Abstract: A highly programmable data processing unit includes multiple processing units for processing streams of information, such as network packets or storage packets. The data processing unit includes one or more specialized hardware accelerators configured to perform acceleration for various data-processing functions. The data processing unit is configured to retrieve speculative probability values for range coding a plurality of bits with a single read instruction to an on-chip memory that stores a table of probability values. The data processing unit is configured to store state information used for context-coding packets of a data stream so that the state information is available after switching between data streams.

Abstract: An engine architecture for processing finite automata includes a hyper non-deterministic automata (HNA) processor specialized for non-deterministic finite automata (NFA) processing. The HNA processor includes a plurality of super-clusters and an HNA scheduler. Each super-cluster includes a plurality of clusters. Each cluster of the plurality of clusters includes a plurality of HNA processing units (HPUs). A corresponding plurality of HPUs of a corresponding plurality of clusters of at least one selected super-cluster is available as a resource pool of HPUs to the HNA scheduler for assignment of at least one HNA instruction to enable acceleration of a match of at least one regular expression pattern in an input stream received from a network.

Abstract: A hardware-based programmable text analytics processor has a plurality of components including at least a tokenizer, a tagger, a parser, and a classifier. The tokenizer processes an input stream of unstructured text data and identifies a sequence of tokens along with their associated token ids. The tagger assigns a tag to each of the sequence of tokens from the tokenizer using a trained machine learning model. The parser parses the tagged tokens from the tagger and creates a parse tree for the tagged tokens via a plurality of shift, reduce and/or finalize transitions based on a trained machine learning model. The classifier performs classification for tagging and parsing by accepting features extracted by the tagger and the parser, classifying the features and returning classes of the features back to the tagger and the parser, respectively. The TAP then outputs structured data to be processed for various text analytics processing applications.

Abstract: In an embodiment, a method of updating a memory with a plurality of memory lines, the memory storing a tree, a plurality of buckets, and a plurality of rules, can include maintaining a copy of the memory with a plurality of memory lines. The method can further include writing a plurality of changes to at least one of the tree, the plurality of buckets, and the plurality of rules to the copy. The method can additionally include determining whether each of the plurality of changes is an independent write or a dependent write. The method can further include merging independent writes to the same line of the copy. The method further includes transferring updates from the plurality of lines of the copy to the plurality of lines of the memory.

Abstract: A system, apparatus, and method are provided for receiving one or more incremental updates including adding, deleting, or modifying rules of a Rule Compiled Data Structure (RCDS) used for packet classification. Embodiments disclosed herein may employ at least one heuristic for maintaining quality of the RCDS. At a given one of the one or more incremental updates received, a section of the RCDS may be identified and recompilation of the identified section may be triggered, altering the RCDS shape or depth in a manner detected by the at least one heuristic employed. The at least one heuristic employed enables performance and functionality of an active search process using the RCDS to be improved by advantageously determining when and where to recompile one or more sections of the RCDS being searched.

Abstract: At least one processor may be operatively coupled to a plurality of memories and a node cache and configured to walk nodes of a per-pattern non-deterministic finite automaton (NFA). Nodes of the per-pattern NFA may be stored amongst one or more of the plurality of memories based on a node distribution determined as a function of hierarchical levels mapped to the plurality of memories and per-pattern NFA storage allocation settings configured for the hierarchical levels, optimizing run time performance of the walk.

Abstract: A system, apparatus, and method are provided for adding, deleting, and modifying rules in one update from the perspective of an active search process for packet classification. While a search processor searches for one or more rules that match keys generated from received packets, there is a need to add, delete, or modify rules. By organizing a plurality incremental updates for adding, deleting, or modifying rules into a batch update, several operations for incorporating the incremental updates may be made more efficient by minimizing a number of updates required.

Abstract: At least one per-pattern non-deterministic finite automaton (NFA) may be generated for a single regular expression pattern and may include a respective set of nodes. Nodes of the respective set of nodes of each per-pattern NFA generated may be distributed for storing in a plurality of memories based on hierarchical levels mapped to the plurality of memories and per-pattern NFA storage allocation settings configured for the hierarchical levels, optimizing run time performance for matching regular expression patterns in an input stream.

Abstract: A method, and corresponding apparatus and system are provided for optimizing matching of at least one regular expression pattern in an input stream by storing a context for walking a given node, of a plurality of nodes of a given finite automaton of at least one finite automaton, the store including a store determination, based on context state information associated with a first memory, for accessing the first memory and not a second memory or the first memory and the second memory. Further, to retrieve a pending context, the retrieval may include a retrieve determination, based on the context state information associated with the first memory, for accessing the first memory and not the second memory or the second memory and not the first memory. The first memory may have read and write access times that are faster relative to the second memory.

Abstract: An engine architecture for processing finite automata includes a hyper non-deterministic automata (HNA) processor specialized for non-deterministic finite automata (NFA) processing. The HNA processor includes a plurality of super-clusters and an HNA scheduler. Each super-cluster includes a plurality of clusters. Each cluster of the plurality of clusters includes a plurality of HNA processing units (HPUs). A corresponding plurality of HPUs of a corresponding plurality of clusters of at least one selected super-cluster is available as a resource pool of HPUs to the HNA scheduler for assignment of at least one HNA instruction to enable acceleration of a match of at least one regular expression pattern in an input stream received from a network.

Abstract: A method and corresponding apparatus relate to converting a nondeterministic finite automata (NFA) graph for a given set of patterns to a deterministic finite automata (DFA) graph having a number of states. Each of the DFA states is mapped to one or more states of the NFA graph. A hash value of the one or more states of the NFA graph mapped to each DFA state is computed. A DFA states table correlates each of the number of DFA states to the hash value of the one or more states of the NFA graph for the given pattern.

Abstract: Matching at least one regular expression pattern in an input stream may be optimized by initializing a search context in a run stack based on (i) partial match results determined from walking segments of a payload of a flow through a first finite automation and (ii) a historical search context associated with the flow. The search context may be modified via push or pop operations to direct at least one processor to walk segments of the payload through the at least one second finite automation. The search context may be maintained in a manner that obviates overflow of the search context and obviating stalling of the push or pop operations to increase match performance.

Abstract: A hardware-based programmable text analytics processor has a plurality of components including at least a tokenizer, a tagger, a parser, and a classifier. The tokenizer processes an input stream of unstructured text data and identifies a sequence of tokens along with their associated token ids. The tagger assigns a tag to each of the sequence of tokens from the tokenizer using a trained machine learning model. The parser parses the tagged tokens from the tagger and creates a parse tree for the tagged tokens via a plurality of shift, reduce and/or finalize transitions based on a trained machine learning model. The classifier performs classification for tagging and parsing by accepting features extracted by the tagger and the parser, classifying the features and returning classes of the features back to the tagger and the parser, respectively. The TAP then outputs structured data to be processed for various text analytics processing applications.

Abstract: A hardware-based programmable deep learning processor (DLP) is proposed, wherein the DLP comprises with a plurality of accelerators dedicated for deep learning processing. Specifically, the DLP includes a plurality of tensor engines configured to perform operations for pattern recognition and classification based on a neural network. Each tensor engine includes one or more matrix multiplier (MatrixMul) engines each configured to perform a plurality of dense and/or sparse vector-matrix and matrix-matrix multiplication operations, one or more convolutional network (ConvNet) engines each configured to perform a plurality of efficient convolution operations on sparse or dense matrices, one or more vector floating point units (VectorFPUs) each configured to perform floating point vector operations, and a data engine configured to retrieve and store multi-dimensional data to both on-chip and external memories.

Abstract: An engine architecture for processing finite automata includes a hyper non-deterministic automata (HNA) processor specialized for non-deterministic finite automata (NFA) processing. The HNA processor includes a plurality of super-clusters and an HNA scheduler. Each super-cluster includes a plurality of clusters. Each cluster of the plurality of clusters includes a plurality of HNA processing units (HPUs). A corresponding plurality of HPUs of a corresponding plurality of clusters of at least one selected super-cluster is available as a resource pool of HPUs to the HNA scheduler for assignment of at least one HNA instruction to enable acceleration of a match of at least one regular expression pattern in an input stream received from a network.

Abstract: In a processor of a security appliance, an input of a sequence of characters is walked through a finite automata graph generated for at least one given pattern. At a marked node of the finite automata graph, if a specific type of the at least one given pattern is matched at the marked node, the input sequence of characters is processed through a reverse non-deterministic finite automata (rNFA) graph generated for the specific type of the at least one given pattern by walking the input sequence of characters backwards through the rNFA beginning from an offset of the input sequence of characters associated with the marked node. Generating the rNFA for a given pattern includes inserting processing nodes for processing an input sequence of patterns to determine a match for the given pattern. In addition, the rNFA is generated from the given type of pattern.

Abstract: A method, and corresponding apparatus and system are provided for optimizing matching at least one regular expression pattern in an input stream by walking at least one finite automaton in a speculative manner. The speculative manner may include iteratively walking at least two nodes of a given finite automaton, of the at least one finite automaton, in parallel, with a segment, at a current offset within a payload, of a packet in the input stream, based on positively matching the segment at a given node of the at least two nodes walked in parallel, the current offset being updated to a next offset per iteration.

Abstract: In one embodiment, a method includes encoding a key matching rule having at least one dimension by storing in a memory (i) a header of the key matching rule that has at least one header field, and (ii) at least one rule value field of the key matching rule corresponding to one of the dimensions.

Abstract: In an embodiment, a method of compiling a pattern into a non-deterministic finite automata (NFA) graph includes examining the pattern for a plurality of elements and a plurality of node types. Each node type can correspond with an element. Each element of the pattern can be matched at least zero times. The method further includes generating a plurality of nodes of the NFA graph. Each of the plurality of nodes can be configured to match for one of the plurality of elements. The node can indicate the next node address in the NFA graph, a count value, and/or node type corresponding to the element. The node can also indicate the element representing a character, character class or string. The character can also be a value or a letter.

Abstract: A multi-rule approach for encoding rules grouped in a rule chunk is provided. The approach includes a multi-rule with a multi-rule header representing headers of the rules and, in some cases, dimensional data representing dimensional data of the rules. The approach further includes disabling dimension matching of always matching dimensions, responding to an always match rule with a match response without matching, interleaving minimum/maximum values in a range field, interleaving value/mask values in a mask field, and for a given rule of rule chunk, encoding a priority field at the end of dimension data stored for the rule in the multi-rule. Advantageously, this approach provides efficient storage of rules and enables the efficient comparison of rules to keys.