That offset may be used to identify the vertex as distinct in that vertex type, although not necessarily distinct across multiple vertex types. Globally unique identifiers are discussed later herein.
Step 205 stores vertex properties. A value from each data field of the record is stored into a distinct vertex property vector associated with that field. Values are stored into property vectors at a same offset as was assigned to the vertex in step 204. For example, the value blue is copied from field 131 of record 142 into color property 187 of vertex 182, which is stored at offset one within vector 191.
After step 205, graph loading is finished, and graph 160 is ready to use. In an embodiment, graph 160 is made read only after step 205.
Graph use occurs during step 206. For example, a property of a vertex may be read. That read is accelerated due to the streamlined storage format as described above. Both of random access of a property for one vertex or scanning the property for many vertices are accelerated. For example, computer 100 may receive and execute a query for graph 160 to find individual vertices or multi-vertex paths or subgraphs based on vertex properties. Query processing is further discussed in
4.0 Example Query Process
In this example, query execution entails graph navigation, which entails repeated traversal from one vertex to another along an edge. Like vertices, edges may have types. Each edge connects two vertices, or one vertex to itself. Like vertices, edges may have types, properties, and property vectors with offsets aligned to an edge vector of an edge type. Edge storage and use are discussed later herein.
Step 301 is preparatory. Step 301 presumes a graph loaded in memory per
Step 301 associate vertex types with edge types and loads edges. Graph traversal depends on graph topology because graph traversal occurs outwards through edges of a vertex. Thus, step 301 establishes which edge type (and edges) are available when traversing from a given vertex.
Steps 302-306 may occur during step 206 of
4.1 Query Acceleration
Efficient execution of the query occurs during steps 303-306. Step 303 detects which types and properties of vertices (and edges) are relevant to the query. Step 303 may construct and semantically analyze a parse tree (AST) of the query. At a minimum, types and properties of vertices and edges are relevant if recited in the query (and/or appear in the AST).
Steps 304-306 are repeated for each vertex type. Step 304 detects whether or not the vertex type participates in the query. Steps 303-304 may contribute to formulating a query execution plan.
According to step 304, irrelevant vertex types are skipped, which is efficient. In an embodiment that lazily loads vertices and/or vertex types, skipping may mean deferring or avoiding loading some vertices. Likewise, vertex properties may be individually irrelevant, skipped, and not loaded. For example, a first query of a cold (i.e. incompletely loaded) graph may cause vertex property vector 191 to be loaded, but not vector 192, because only color 187 is accessed by the query, and not property 186.
Step 305 accesses relevant property vectors of the vertex type. Step 305 may entail random access to properties of particular vertices or scanning of properties of many vertices. For example, multiple values contiguously stored in vertex property vector 191 may be simultaneously loaded (e.g. packed) into a same CPU register and/or processed together by single instruction multiple data (SIMD) or other vector hardware acceleration, such as with a graphics processing unit (GPU).
4.2 Improved Performance
Thus, efficiencies of space savings and acceleration arise at two levels. First, heterogeneity of vertex types facilitates minimizing how many vertices are accessed by the query or even loaded. Second, column (i.e. property) shredding and storage in separate vectors facilitates both of: a) minimizing how many property values are accessed by the query or even loaded, b) property vector compression such as by dictionary or run-length encoding, which may occur during or after loading a property vector, and c) vector data parallelism such as by SIMD or GPU.
These many efficiencies are more or less unavailable with graph loading techniques that treat vertices and edges as homogeneous or generalized. Various space savings (e.g. loading avoidance, reading avoidance, vector packing and/or other vector compression) may cause various levels of acceleration such as decreased virtual memory thrashing, increased hardware cache locality, decreased bus saturation or wait states, and hardware data parallelism.
Edge traversal occurs during step 306, which inspects edges of vertex(ices) to steer along a path or prune a subgraph. Step 306 may facilitate exploration algorithms such as breadth first search (BFS) or depth first search (DFS), because edge traversal facilitates iteration of neighboring vertices. Edge storage and access techniques are discussed later herein.
5.0 Globally Unique Identifiers
5.1 Ram Content
RAM 405 stores a heterogeneous graph (not shown) that includes vertex type 470, which may have vertices, such as 480, and various vertex properties that may each be stored in a property vector such as 491-492. Each of vectors 491-492 store property values of multiple vertices. The contents of vectors 491-492 have a same ordering by vertex as does the vertex array (not shown) of vertex type 470. For example, vertex 480 may be a first vertex (i.e. offset zero) in the vertex array. Thus, the top row (i.e. offset zero) of vectors 491-492 have property values for vertex 480.
5.2 Identifying Pair
Thus, a zero is recorded in the offset field within identifying pair 420. Identifying pair 420 uniquely and globally identifies vertex 480. However, a vertex offset is insufficient to uniquely and globally identify a vertex. Thus, identifying pair 420 also has an additional field to ensure global uniqueness. The additional field should identify which type is vertex 480. In an embodiment, the additional field records the name of vertex type 470. In an embodiment, the additional field is a serial number that uniquely identifies vertex type 470. Vertex type 470 has serial number 410 with a value of five, which is recorded in the identifying pairs of vertices such as 480.
The serial number and offset fields of identifying pair 420 may be integers. The serial number need not have a same width (i.e. storage size) as the offset, and they may be byte packed or bit packed together, such that the width of identifying pair 420 comprises the sum of the widths of the serial number and offset fields. For speed, the width of identifying pair 420 should not exceed a machine word (e.g. CPU register width), which usually is 64 bits. A graph with too many vertices may need as many as 128 bits (i.e. double word) for identifying pair 420.
A more or less same technique may be used to globally identify edges that connect vertices. RAM storage of edges is discussed later herein. Besides
6.0 Topology Encoding
6.1 Relationships
Computer 500 loads graph 560 into RAM 505 based on data from data source 510 as follows. In addition to providing records (not shown) for vertices 0-2, data source 510 also provides relationship 520 for creating edges 0-4. Each instance 0-4 of relationship 520 provides data for respective edges 0-4. For example as shown, edge 1 is generated from instance 1.
For demonstration, relationship 520 is depicted as tabular. Each row represents one instance of relationship 520. Relationship 520 has columns 530 and 540 that identify the directed endpoint vertices of each instance. For example, instances 1-2 are edges in opposite directions between a same two vertices. Instance 4 is reflexive (i.e. connects a vertex to itself).
In an embodiment, data source 520 is a relational database. Relationship 520 may be encoded within a relational database table in various ways based on one or two foreign keys in one or three tables as follows. Each of instance 0-4 connects (i.e. joins) two tables.
6.2 Foreign Key
For example, columns 530 and 540 may identify rows of respective source and destination tables (not shown). Source 530 may be a primary key column within the source table. Destination 540 may be replicated as two columns: a) a primary key column in the destination table, and b) a foreign key column in the source table. This may achieve a one-to-one or one-to-many relationship. In another example, source 530 may instead be the replicated column, and the destination table may store the foreign key column. When the source and destination tables are a same table, then reflexivity such as with instance 4 is possible.
Another example may involve a third table as an associative table. For example relationship 520 may be an associative table that is separate from the source and destination tables. As such, both of columns 530 and 540 are foreign keys that cross reference the source and destination tables respectively. This may achieve a one-to-one, one-to-many, or many-to-many relationship. Regardless of how many tables and keys are used for relationship 520, the instance column might be implied (i.e. demonstrative and not actually stored).
6.3 Endpoint Vectors
Edges 0-4 are created from instances 0-4 by populating edge endpoint vectors 570 and 580 and a set of property vectors (not shown). Edge property vectors are encoded in a similar way to vertex property vectors, as described above. For example, there may be an array (not shown) of edges for each type of edge. The contents of edge property vectors may share the edge array's ordering of edges.
Edge endpoint vectors 570 and 580 are encoded based on a special ordering of edges as follows. The special ordering of edges 0-4 depends on the ordering of vertices 0-2 in the vertex array (not shown). The following discussion of
7.0 Example Topology Encoding Process
The depicted process is performed by computer 500 in two phases, loading and use. During loading, the graph is created in memory based on data from data source(s). Loading occurs during steps 601-605, which are repeated for each relationship provided by the data source(s) for the graph. Loading may be parallelized per data source or per relationship.
Loading occurs in two phases, preparation and encoding. Preparation occurs during steps 601-602, during which empty data structures are created and linked within memory. Step 601 associates each vertex of a graph with an offset that is unique to the type of vertex, although not necessarily globally unique. In an embodiment, RAM 505 is an implementation of memory 105 of
Steps 602-605 are repeated for each relationship in data source(s). With the relationship, step 602 associates a distinct edge type that connects a source vertex type to a destination vertex type, which need not be a same vertex type. Also with the relationship, step 602 associates a source vector and a destination vector. For example for an edge type associated with relationship 520, endpoint vectors 570 and 580 are created. Vectors 570 and 580 are initially empty. Subsequent steps populate vectors 570-580 according to special orderings. In that sense, contents of vectors 570 and 580 are effectively sorted, without actually invoking a sorting algorithm.
Edge encoding occurs during steps 603-605. Step 603 is repeated for each instance of the relationship. Each repetition of step 603 may populate one edge's properties within edge property vectors (not shown). In an embodiment, empty graph 560 is created in step 601, and edges may be added to graph 560 as each edge is initialized. In an embodiment, empty graph 560 is not created until later, but not later than immediately after step 605, and no edges are added to graph 560 until all edges are initialized.
Step 603 creates an edge, having an edge type, from an instance of the relationship that corresponds to the edge type. For example, edge 1 is created from instance 1. In an embodiment, the edge is stored in an array for that edge type.
7.1 Destination Vector
Edge encoding and special ordering within vectors 570 and 580 are achieved by steps 604-605 as follows. Steps 604-605 are repeated for each vertex of the source vertex type. Step 604 is repeated for each edge of the edge type that has that vertex as a source vertex. In other words, step 604 is repeated for each edge of the edge type that originates from that vertex.
For example, vertex 0 is the source vertex of edges 0-1, which are of a same edge type. Although edge 2 has the same edge type and also connects to vertex 0, the source vertex of edge 2 is not vertex 0. Edge 0 connects source vertex 0 to destination vertex 1.
Into the next sequential (i.e. vacant) position within the destination vector of the edge type, step 604 stores the value of the offset that was assigned to the destination vertex in step 601. For example, vertex 1 is stored at zero-based offset one of a vertex array (not shown) of a vertex type. For edge 0, step 604 appends a one to empty destination vector 580. Because destination vector 580 is empty, the one is appended (i.e. stored) at position zero of destination vector 580. Also from same source vertex 0 as edge 0, is edge 1 that has vertex 2 as its destination vertex, a two is appended to the next position of destination vector 580, which is at zero-based offset 1.
Thus edges, of the edge type, from a same source vertex are contiguously recorded in destination vector 580, after which, step 604 repeats for a next vertex of the same source vertex type. The next source vertex may arise during a sequential iteration of the vertex array (not shown) of that vertex type. Thus, the contents of destination vector 580 are implicitly sorted by source vertex.
7.2 Source Vector
Step 605 stores, into the source vector of the edge type, at the same offset as was assigned to the source vertex in step 601, the last position, within the destination vector, that contains a unique offset of a destination vertex of a directed edge that has the vertex as source vertex. That last position should be the last position that was written to in step 604. For example for source vertex 0, positions 0-1 were written to within destination vector 580, with 1 being the last position written. Thus, step 605 stores a 1 into source vector 570 at the same offset as was assigned to vertex 0, which is offset 0.
Whereas next vertex 1 of the source vertex type is not a source vertex for any edge of the edge type. In other words, no edge of the edge type originates from vertex 1. Thus, step 604 records nothing into destination vector 580 for vertex 1. Thus, the last written position within destination vector 580 is the same for vertices 0-1. Thus, step 605 writes the same value 1 into source vector 570, but at the offset assigned to vertex 1, which is 1.
After step 605, population of endpoint vectors 570 and 580 is finished. Source vector 570 has as many values as there are vertices of the source vertex type of the edge type. By design and without further sorting, values within source vector 570 are ordered by source vertex and are monotonically increasing.
7.3 Performance Improvement
Destination vector 580 has as many values as there are edges of the edge type. By design and without further sorting, values within destination vector 580 too are ordered by source vertex. However, those values (e.g. 1, 2, 0, 1, 2 as shown) are clearly not naturally sorted. As shown, edges are recorded within destination vector 580 in their natural ordering of 0-4. The efficiency of these special orderings of the contents of endpoint vectors 570 and 580 manifests in step 606 as follows.
Graph use occurs during step 606. For example, an edge may be traversed. That traversal is accelerated due to the streamlined storage format as described above. Both of random access for traversal of one edge, such as for probing during a depth first search, or traversal of many edges, such as for reaching neighboring vertices during a breadth first search, are accelerated. For example, computer 500 may receive and execute a query for graph 560 to find multi-edge paths or subgraphs based on types and properties of edges and neighboring vertices.
For example, source vertex 2 originates edges 2-4. Endpoint vectors 570 and 580 may be used for fast traversal of those edges during step 606 as follows. The offset assigned to vertex 2 during step 601 is 2. At offset 2 within source vector 570 is value 4, with a preceding value of 1 at offset 1 within source vector 570. Four minus one is three. Thus, vertex 2 originates three directed edges, which are recorded within destination vector 580 after (i.e. excluding) offset 1 and until (i.e. including) offset 4, because values 1 and 4 are contiguously stored in source vector 570. One exclusive through four inclusive is the same as two inclusive through four. Thus, the three edges originating from vertex 2 are recorded within destination vector 580 at offsets 2-4, which stores values 0, 1, 2, which are the identifying offsets that were assigned to destination vertices 0-2 in step 601. Thus, destination vertices 0-2 are immediate neighbors of source vertex 2 by way of edges of the given edge type.
Because each edge type has its own source vector and destination vector, filtration and/or traversal by edge type entails minimal data access and maximal data locality, thereby increasing efficient reading of graph 560. Edge properties stored in property vectors may efficiently be randomly or sequentially accessed during edge traversal or iteration if the property vectors have an edge ordering that matches the ordering of destination vector 580.
Endpoint vectors 570 and 580 together achieve compressed sparse row (CSR) encoding of graph 560's topology (i.e. edges), which is space and time efficient for traversal and filtration of edges. In an embodiment, vertex offsets recorded in endpoint vector 570 and/or 580 are bit or byte packed (i.e. not byte aligned or not word aligned, respectively for increased density). However, CSR may be inefficient to adjust, such as when graph 560's topology evolves by addition or removal of an edge. Thus, endpoint vectors 570 and 580 may become read only in an embodiment. While read only endpoint vectors may more or less prevent topological mutation, edge property vectors may still be writable, such that edge property values may change. For example, a value of a weight property of an edge may change, even if the edge cannot be removed. For example, removal of an edge may be approximated by setting its weight property to zero or setting its deletion flag property to true. Edge property vectors are discussed further in
8.0 Edge Property Storage
Computer 700 populates edge property vector 790 within RAM 705 based on data from instances 0-1 of data source 710 as follows. Data source 710 provides relationship 720 that includes data fields 751-752. A separate edge of type 770 may be created for each of instances 0-1 to populate an edge table (not shown) of edge type 770. Values within field 751 of instances 0-1 are copied into vector 790 at offsets 0-1 according to a same edge ordering as the edge table, which facilitates random access into vector 790 for a given edge. The offset column of vector 790 may be implied (i.e. demonstrative and not stored). Thus, edge property vector 790 need only store values red and blue, perhaps without any metadata. Edge loading is further discussed in
9.0 Populating Edge Properties
Step 802 associates a data field of a relationship of a data source with an edge type and a property vector. For example, edge type 770 may become associated with empty property vector 790 and data field 751.
Steps 804 and 806 are repeated for each instance, such as 0-1 of
Step 804 associates the edge with an offset that is unique for the edge type, though not necessarily across all edge types. For example, the edges are stored into an edge array (not shown) at their offsets.
Step 806 copies a value of a data field of the relationship instance into a property vector, which is associated with the data field, at the offset assigned to the edge. For example, a red value is copied from data field 751 of instance 0 into edge property vector 790 at offset 0 because 0 is the offset assigned to the edge (not shown) generated from instance 0. For example, that edge is the first edge in the zero-based edge array of edge type 770.
10.0 Bidirectional Traversal
10.1 Reverse Endpoint Vectors
Reverse endpoint vectors 970 and 980 are analogous to direct endpoint vectors 570 and 580 (not shown in
Population of reverse data structures 910, 970, and 980 are described later herein for
11.0 Example Reverse Encoding Process
Step 1002 associates the edge type with a distinct reverse source vector and a distinct reverse destination vector. For example, empty reverse endpoint vectors 970 and 980 are created within RAM 505.
Steps 1004 and 1006 are repeated for each vertex, such as 0-2, of the source vertex type of the edge type. A prerequisite of step 1004 is that at least step 603 of
Edge reverse encoding and special ordering within vectors 970 and 980 are achieved by steps 1004 and 1006 as follows. Steps 1004 and 1006 are repeated for each vertex of the destination vertex type of the edge type. Step 1004 is repeated for each edge of the edge type that has that vertex as a source vertex. In other words, step 1004 is repeated for each edge of the edge type that terminates at that vertex.
For example, vertex 0 is the destination vertex of edge 2. Although edges 0-1 have the same edge type and also connect to vertex 0, the destination vertex of edges 0-1 are not vertex 0. Edge 2 connects destination vertex 0 to source vertex 2.
11.1 Reverse Destination Vector
Into the next sequential (i.e. vacant) position within the reverse destination vector of the edge type, step 1004 Stores the Value of the Offset that was Assigned to the Source Vertex in step 601 of
Edges, of the edge type, to a same destination vertex are contiguously recorded in reverse destination vector 980, after which, step 1004 repeats for a next vertex of the same destination vertex type. The next destination vertex may arise during a sequential iteration of the vertex array (not shown) of that vertex type. Thus, the contents of reverse destination vector 980 are implicitly sorted by destination vertex.
11.2 Reverse Source Vector
Step 1006 stores, into the reverse source vector of the edge type, at the same offset as was assigned to the destination vertex in step 601 of
After step 1006, population of reverse endpoint vectors 970 and 980 is finished. Reverse source vector 970 has as many values as there are vertices of the destination vertex type of the edge type. By design and without further sorting, values within reverse source vector 570 are ordered by destination vertex and are monotonically increasing.
11.3 Edge Offset Translation
As explained for
In an embodiment that lacks map 910, step 1008 is not performed. Step 1008 creates and populates offset translation map 910. For each edge of the edge type, step 1008 stores into an edge mapping array, in a same ordering of edges as the reverse destination vector, the position, within the direct destination vector, of the edge. For example, edge 2 is the first/only edge, of the edge type, that has vertex 0 as its destination vertex. Edge 2 has vertex 2 as its source vertex. Thus, two is recorded into offset 0 of reverse destination vector 980. The offset 0 may be used as an index into map 910, per implied (i.e. demonstrative, not actually stored) column 920. Thus, the first row of actual data in map 910 is for edge 2, as indicated by column 930, which is the only column that need actually be stored for map 910.
12.0 Undirected Arcs
Heterogeneous graph 1160 contains undirected arcs and directed edges of various types including a drives edge type, a cooperates arc type, and a co-located arc type. Drives directed edge E-0 indicates that person P-0 is the driver of car C-0. A person can drive a car, but a car cannot drive a person. Thus, the drives edge type is necessarily directed due to the semantics (e.g. schema) of graph 1160.
Cooperates arc A-0 indicates that persons P-1 and P-2 mutually cooperate with each other. Mutual cooperation has no notion of direction. Thus, the cooperates arc type is necessarily undirected due to the semantics of graph 1160. Co-located arc L-0 is also undirected and indicates that person P-1 and car C-0 are in a same spatial location (not shown).
Instances of each of the connection (i.e. edge or arc) types of graph 1160 has enough endpoint vectors to achieve bidirectional (e.g. direct and reverse) access. For example, drives edge type has direct source and destination vectors (not shown) and reverse source and destination vectors (not shown), for a total of four endpoint vectors for bidirectional access.
Likewise, co-located arc type has a similar set of four endpoint vectors 1171-1172 and 1181-1182 for bidirectional access, although the comparative semantic notions of source and destination are absent because arcs are naturally undirected. Thus although vectors 1171 and 1181 operate together as a pair, and vectors 1172 and 1182 operate together as another pair, which pair is direct and which is reverse is logically undefined. However in practice, one of the vector pairs may have a same ordering of arcs as an arc array (not shown) of the co-located arc type. That vector pair may be direct vectors, and the other pair may be reverse vectors. Although not shown, the co-located arc type may have a translation map for use with the reverse pair, as described for
12.1 Arcs with Homogeneous Endpoints
When both endpoint vertices of an undirected arc type have a same vertex type, then bidirectional arc access may be achieved with only one pair of endpoint vectors and no translation map as follows. For example, cooperates arc type connects two persons, such as persons P-0 and P-1 connected by cooperates arc A-0. In this case, endpoint vectors 1175 and 1185 are sufficient for bidirectional access to cooperates arcs. Unlike endpoint vectors 1171-1172 and 1181-1182 that each have one value (shown dark) per undirected arc, each of endpoint vectors 1175 and 1185 have two values (shown dark) per undirected arc.
Legend 1110 indicates that only the dark values of endpoint vectors 1171-1172, 1175, 1181-1182, and 1185 have actual data. The light value in the endpoint vectors indicate implied data not actually stored, which are offsets into those endpoint vectors. Bidirectional access occurs as follows for cooperates arc A-0. If exploring from person P-0, then the left dark column of cooperates source vector 1175 is used. If exploring from person P-1, then the right dark column is used. Population of endpoint vectors 1171-1172 and 1181-1182 are described for
13.0 Bidirectionally Encoding Arcs Having Homogeneous Endpoints
13.1 Forward Encoding
Step 1202 loads edges, vertices, and properties more or less as described in
However, undirected arcs having homogeneous endpoints (i.e. arc type has same vertex type for both endpoints) entail special loading as follows. In step 1202, reverse endpoint vectors and map 910, per
13.2 Backward Encoding
In step 1204, for each arc of a given arc type, a unique offset of a source vertex is stored into the arc type's destination vector. For that arc, the offset of the destination vertex was already stored into the destination vector during step 1202. For example, cooperate arc A-0 was stored as traversable from person P-1 to person P-0. Whereas, step 1204 instead records the source vertex into the destination vector. For example during step 1204, the second column (i.e. offset 1) of cooperates destination vector 1185 stores value 1 to indicate that cooperate arc A-0 may be traversed in an opposite direction from person P-0 to person P-1.
In step 1206, for each vertex having the particular source vertex type, the computer stores, into the source vector, at a unique offset of the vertex, the last position, within the destination vector, that contains a unique offset of a source vertex of an undirected arc that has the vertex as a destination vertex. For example, the last position in cooperates destination vector 1185 that was written to (during step 1204) is offset 1. Thus, step 1206 stores a 1 into the second column's value in cooperates source vector 1175. Thus, endpoint vectors 1175 and 1185 as shown support traversing cooperates arc A-0 in either direction. After step 1206, bidirectional storage of undirected arcs that have homogeneous endpoint types is complete, at which time endpoint vectors 1175 and 1185 may optionally become read only.
14.0 Hardware Overview
According to one embodiment, the techniques described herein are implemented by one or more special-purpose computing devices. The special-purpose computing devices may be hard-wired to perform the techniques, or may include digital electronic devices such as one or more application-specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) that are persistently programmed to perform the techniques, or may include one or more general purpose hardware processors programmed to perform the techniques pursuant to program instructions in firmware, memory, other storage, or a combination. Such special-purpose computing devices may also combine custom hard-wired logic, ASICs, or FPGAs with custom programming to accomplish the techniques. The special-purpose computing devices may be desktop computer systems, portable computer systems, handheld devices, networking devices or any other device that incorporates hard-wired and/or program logic to implement the techniques.
For example,
Computer system 1300 also includes a main memory 1306, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 1302 for storing information and instructions to be executed by processor 1304. Main memory 1306 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor 1304. Such instructions, when stored in non-transitory storage media accessible to processor 1304, render computer system 1300 into a special-purpose machine that is customized to perform the operations specified in the instructions.
Computer system 1300 further includes a read only memory (ROM) 1308 or other static storage device coupled to bus 1302 for storing static information and instructions for processor 1304. A storage device 136, such as a magnetic disk or optical disk, is provided and coupled to bus 1302 for storing information and instructions.
Computer system 1300 may be coupled via bus 1302 to a display 1312, such as a cathode ray tube (CRT), for displaying information to a computer user. An input device 1314, including alphanumeric and other keys, is coupled to bus 1302 for communicating information and command selections to processor 1304. Another type of user input device is cursor control 1316, such as a mouse, a trackball, or cursor direction keys for communicating direction information and command selections to processor 1304 and for controlling cursor movement on display 1312. This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), that allows the device to specify positions in a plane.
Computer system 1300 may implement the techniques described herein using customized hard-wired logic, one or more ASICs or FPGAs, firmware and/or program logic which in combination with the computer system causes or programs computer system 1300 to be a special-purpose machine. According to one embodiment, the techniques herein are performed by computer system 1300 in response to processor 1304 executing one or more sequences of one or more instructions contained in main memory 1306. Such instructions may be read into main memory 1306 from another storage medium, such as storage device 136. Execution of the sequences of instructions contained in main memory 1306 causes processor 1304 to perform the process steps described herein. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions.
The term “storage media” as used herein refers to any non-transitory media that store data and/or instructions that cause a machine to operation in a specific fashion. Such storage media may comprise non-volatile media and/or volatile media. Non-volatile media includes, for example, optical or magnetic disks, such as storage device 136. Volatile media includes dynamic memory, such as main memory 1306. Common forms of storage media include, for example, a floppy disk, a flexible disk, hard disk, solid state drive, magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with patterns of holes, a RAM, a PROM, and EPROM, a FLASH-EPROM, NVRAM, any other memory chip or cartridge.
Storage media is distinct from but may be used in conjunction with transmission media. Transmission media participates in transferring information between storage media. For example, transmission media includes coaxial cables, copper wire and fiber optics, including the wires that comprise bus 1302. Transmission media can also take the form of acoustic or light waves, such as those generated during radio-wave and infra-red data communications.
Various forms of media may be involved in carrying one or more sequences of one or more instructions to processor 1304 for execution. For example, the instructions may initially be carried on a magnetic disk or solid state drive of a remote computer. The remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem. A modem local to computer system 1300 can receive the data on the telephone line and use an infra-red transmitter to convert the data to an infra-red signal. An infra-red detector can receive the data carried in the infra-red signal and appropriate circuitry can place the data on bus 1302. Bus 1302 carries the data to main memory 1306, from which processor 1304 retrieves and executes the instructions. The instructions received by main memory 1306 may optionally be stored on storage device 136 either before or after execution by processor 1304.
Computer system 1300 also includes a communication interface 1318 coupled to bus 1302. Communication interface 1318 provides a two-way data communication coupling to a network link 1320 that is connected to a local network 1322. For example, communication interface 1318 may be an integrated services digital network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, communication interface 1318 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links may also be implemented. In any such implementation, communication interface 1318 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.
Network link 1320 typically provides data communication through one or more networks to other data devices. For example, network link 1320 may provide a connection through local network 1322 to a host computer 1324 or to data equipment operated by an Internet Service Provider (ISP) 1326. ISP 1326 in turn provides data communication services through the world wide packet data communication network now commonly referred to as the “Internet” 1328. Local network 1322 and Internet 1328 both use electrical, electromagnetic or optical signals that carry digital data streams. The signals through the various networks and the signals on network link 1320 and through communication interface 1318, which carry the digital data to and from computer system 1300, are example forms of transmission media.
Computer system 1300 can send messages and receive data, including program code, through the network(s), network link 1320 and communication interface 1318. In the Internet example, a server 1330 might transmit a requested code for an application program through Internet 1328, ISP 1326, local network 1322 and communication interface 1318.
The received code may be executed by processor 1304 as it is received, and/or stored in storage device 136, or other non-volatile storage for later execution.
15.0 Software Overview
Software system 1400 is provided for directing the operation of computing system 1300. Software system 1400, which may be stored in system memory (RAM) 1306 and on fixed storage (e.g., hard disk or flash memory) 136, includes a kernel or operating system (OS) 1410.
The OS 1410 manages low-level aspects of computer operation, including managing execution of processes, memory allocation, file input and output (I/O), and device I/O. One or more application programs, represented as 1402A, 1402B, 1402C . . . 1402N, may be “loaded” (e.g., transferred from fixed storage 136 into memory 1306) for execution by the system 1400. The applications or other software intended for use on computer system 1300 may also be stored as a set of downloadable computer-executable instructions, for example, for downloading and installation from an Internet location (e.g., a Web server, an app store, or other online service).
Software system 1400 includes a graphical user interface (GUI) 1415, for receiving user commands and data in a graphical (e.g., “point-and-click” or “touch gesture”) fashion. These inputs, in turn, may be acted upon by the system 1400 in accordance with instructions from operating system 1410 and/or application(s) 1402. The GUI 1415 also serves to display the results of operation from the OS 1410 and application(s) 1402, whereupon the user may supply additional inputs or terminate the session (e.g., log off).
OS 1410 can execute directly on the bare hardware 1420 (e.g., processor(s) 1304) of computer system 1300. Alternatively, a hypervisor or virtual machine monitor (VMM) 1430 may be interposed between the bare hardware 1420 and the OS 1410. In this configuration, VMM 1430 acts as a software “cushion” or virtualization layer between the OS 1410 and the bare hardware 1420 of the computer system 1300.
VMM 1430 instantiates and runs one or more virtual machine instances (“guest machines”). Each guest machine comprises a “guest” operating system, such as OS 1410, and one or more applications, such as application(s) 1402, designed to execute on the guest operating system. The VMM 1430 presents the guest operating systems with a virtual operating platform and manages the execution of the guest operating systems.
In some instances, the VMM 1430 may allow a guest operating system to run as if it is running on the bare hardware 1420 of computer system 1400 directly. In these instances, the same version of the guest operating system configured to execute on the bare hardware 1420 directly may also execute on VMM 1430 without modification or reconfiguration. In other words, VMM 1430 may provide full hardware and CPU virtualization to a guest operating system in some instances.
In other instances, a guest operating system may be specially designed or configured to execute on VMM 1430 for efficiency. In these instances, the guest operating system is “aware” that it executes on a virtual machine monitor. In other words, VMM 1430 may provide para-virtualization to a guest operating system in some instances.
A computer system process comprises an allotment of hardware processor time, and an allotment of memory (physical and/or virtual), the allotment of memory being for storing instructions executed by the hardware processor, for storing data generated by the hardware processor executing the instructions, and/or for storing the hardware processor state (e.g. content of registers) between allotments of the hardware processor time when the computer system process is not running. Computer system processes run under the control of an operating system, and may run under the control of other programs being executed on the computer system.
16.0 Cloud Computing
The term “cloud computing” is generally used herein to describe a computing model which enables on-demand access to a shared pool of computing resources, such as computer networks, servers, software applications, and services, and which allows for rapid provisioning and release of resources with minimal management effort or service provider interaction.
A cloud computing environment (sometimes referred to as a cloud environment, or a cloud) can be implemented in a variety of different ways to best suit different requirements. For example, in a public cloud environment, the underlying computing infrastructure is owned by an organization that makes its cloud services available to other organizations or to the general public. In contrast, a private cloud environment is generally intended solely for use by, or within, a single organization. A community cloud is intended to be shared by several organizations within a community; while a hybrid cloud comprise two or more types of cloud (e.g., private, community, or public) that are bound together by data and application portability.
Generally, a cloud computing model enables some of those responsibilities which previously may have been provided by an organization's own information technology department, to instead be delivered as service layers within a cloud environment, for use by consumers (either within or external to the organization, according to the cloud's public/private nature). Depending on the particular implementation, the precise definition of components or features provided by or within each cloud service layer can vary, but common examples include: Software as a Service (SaaS), in which consumers use software applications that are running upon a cloud infrastructure, while a SaaS provider manages or controls the underlying cloud infrastructure and applications. Platform as a Service (PaaS), in which consumers can use software programming languages and development tools supported by a PaaS provider to develop, deploy, and otherwise control their own applications, while the PaaS provider manages or controls other aspects of the cloud environment (i.e., everything below the run-time execution environment). Infrastructure as a Service (IaaS), in which consumers can deploy and run arbitrary software applications, and/or provision processing, storage, networks, and other fundamental computing resources, while an IaaS provider manages or controls the underlying physical cloud infrastructure (i.e., everything below the operating system layer). Database as a Service (DBaaS) in which consumers use a database server or Database Management System that is running upon a cloud infrastructure, while a DbaaS provider manages or controls the underlying cloud infrastructure and applications.
The above-described basic computer hardware and software and cloud computing environment presented for purpose of illustrating the basic underlying computer components that may be employed for implementing the example embodiment(s). The example embodiment(s), however, are not necessarily limited to any particular computing environment or computing device configuration. Instead, the example embodiment(s) may be implemented in any type of system architecture or processing environment that one skilled in the art, in light of this disclosure, would understand as capable of supporting the features and functions of the example embodiment(s) presented herein.
In the foregoing specification, embodiments of the invention have been described with reference to numerous specific details that may vary from implementation to implementation. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. The sole and exclusive indicator of the scope of the invention, and what is intended by the applicants to be the scope of the invention, is the literal and equivalent scope of the set of claims that issue from this application, in the specific form in which such claims issue, including any subsequent correction.
Number | Name | Date | Kind |
---|---|---|---|
4613946 | Forman | Sep 1986 | A |
5515487 | Beaudet | May 1996 | A |
5553218 | Li | Sep 1996 | A |
5822592 | Zhu | Oct 1998 | A |
5983215 | Ross et al. | Nov 1999 | A |
6400383 | Geisler | Jun 2002 | B1 |
6774899 | Ryall | Aug 2004 | B1 |
7624081 | Zhao et al. | Nov 2009 | B2 |
7849106 | Agrawal | Dec 2010 | B1 |
8499284 | Pich | Jul 2013 | B2 |
8527909 | Mullany | Sep 2013 | B1 |
8639818 | Vladimirov | Jan 2014 | B1 |
8713063 | Abrams | Apr 2014 | B2 |
8810576 | Duplessis | Aug 2014 | B2 |
8910084 | Helfman | Dec 2014 | B2 |
8937618 | Erez | Jan 2015 | B2 |
8990675 | Folting | Mar 2015 | B2 |
8996492 | Paradies et al. | Mar 2015 | B2 |
9092744 | Tito | Jul 2015 | B2 |
9195941 | Mojsilovic et al. | Nov 2015 | B2 |
9208257 | Milenova | Dec 2015 | B2 |
9262537 | Kim | Feb 2016 | B2 |
9466138 | Singh | Oct 2016 | B2 |
9477732 | Helfman | Oct 2016 | B2 |
9495777 | Srivastava | Nov 2016 | B2 |
9508167 | Kim | Nov 2016 | B2 |
9519677 | Wong | Dec 2016 | B2 |
9600603 | Weise | Mar 2017 | B2 |
9658943 | Gounares | May 2017 | B2 |
9672656 | McPhail | Jun 2017 | B1 |
9679247 | Koutra | Jun 2017 | B2 |
9705815 | Mattson | Jul 2017 | B2 |
9710544 | Smith | Jul 2017 | B1 |
9734040 | Gounares | Aug 2017 | B2 |
9740368 | Love | Aug 2017 | B1 |
9740536 | Van Der Lugt | Aug 2017 | B2 |
9779632 | Shakeri | Oct 2017 | B2 |
9826183 | Love | Dec 2017 | B2 |
9836183 | Love | Dec 2017 | B1 |
9836508 | Morton | Dec 2017 | B2 |
9836524 | Bayliss | Dec 2017 | B2 |
9864672 | Seto | Jan 2018 | B2 |
9886521 | Wu | Feb 2018 | B2 |
9892531 | Csenteri | Feb 2018 | B2 |
9911211 | Damaraju | Mar 2018 | B1 |
9984109 | Da Silva | May 2018 | B2 |
9996236 | Spencer-Harper | Jun 2018 | B1 |
10061841 | Xiong | Aug 2018 | B2 |
10120956 | Chen | Nov 2018 | B2 |
10140121 | Ravindhran | Nov 2018 | B2 |
10268735 | White | Apr 2019 | B1 |
10347017 | Ruble | Jul 2019 | B2 |
10459978 | Hong | Oct 2019 | B2 |
10585575 | Kindelsberger | Mar 2020 | B2 |
10599656 | Sharma | Mar 2020 | B1 |
10606892 | Broecheler | Mar 2020 | B1 |
10672155 | Contractor | Jun 2020 | B2 |
10776371 | Bross | Sep 2020 | B2 |
10809881 | Kindelsberger | Oct 2020 | B2 |
20020161736 | Beygelzimer | Oct 2002 | A1 |
20020184225 | Ghukasyan | Dec 2002 | A1 |
20030195885 | Emmick | Oct 2003 | A1 |
20040015480 | Federighi | Jan 2004 | A1 |
20050007383 | Potter | Jan 2005 | A1 |
20050160432 | Gallagher | Jul 2005 | A1 |
20070109302 | Tsuboshita | May 2007 | A1 |
20070266039 | Boykin | Nov 2007 | A1 |
20080005086 | Moore | Jan 2008 | A1 |
20080052378 | Matsuyama | Feb 2008 | A1 |
20080052379 | Matsuyama | Feb 2008 | A1 |
20100063785 | Pich | Mar 2010 | A1 |
20100161680 | Atre | Jun 2010 | A1 |
20110148880 | De Peuter | Jun 2011 | A1 |
20110249002 | Duplessis | Oct 2011 | A1 |
20110267350 | Cubera | Nov 2011 | A1 |
20110307685 | Song | Dec 2011 | A1 |
20120005631 | B'Far | Jan 2012 | A1 |
20120087377 | Lai | Apr 2012 | A1 |
20120209886 | Henderson | Aug 2012 | A1 |
20120254669 | Xia | Oct 2012 | A1 |
20120297145 | Castillo et al. | Nov 2012 | A1 |
20120317142 | Broecheler | Dec 2012 | A1 |
20130103719 | Gotz | Apr 2013 | A1 |
20130187941 | Noon | Jul 2013 | A1 |
20130212131 | Reddy | Aug 2013 | A1 |
20130212479 | Willis | Aug 2013 | A1 |
20130222388 | McDonald | Aug 2013 | A1 |
20130229416 | Krajec | Sep 2013 | A1 |
20130238667 | Carvalho | Sep 2013 | A1 |
20130250060 | Andersson | Sep 2013 | A1 |
20130282889 | Tito | Oct 2013 | A1 |
20130332387 | Mirra et al. | Dec 2013 | A1 |
20140050417 | Jiang | Feb 2014 | A1 |
20140098103 | Hirsch | Apr 2014 | A1 |
20140177008 | Raymond | Jun 2014 | A1 |
20140189665 | Hong et al. | Jul 2014 | A1 |
20140244623 | King | Aug 2014 | A1 |
20140244667 | Bararsani | Aug 2014 | A1 |
20140280143 | Milenova | Sep 2014 | A1 |
20140282384 | Pamer | Sep 2014 | A1 |
20140320498 | Ajitomi | Oct 2014 | A1 |
20140320500 | Fletcher | Oct 2014 | A1 |
20140354650 | Singh | Dec 2014 | A1 |
20140372458 | Jurca | Dec 2014 | A1 |
20140380282 | Ravindranath Sivalingam et al. | Dec 2014 | A1 |
20150026158 | Jin | Jan 2015 | A1 |
20150032777 | Dietrich | Jan 2015 | A1 |
20150074044 | Metrevell | Mar 2015 | A1 |
20150081741 | Xu | Mar 2015 | A1 |
20150089366 | Beckett | Mar 2015 | A1 |
20150099254 | Kamimaeda | Apr 2015 | A1 |
20150112750 | Grichnik | Apr 2015 | A1 |
20150120775 | Shao et al. | Apr 2015 | A1 |
20150138203 | Nachmanson | May 2015 | A1 |
20150138206 | Mizell | May 2015 | A1 |
20150227582 | Gu | Aug 2015 | A1 |
20150254370 | Sexton | Sep 2015 | A1 |
20150261886 | Wu | Sep 2015 | A1 |
20150310644 | Zhou | Oct 2015 | A1 |
20150324444 | Chercoles | Nov 2015 | A1 |
20150331932 | Georges | Nov 2015 | A1 |
20150339835 | Mohr | Nov 2015 | A1 |
20150347559 | Elias | Dec 2015 | A1 |
20150370919 | Bornhoevd | Dec 2015 | A1 |
20150378564 | Ramanathan | Dec 2015 | A1 |
20150379054 | Kernert | Dec 2015 | A1 |
20160019228 | Hong | Jan 2016 | A1 |
20160026939 | Schiffer | Jan 2016 | A1 |
20160048787 | Shinkuma | Jan 2016 | A1 |
20160035114 | Hesse | Feb 2016 | A1 |
20160063037 | Savkli | Mar 2016 | A1 |
20160063132 | Chen et al. | Mar 2016 | A1 |
20160071233 | Macko et al. | Mar 2016 | A1 |
20160078344 | Agarwal et al. | Mar 2016 | A1 |
20160103933 | Glover | Apr 2016 | A1 |
20160110476 | Shinkuma | Apr 2016 | A1 |
20160117358 | Schmid et al. | Apr 2016 | A1 |
20160125093 | Cho | May 2016 | A1 |
20160140152 | Sevenich et al. | May 2016 | A1 |
20160140166 | Schechter | May 2016 | A1 |
20160171068 | Hardin | Jun 2016 | A1 |
20160179887 | Lisonbee | Jun 2016 | A1 |
20160224532 | Miller | Aug 2016 | A1 |
20160259886 | Li | Sep 2016 | A1 |
20160292303 | Hong | Oct 2016 | A1 |
20160299991 | Hong et al. | Oct 2016 | A1 |
20160306922 | Van Rooyen | Oct 2016 | A1 |
20160321213 | Miyamoto | Nov 2016 | A1 |
20160342737 | Kaye | Nov 2016 | A1 |
20160359678 | Madani | Dec 2016 | A1 |
20170004638 | Csenteri | Jan 2017 | A1 |
20170032052 | Raman | Feb 2017 | A1 |
20170042495 | Matsuzaki | Feb 2017 | A1 |
20170060958 | Van Rest | Mar 2017 | A1 |
20170068746 | Levin | Mar 2017 | A1 |
20170085445 | Layman | Mar 2017 | A1 |
20170109907 | Hamedani | Apr 2017 | A1 |
20170116315 | Xiong | Apr 2017 | A1 |
20170140068 | Oh | May 2017 | A1 |
20170154445 | Maruhashi | Jun 2017 | A1 |
20170169133 | Kim | Jun 2017 | A1 |
20170177681 | Potiagalov | Jun 2017 | A1 |
20170177744 | Potiagalov | Jun 2017 | A1 |
20170185655 | Jahankhani | Jun 2017 | A1 |
20170195854 | Shi-Nash | Jul 2017 | A1 |
20170199905 | Ott | Jul 2017 | A1 |
20170200293 | Solem | Jul 2017 | A1 |
20170221240 | Stetson | Aug 2017 | A1 |
20170255945 | McCord | Sep 2017 | A1 |
20170277768 | Doan | Sep 2017 | A1 |
20170286122 | Wu et al. | Oct 2017 | A1 |
20170293697 | Youshi et al. | Oct 2017 | A1 |
20170308567 | Xiao | Oct 2017 | A1 |
20170323463 | Leiba | Nov 2017 | A1 |
20170330357 | Siegel | Nov 2017 | A1 |
20170351406 | Rossi | Dec 2017 | A1 |
20170357429 | Dixon | Dec 2017 | A1 |
20170371879 | Jang | Dec 2017 | A1 |
20180017746 | Stanfill | Jan 2018 | A1 |
20180018386 | Orlova | Jan 2018 | A1 |
20180025035 | Xia | Jan 2018 | A1 |
20180025040 | Goldberg | Jan 2018 | A1 |
20180025073 | Singh | Jan 2018 | A1 |
20180025093 | Xia | Jan 2018 | A1 |
20180025361 | Ferrarons | Jan 2018 | A1 |
20180067981 | Ahuja | Mar 2018 | A1 |
20180068406 | Van De Sande | Mar 2018 | A1 |
20180081937 | Broecheler | Mar 2018 | A1 |
20180089237 | Dunne | Mar 2018 | A1 |
20180121414 | Nanavati | May 2018 | A1 |
20180136830 | Kindelsberger | May 2018 | A1 |
20180137667 | Kinelsberger | May 2018 | A1 |
20180144424 | Sahu | May 2018 | A1 |
20180144534 | Lee | May 2018 | A1 |
20180157404 | Noel | Jun 2018 | A1 |
20180157761 | Halstead | Jun 2018 | A1 |
20180173372 | Greenspan | Jun 2018 | A1 |
20180174060 | Velez-Rojas | Jun 2018 | A1 |
20180181615 | Asher | Jun 2018 | A1 |
20180189416 | Lee | Jul 2018 | A1 |
20180198819 | Lee | Jul 2018 | A1 |
20180203918 | Chen | Jul 2018 | A1 |
20180219888 | Apostolopoulos | Aug 2018 | A1 |
20180239959 | Bui | Aug 2018 | A1 |
20180246986 | Haubenschild | Aug 2018 | A1 |
20180254101 | Gilmore | Sep 2018 | A1 |
20180285536 | Hanusiak | Oct 2018 | A1 |
20180293330 | Kovac | Oct 2018 | A1 |
20180307777 | Sevenich | Oct 2018 | A1 |
20180315228 | Boettcher | Nov 2018 | A1 |
20180329953 | Weld | Nov 2018 | A1 |
20180349002 | Kindelsberger | Dec 2018 | A1 |
20180357329 | Meyer | Dec 2018 | A1 |
20180357330 | Rodriguez | Dec 2018 | A1 |
20180367414 | Raghavendra | Dec 2018 | A1 |
20190005023 | Olsen | Jan 2019 | A1 |
20190065597 | Murray | Feb 2019 | A1 |
20190121810 | Zhuang | Apr 2019 | A1 |
20190155961 | Alonso | May 2019 | A1 |
20190179808 | Xu | Jun 2019 | A1 |
20190220695 | Nefedov | Jul 2019 | A1 |
20190278808 | Xia | Sep 2019 | A1 |
20190286504 | Muntes-Mulero | Sep 2019 | A1 |
20190286692 | Odate | Sep 2019 | A1 |
20190286757 | Muntes-Mulero | Sep 2019 | A1 |
20190325075 | Hilloulin | Oct 2019 | A1 |
20190332698 | Cho | Oct 2019 | A1 |
20190379700 | Canzanese, Jr. | Dec 2019 | A1 |
20190394103 | Korkin | Dec 2019 | A1 |
20200004746 | Randall | Jan 2020 | A1 |
20200035002 | Epasto | Jan 2020 | A1 |
20200051293 | Hou | Feb 2020 | A1 |
20200065316 | Mizuno | Feb 2020 | A1 |
20200082026 | Tian et al. | Mar 2020 | A1 |
20200117662 | Abe | Apr 2020 | A1 |
20200133663 | Dwars | Apr 2020 | A1 |
20200134359 | Kim | Apr 2020 | A1 |
20200151216 | Sevenich | May 2020 | A1 |
20200153934 | Burbank | May 2020 | A1 |
20200192880 | Hashemi | Jun 2020 | A1 |
20200192938 | Chang | Jun 2020 | A1 |
20200201875 | Wu | Jun 2020 | A1 |
20200210481 | Niebielski | Jul 2020 | A1 |
20200226124 | Chishti | Jul 2020 | A1 |
20200226156 | Borra | Jul 2020 | A1 |
20200226512 | Epstein | Jul 2020 | A1 |
20200233860 | Laino | Jul 2020 | A1 |
20200257730 | Srinivas | Aug 2020 | A1 |
20200265090 | Hilloulin | Aug 2020 | A1 |
20200301972 | Wang | Sep 2020 | A1 |
20200372075 | Rogynskyy | Nov 2020 | A1 |
20200387547 | Yoshitake | Dec 2020 | A1 |
20210132410 | Bhat et al. | May 2021 | A1 |
Number | Date | Country |
---|---|---|
WO1999045479 | Sep 1999 | WO |
Entry |
---|
Yonathan Perez et al., Ringo: Interactive Graph Analytics on Big-Memory Machines:, Proc ACM SIGMOD Int Conference Manag Data. Dated May-Jun. 2015, 21 pages. |
De Virgilio et al., “Converting Relational to Graph Databases”, Proceedings of the First International Workshop on Graph Data Management Experience and Systems (Grades 2013), Jun. 23, 2013, 6 pages. |
Anonymous, “Universally Unique Identifer”, Wikipedia, dated Feb. 13, 2018, www.web/2018020325/https://en.wikipedia.org/wiki/Universally_unique_identifer, 10 pages. |
Ahmed Elgohary et al., “Compressed Linear Algebra for Large-Scale Machine Learning”, VLDB Journal, No. 27, No. 5, dated Sep. 12, 2017, pp. 719-744. |
Oracle, “Using Property Graphs in an Oracle Database Environment”, docs.oracle.com/database/122/SPGDG/using-property-graphs-oracle database.htm#BDSPA191, Apr. 23, 2018, 152 pages. |
Linkurious, A Web-Based Graph Data Visualization Tool for Investigative Journalism, http://www.influencemapping.org/blog/2015/09/30/linkurious.html, 3 pages. Dated Sep. 2015, 3 pages. |
Ellson et al., “Graphviz and Dynagraph-Static and Dynamic Graph Drawing Tools”, AT&T Labs—Research, dated 2004, 23 pages. |
Farber et al., “The SAP HANA Database-An Architecture Overview” IEEE, dated 2012, 6 pages. |
Freeman, Eve, “Getting Sgtarted with Neo4j and Cypher”, https://www.airpair.com/neo4j/posts/getting-started-with-neo4j-and-cypherdated Aug. 2016, 12 pages. |
Ganser et al., “Topological Fisheye Views for Visualizing Large Graphs”, dated Jul. 2005, 8 pages. |
Gephi Tutorial Visualization, dated May 26, 2010, 24 pages. |
Graph Data, “Turi Machine Learning Platform User Guide”, Working with Graphs, https://turi.com/learn/userguide/sgraph/sgraph.html, dated Jun. 2016, 1 page. |
Ediger et al., “STINGER: High Performance Data Structure for Streaming Graphs”, IEEE, dated 2012, 5 pages. |
Kofia, et al., “Social Network: a Cytoscape app for visualizing co-publication networks [version 1; referees: 2 approved with reservations]”, dated Aug. 5, 2015, 14 pages. |
Xirogiannopoulos et al., “Extracting and Analyzing Hidden Graphs from Relational Databases”, ACM, SIGMOD dated 2017, 18 pages. |
Macko et al., “Llama: Efficient Graph Analytics Using Large Multiversioned Arrays.” In Proceedings of the 31st IEEE International Conference on Data Engineering, dated Apr. 13-17, 2015, 13 pages. |
Neo4j Data Import: Moving Data from RDBMS to Graph, Tutorial, https://neo4j.com/developer/guide-importing-data-and-etl/, dated Aug. 2016, 9 pages. |
Nick Quinn, “Making sense of the graph revolution”, www.JAXenter.com, dated Sep. 2013, 5 pages. |
Oracle, “Using Property Graphs in an Oracle Database Environment”, docs.oracle.com/database/122/SPGDG/using-property-graphs-oracle-database.htm#BDSPA191, Apr. 23, 2018, Apr. 23, 2018, 152 pages. |
SQL Server Blog, “Graph Data Processing with SQL Server 2017 and Azure SQL Database”, dated Aug. 20, 2017, 8 pages. |
TensorFlow, “TensorBoard: Graph Visualization”, dated Nov. 2015, https://www.tensorflow.org/versions/r0.11/how_tos/graph_viz/, 10 pages. |
Kaur, Sawinder, “Visualizing Class Diagram Using OrientDB NoSQL Data-Store”, dated Jul. 2016, 5pages. |
Zhou et al., “Implementing Database Operations Using SIMD Instructions”, ACM SIGMOD Jun. 4-6, '2002, Madison, Wisconsin, USA, 12 pages. |
Wheatman et al., “Packed Compressed Sparse Row: A Dynamic Graph Representation”, dated 2018, 7 pages. |
Oracel Help Center, “ROWID Pseudocolum”, dated Nov. 20, 2019, 2 pages. |
Oracel Help Center, “Pseudoclumns”, dated Nov. 20, 2019, 1 page. |
IMB Knowledge Center, “Expand Your IBM Z Capabilities”, dated Nov. 20, 2019, 3 pages. |
Hong et al., “PGX.D: A Fast Distributed Graph Processing Engine”, SC, dated Nov. 2015, Austin TX, USA, 12 pages. |
Hassan, “Extending In-Memory Relational Database Engines with Native Graph Support”, dagted Mar. 2018, 12 pages. |
Xirogiannopoulos et al., “GraphGen: Exploring Interesting Graphs in Relational Data”, Proceedings of the VLDB Endowment, vol. 8, No. 12 Copyright 2015 VLDB Endowment, 4 pages. |
Mathew et al., “An Efficient Index based Query handling model for Neo4j”, International Journal of Advances in Computer Science and Technology (IJACST), vol. 3 No. 2, dated 2014, 7 pages. |
Dave et al., “GraphFrames: An Integrated API for Mixing Graph and Relational Queries”, ACM, dated 2016, 8 pages. |
Merril et al., Scalable GPU Graph Traveral, dated 2012, 11 pages. |
Merril et al., High Performance and Scalable GPU Graph Traversal, dated 2015, 30 pages. |
Djdates, Access 2007 Tutorial, Dated Nov. 15, 2009, You Tube.com, link: https://wwww.youtube.com/watch?v=nszRT3nRUMU, 8 pages. |
Valiyev, Mahammad, “Graph Storage: How Good is CSR Really?”, dated Dec. 10, 2017, 8 pages. |
Kesheng Wu et al., “A Performance Comparison of Bitmap Indexes”, proceedings of the 2001 ACM CIKM 10th International Conference on Information and Knowledge Management, dated Oct. 5, 2001, 2 pages. |
Kindelsberger, U.S. Appl. No. 15/603,218, filed May 23, 2017, Office Action, dated Nov. 22, 2019. |
Kindelsberger, U.S. Appl. No. 15/603,218, filed May 23, 2017, Notice of Allowance, dated Apr. 10, 2020. |
Kindelsberger, U.S. Appl. No. 15/469,674, filed Mar. 27, 2017, Interview Summary, dated Mar. 20, 2019. |
Kindelsberger, U.S. Appl. No. 15/469,674, filed Mar. 27, 2017, Advisory Action, dated Apr. 18, 2019. |
Yousfi, U.S. Appl. No. 15/096,034, filed Apr. 11, 2016, Final Office Action, dated Sep. 20, 2018. |
Kindelsberger, U.S. Appl. No. 15/469,674, filed Mar. 27, 2017, Office Action, dated Jun. 13, 2019. |
Kindelsberger, U.S. Appl. No. 15/469,674, filed Mar. 27, 2017, Office Action, dated Jul. 13, 2018. |
Kindelsberger, U.S. Appl. No. 15/603,218, filed May 23, 2017, Final Office Action, dated Feb. 21, 2020. |
Yousfi, U.S. Appl. No. 15/096,034, filed Apr. 11, 2016, Interview Summary, dated Nov. 23, 2018. |
Yousfi, U.S. Appl. No. 15/096,034, filed Apr. 11, 2016, Advisory Action, dated Dec. 5, 2018. |
Kindelsberger, U.S. Appl. No. 15/603,218, filed May 23, 2017, Office Action, dated Dec. 14, 2018. |
Haubenschild, U.S. Appl. No. 15/443,327, filed Feb. 27, 2017, Notice of Allowance, dated Nov. 2, 2018. |
Kindelsberger, U.S. Appl. No. 15,469,674, filed Mar. 27, 2017, Final Office Action, dated Sep. 30, 2020. |
Kindelsberger, U.S. Appl. No. 15/469,674, filed Mar. 27, 2017, Office Action, dated Jun. 25, 2020. |
Merrill et al. Scalable GPU Graph Traversal, 2012, 11 pages (Year: 2012). |
Padmanabhan: “HDB-Subdue, A Relational Database Approach To Graph Mining and Hierarchical Reduction”, Dec. 2005. |
Number | Date | Country | |
---|---|---|---|
20190325075 A1 | Oct 2019 | US |