Patent Application
20230130239
Decision trees are one approach to machine learning, with decision tree adoption being widespread across many disciplines. Inference performance of a decision tree in a production environment can be a significant issue. Conventionally, decision tree inference processing is inherently sequential. Decision node comparisons are tested in hierarchical sequence as prescribed by the decision tree structure. Decision nodes branch to other decision nodes depending on the results of the comparisons, where the logic fetches the next node of the tree to process. These branches are skip sequential memory accesses, which can cause a pipeline slowdown in the central processing unit due to memory access delays. This memory access pattern is endemic to the decision tree structure. To assist performance, a decision tree inference workload can be offloaded to a peripheral computational device to reduce resource consumption on, for instance, one or more central processing units (CPUs).
Certain shortcomings of the prior art are overcome and additional advantages are provided through the provision, in one or more aspects, of a computer system for facilitating processing within a computing environment. The computer system includes a memory, and a processing circuit in communication with the memory, and configured to perform a method, which includes establishing, by the processing circuit, an inflated decision tree from a source decision tree. The establishing includes inserting one or more phantom decision nodes into the source decision tree to obtain the inflated decision tree. Further, the method includes ascertaining, by the processing circuit, decision node data and leaf node data for the inflated decision tree, and providing, by the processing circuit, the decision node data and the leaf node data to an inference accelerator to facilitate accelerated processing of the inflated decision tree, and determining therefrom which leaf node of a plurality of leaf nodes of the inflated decision tree is selected.
Advantageously, inflating the decision tree enables the decision tree to be processed more efficiently by the inference accelerator. With the inflated decision tree, or fully-populated decision tree, each leaf node has a fixed and known path through the decision nodes. That knowledge is leveraged by the inference accelerator to reduce the number of parameters that need to be transmitted to the inference accelerator, hence improving overall performance by reducing the input/output to the inference accelerator, which reduces inference processing turnaround time, as well as freeing logic real estate for other work.
In one example, the source decision tree is a pruned decision tree, and the decision node data includes adjusted position decision node data for the inflated decision tree, and the leaf node data includes adjusted position leaf node data for the inflated decision tree. Advantageously, the generalization benefits of a pruned source decision tree are maintained, while inference performance is improved using the inflated, or fully-populated, decision tree for accelerated processing by the inference accelerator.
In one or more embodiments, the accelerated processing includes processing accelerated, based on respective predetermined path vectors through the inflated decision tree for each leaf node of the plurality of leaf nodes, by processing a decision node result vector by the plurality of leaf nodes in parallel. Advantageously, the inference accelerator is configured to facilitate full-parallel processing by the plurality of leaf nodes of the decision node result vector based on the respective, predetermined path vector through the decision tree to each leaf node of the plurality of leaf nodes. In this manner, latency of decision tree inference processing is reduced, enhancing processing throughput of the decision tree and/or decision tree ensemble.
In one embodiment, the ascertaining includes, for a leaf node of the plurality of leaf nodes, descending the leaf node to a maximum depth of the inflated decision tree and based thereon, adjusting a path vector for the leaf node to obtain the respective, predetermined path vector for the leaf node. Descending each leaf node to a maximum depth of the inflated decision tree removes the requirement that the path to each leaf node be transmitted to the inference accelerator, thereby reducing input/output and configuration processing for the inference accelerator. Descending the leaf node the maximum depth of the inflated decision tree results in a fully-populated inflated decision tree, meaning that one or more masks derived from the respective, predetermined path vectors to facilitate inference processing in the inference accelerator do not need to be transmitted to the accelerator.
In one example, the adjusting includes determining a string value to insert into the path vector based on a distance that the leaf node descends pursuant to the descending, and inserting the string value into the path vector to obtain the respective, predetermined path vector for the leaf node. In one embodiment, the string value to insert is equivalent to a last entry in a fully-populated path table for the descendent-level size.
In one example, the adjusting includes determining an insertion position for the string value by identifying a last decision node in the path vector, and based on the last decision node being true, or the last decision node being at the end of the path, the insertion position is one position beyond the decision node. Based on the last decision node being false and other than the end of the path, the insertion position is two positions beyond the last decision node.
In one implementation, for at least one other leaf node of the plurality of leaf nodes, a ‘don't care’ string is inserted at the same insertion position into its path vector, where the ‘don't care’ string is same-sized as the determined string value.
Computer-implemented methods and computer program products relating to one or more aspects are also described and claimed herein. Further, services relating to one or more aspects are also described and can be claimed herein.
Additional features and advantages are realized through the techniques described herein. Other embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed aspects.
One or more aspects of the present invention are particularly pointed out and distinctly claimed as examples in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the invention are apparent from the following detailed description taken in conjunction with the accompanying drawings in which:
The accompanying figures, in which like reference numerals refer to identical or functionally similar elements throughout the separate views, and which are incorporated in and form a part of this specification, further illustrate the present invention and, together with the detailed description of the invention, serve to explain aspects of the present invention. Note in this regard that descriptions of well-known systems, devices, accelerators, processing techniques, etc., are omitted so as to not unnecessarily obscure the invention in detail. It should be understood, however, that the detailed description and this specific example(s), while indicating aspects of the invention, are given by way of illustration only, and not limitation. Various substitutions, modifications, additions, and/or other arrangements, within the spirit or scope of the underlying inventive concepts will be apparent to those skilled in the art from this disclosure. Note further, that numerous inventive aspects and features are disclosed herein, and unless inconsistent, each disclosed aspect or feature is combinable with any other disclosed aspect or feature as desired for a particular application of the concepts disclosed herein.
Note also that illustrative embodiments are described below using specific code, designs, architectures, accelerators, protocols, layouts, schematics or tools, only as examples, and not by way of limitation. Further, the illustrative embodiments are described in certain instances using particular hardware, software, tools, or data processing environments only as example for clarity of description. The illustrative embodiments can be used in conjunction with other comparable or similarly purposed structures, systems, applications, or architectures. One or more aspects of an illustrative embodiment can be implemented in hardware, software, or a combination thereof.
As understood by one skilled in the art, program code, as referred to in this application, can include both hardware and software. For example, program code in certain embodiments of the present invention can include fixed function hardware, but other embodiments can utilize a software-based implementation of the functionality described. Certain embodiments combine both types of program code. One example of program code, also referred to as one or more programs or program instructions, is depicted in
As noted, decision tree inference processing is conventionally inherently sequential, with decision node comparisons typically being tested in hierarchical sequence, as prescribed by the decision tree structure. Decision nodes branch to other decision nodes depending on the results of the comparisons, where the logic fetches the next node to process. These branches are skip sequential memory accesses, which cause a pipeline slowdown in the CPU due to memory access delays. This memory access pattern is endemic to the decision tree structure.
In contrast, decision tree inference processing, via an inference accelerator such as disclosed herein, provides full parallelization with processing time measured in clock cycles, not milliseconds, as with other machine learning techniques. In one embodiment, section-wide parallelism of decision tree ensembles is disclosed with, for instance, all decision nodes being processed in parallel at once, and all leaf nodes being processed in parallel at once, as described herein. A parallel adder, or adder tree, can be used to summarize the parallel results. Further, as discussed herein, library support can be provided with the interface accelerator, and if demand exceeds accelerator supply of any internal resource, then the processing can be performed with multiple iterations. Further, in one embodiment, decision trees can be inflated to conform to a fully populated form, as needed, depending on the implementation. Decision trees can be translated into a format required by the inference accelerator, and the decision tree or decision tree ensemble can be separated into different processing sections, as needed. For instance, as noted, all decision nodes can be processed in parallel by the accelerator, followed by all leaf nodes in parallel.
In one embodiment, a decision tree ensemble can include multiple independent decision trees, with each tree being a weak learner. Decision tree results can be weighted, with, for instance, a signal or value from an accurate tree being amplified, and a signal or value from an inaccurate tree being attenuated. By way of example, a random forest is a machine learning model based on decision tree ensembles.
As illustrated, the inference accelerator is trained using training or testing data 101 to produce, in one embodiment, a trained machine learning model 111. The decision tree or decision tree ensemble training can develop the decision tree using, for instance, existing software processes, such as SKLearn (i.e., a machine learning library for the Python™ programming language) and be saved in the software-determined format. The trained machine learning model 111 is then compiled 120, which in one embodiment, is performed by a special-purpose compiler that converts or formats the decision tree or decision tree ensemble into a form usable by inference accelerator 140. For instance, in one implementation, a compilation process converts the SKLearn decision tree (or decision tree ensemble) into a format required by the inference accelerator. In addition, the compilation process can include, for instance, inflating pruned trees (as described herein) to fully-populated trees at a specified depth. Lists of decision nodes and leaf nodes are generated, in one or more embodiments, for transmission to the inference accelerator. The machine learning model 121 prepared for processing by inference accelerator 140 can, in one embodiment, be saved to an inference library 130 for invoking by a user program 131 to drive inference accelerator 140, sending, for instance, an input data vector and the trained decision tree(s) to the inference accelerator. In one embodiment, the library interface responds to a caller's request, performs data conversion for provision to inference accelerator 140, transmits the decision nodes and input data vector to the inference accelerator, and waits for the output results.
The inference accelerator can be housed, in one or more embodiments, external from a central processing unit (CPU), and can be, in one or more embodiments, a multi-core inference engine, configured to read decision nodes, obtain data from the caller's memory, and perform processes on the decision tree (or decision tree ensemble) including, in one embodiment, processing the decision nodes in parallel or concurrently, and separately processing the leaf nodes in parallel, for instance, using a bit-level parallelism. Further, in one embodiment, an adder or adder tree can be provided to accumulate output values in parallel, with the output of the inference accelerator being returned to the caller (e.g., user program). In another embodiment, the encoded result of the leaf node selection process, representing an index of the selected leaf node, can be returned to the calling process for further processing. This would remove the need for an adder in the inference accelerator and would serve to reduce the amount of information needed to be transferred to the inference accelerator. Thus, the embodiments depicted and described herein with reference to the inclusion of one or more adders or adder trees are one example only of an inference accelerator with enhanced decision tree leaf node processing such as described.
Referring to
By way of further example,
In one or more implementations, a decision tree includes a hierarchy of decision nodes, with a leaf node residing at the terminus of each path through the decision tree. Each decision node can have a comparison of a variable against a constant value, which results in a Boolean value, either true or false. The decision node includes two edges that represent the paths to the next nodes in the tree, which can either be another decision node or a leaf node. A leaf node is reached depending on the results of the decision nodes test, and once reached, is said to be selected. Each leaf node has one or more values associated with it that, once selected, are output as the result of the decision tree.
Disclosed herein is an inference accelerator that uses the path through the decision tree for each leaf node of the plurality of leaf nodes of the decision tree to infer whether the leaf node is selected. In one embodiment, all decision node comparisons of the decision tree are first determined to obtain by the inference accelerator a decision node result vector. In one example, the decision nodes can be processed in parallel, though that is not required.
In one or more embodiments, the inference accelerator utilizes two masks for each leaf node derived from the leaf node's path. These masks are used to determine whether the leaf node is the one leaf node selected among the set of leaf nodes of the decision tree. Only one leaf node is selected for the decision tree for any given vector of input data using inference processing. In this manner, the entire set of leaf nodes of the decision tree can be processed in parallel by the inference accelerator. Advantageously, this approach reduces latency of the decision tree inference and provides a rapid turnaround of results of the decision tree, or decision tree ensemble, which leads to higher processing throughput.
As illustrated in the decision tree example of
The depth of a leaf node 301 is the distance from the root decision node to the leaf node.
The decision nodes 300 are ordered in a depth-first sequence, which is a common technique for ordering decision nodes of a decision tree. The leaf nodes are also ordered in the depth-first manner, though they can also be seen to be left-to-right ordered.
A fully-populated decision tree has all of its leaf nodes 301 at the same distance from the root decision node. The inference accelerator disclosed herein does not require that the decision tree be in fully-populated form, and only for illustrative purposes is
The leaf node distance is the number of edges between the root decision node and the leaf node, and is referred to as the depth. For a given depth of a fully-populated tree, there are two raised to the power of the depth leaf nodes, and one less decision node. For example, a fully-populated tree of depth=3, will have 23, or 8 leaf nodes, and 7 decision nodes.
As noted, the decision tree has been trained, and is used in the inference accelerator, such as disclosed. In one embodiment, the trained decision tree is passed through a compilation process, which extracts and/or formats the information needed by the inference accelerator.
In one or more implementations, the path of each leaf node in the decision tree is identified through the decision nodes, and two masks are derived for each leaf node from the respective path.
The two masks (referred to herein as masks M1, M2) can be derived from the path vectors, such as the vectors illustrated in
When processing a decision tree, each decision node sets a Boolean indicator of whether the comparison result is true or false. The aggregate of the decision node indicators of the decision tree are a decision node result vector (RV) in
In one solution to decision tree processing, it is possible to use the full set of combinations of decision node outcomes to reach each leaf node, and store this in an internal table. However, the number of combinations explodes because there are two raised to the number of decision node power possible combinations. For a depth-3 decision tree, there are seven decision nodes, and hence 27, or 128 combinations. For a 4-depth tree, 215, or 32K combinations, for a depth-5 tree, 231, or 2G combinations. In comparison, using the predetermined path vector technique described herein, for a depth-3 decision tree, there are 8 leaf nodes, with each having 3 decision nodes in its path. For a depth-4 tree, there are 16 leaf nodes, with each having 4 decision nodes in its path, and for a depth-5 tree, there are 32 leaf nodes, with each having 5 decision nodes in its path.
In one or more embodiments of the parallel inference processing disclosed herein, each leaf node processing is provided its own copy of the decision node result vector from the set of decision node processes. In one implementation, the one mask M1 is applied to the decision node result vector with a bit-wise exclusive OR (XOR), and the other mask M2 is applied to the updated result vector with a bit-wise logical AND (AND). The result of this operation is then compared to the other mask M2 using a not-exclusive-OR (XNOR). A reduction AND can then be applied to all the bits of the result vector to obtain a single result. This single result is (in one embodiment) a single indicator that signals whether this leaf node is selected or not. If all the bits in the leaf node inference result vector are a binary 1, then the combination will be true, and the leaf node is selected, otherwise, the result is false, and the leaf node is not selected. For any given input data vector, only one leaf node is selected for the decision tree.
As illustrated in
As described, each leaf node inference processing is independent of the other leaf nodes, so that the leaf nodes can readily process in parallel the decision node result vector. More particularly, each bit of the result vector that is input to a leaf node, along with the place matching bit in the two masks, are independent of all other bits processed for this leaf node.
Only one leaf node will activate for the decision tree for the given data input vector, and the set of activation bits is a one-hot vector (all “0”, except for the activated leaf node, which is “1”). The set of activation bits can be passed into an encoder that converts the result into a numeric index equivalent to the leaf node's position in the tree, as illustrated in
Generally stated, disclosed is a process to alter the structure of a source decision tree to improve overall performance of decision tree inference processing by the inference accelerator. As described above, in one or more implementations, the inference accelerator is a special-purpose device acting as an adjunct to a general purpose processor, such as a central processing unit (CPU). Advantageously, in one or more embodiments, the structure of the source decision tree is altered to conform to the expectations of the inference accelerator. This advantageously reduces the input/output across the accelerator boundary, further accelerating overall inference process. By restricting processing to fully-populated or inflated decision trees, the masks (M1, M2) described above do not need to be transmitted to the inference accelerator. Note that an inflated decision tree or fully-populated tree refers herein to a decision tree, that for a given depth, has all decision nodes occupied. More particularly, in one or more embodiments, an inflated or fully-populated decision tree refers to a pruned decision tree that has been inflated to full-form. For instance, bypassed decision nodes are inserted as phantom decision nodes, and are “not valid” in the resultant decision tree definition object. Missing leaf nodes are inserted as “not present” in the leaf node portion of the decision tree definition. Once the decision tree conforms to a fully-populated form, leaf node paths take default values for a given depth of the decision tree. Thus, the masks used by the inference accelerator are fixed, and can be embedded in the accelerator. Advantageously, with a fixed and known tree structure, the inference process does not require that the tree structure be included in the inference transmission to the peripheral inference accelerator. In addition, not having to respond to a variable structure improves overall decision tree inference processing by reducing turnaround time. Also, avoiding the extra processing steps that would otherwise be required, frees logic real estate for other work.
Trained decision trees are typically not fully-populated to the constructed maximum depth. During training, splits identified in the data and embedded in the decision nodes can lead to many different forms of decision tree structures. Also, many decision tree training algorithms use pruning, where decision nodes are removed from the trained tree to reduce over-fitting and improve generalization. This can be beneficial to the results of an inference process of the decision tree. This process leads to the expectation that most decision trees (i.e., source decision trees) to be process will not conform to the corresponding, fully-populated decision tree structure (e.g., as shown in
Advantageously, disclosed herein are computer systems, computer-implemented methods and computer program products with program code which when executed performs a method that includes establishing, by a processing circuit, an inflated decision tree from a source decision tree, where the establishing includes inserting one or more phantom decision nodes into the source decision tree to obtain the inflated decision tree. Processing ascertains decision node data and leaf node data for the inflated decision tree, and provides the decision node data and the leaf node data to an inference accelerator to facilitate accelerated processing of the inflated decision tree, and determining therefrom which leaf node of a plurality of leaf nodes of the inflated decision tree is selected, such as described herein. Advantageously, the inflating of the source decision tree is so that the tree takes the form of a fully-populated tree at the selected depth. This enables the pruned or otherwise-shaped source decision tree to be processed more efficiently by the inference accelerator. In this manner, inflating of the decision tree is part of the overall process to accelerate decision tree inference processing, and can be (in one embodiment) embedded or associated with, for instance, a compilation process, such as compilation process 120 of
In accordance with one or more aspects, program code is provided to inflate the decision tree (i.e., source decision tree) so that it matches the corresponding, fully-populated decision tree format expected by the inference accelerator. As part of the inflating process, one or more phantom decision nodes are inserted into the source decision tree structure to represent missing decision nodes. The phantom decision nodes only produce a false comparison result. In this manner, leaf nodes are shifted into appropriate positions based on the decision node insertions. The restructuring of the decision tree advantageously enables, for instance, reuse of leaf node path knowledge in the associated inference accelerator, which advantageously reduces traffic between the CPU and inference accelerator.
Once the decision tree has been inflated to a fully-populated decision tree for the given depth, there is a decision node at each decision position of the tree. There are vacant leaf node positions when leaf nodes shift to accommodate the restructuring. The now missing leaf node positions are unreachable and are therefore ignored by the parallel inference processing described herein. In the fully-populated or inflated decision tree, each leaf node has a fixed and known path through the decision nodes. That knowledge is leveraged by the inference accelerator to reduce the number of parameters that need to be transmitted to the inference accelerator, hence improving overall performance by reducing the input/output to the peripheral inference accelerator, and freeing logic real estate for other work. In this manner, the generalization benefit of a pruned decision tree is maintained, while also gaining performance improvement of using the fixed structure of a fully-populated inflated decision tree, as described herein.
Given a trained decision tree or source decision tree, and the target depth of the tree, program code will (in one embodiment) first determine if any phantom decision nodes need to be inserted into the tree structure. For a fully-populated tree at a given depth, it is known how many decision nodes and leaf nodes are present. For any given depth of tree, a fully-populated or inflated tree will have 2 raised to the depth of the leaf nodes, and one less decision node. Thus, for a tree of depth 3, there are 8 leaf nodes and 7 decision nodes. To reach a leaf node, control will need to pass through the same number of decision nodes as the depth.
As described above,
As depicted in
The string value is inserted into the leaf node's path vector at the insertion position 734, and for other leaf nodes in the tree, a same-sized ‘don't care’ string is inserted at the same position 736. As an example, for a 2-level descent, is a ‘don't care’ string of “. . . ”. Once string values have been inserted to adjust or standardize the leaf node path vectors, processing returns to the flow of
Further details of one embodiment of facilitating processing within a computing environment, as it relates to one or more aspects of the present invention, are described with reference to
Referring to
In one embodiment, the source decision tree is a pruned decision tree, and the decision node data includes adjusted position decision node data for the inflated decision tree and the leaf node data includes adjusted position leaf node data for the inflated decision tree 1006. Advantageously, the generalization benefits of a pruned source decision tree are maintained, while inference performance is improved using the inflated, or fully-populated, decision tree for accelerated processing by the inference accelerator.
In one example, the accelerated processing includes processing accelerated, based on respective, predetermined path vectors through the inflated decision tree for each leaf node of the plurality of leaf nodes, by processing a decision node result vector by the plurality of leaf nodes in parallel 1008. Advantageously, the inference accelerator is configured to facilitate full-parallel processing by the plurality of leaf nodes of the decision node result vector based on the respective, predetermined path vector through the decision tree to each leaf node of the plurality of leaf nodes. In this manner, latency of decision tree inference processing is reduced, enhancing processing throughput of the decision tree and/or decision tree ensemble.
In one embodiment, the ascertaining includes, for a leaf node of the plurality of leaf nodes, descending the leaf node to a maximum depth of the inflated decision tree, and based thereon, adjusting a path vector for the leaf node to obtain the respective, predetermined path vector for the leaf node 1010. Descending each leaf node to a maximum depth of the inflated decision tree removes the requirement that the path to each leaf node being transmitted to the inference accelerator, thereby reducing input/output and configuration processing for the inference accelerator. Descending the leaf node to the maximum depth of the inflated decision tree results in a fully-populated inflated decision tree, meaning that one or more masks derived from the respective, predetermined path vectors to facilitate inference processing in the inference accelerator do not need to be transmitted to the accelerator.
As illustrated in
In one example, the adjusting includes determining an insertion position for the string value by identifying a last decision node in the path vector, and based on the last decision node being true, or the last decision node being at the end of the path, the insertion position is 1 position beyond the last decision node 1016. In another example, where based on the last decision node being false and other than the end of the path, the insertion position is 2 positions beyond the last decision node 1018.
In one example, for at least one other leaf node of the plurality of leaf nodes, a ‘don't care’ string is inserted at the same insertion position into its path vector, where the ‘don't care’ string is same-sized as the determined string value 1020.
Those skilled in the art will note from the description provided herein that a decision tree inference accelerator is provided which obtains a decision node result vector based on processing of an input data vector through decision nodes of, in one embodiment, an inflated decision tree. The inference accelerator accelerates, in one embodiment, leaf node processing of a decision node result vector, and facilitates, based on respective, predetermined path vectors through the inflated decision tree for each leaf node of the plurality of leaf nodes, processing of the decision node result vector by the plurality of leaf nodes in parallel, and determining therefrom which leaf node of the plurality of leaf nodes is selected. In this manner, the inference accelerator enhances processing within the computing environment by, for instance, parallel-processing by the plurality of leaf nodes the decision node result vector based on respective, predetermined path vectors through the inflated decision tree to each leaf node. Thus, reduced latency of decision tree inference processing is obtained, enhancing processing throughput of the decision tree and/or decision tree ensemble.
Other variations and embodiments are possible.
An inference accelerator of one or more aspects of the present invention may be incorporated and used in many computing environments. One example computing environment is described with reference to
As shown in
Bus 1111 is, for instance, a memory or a cache coherence bus, and bus 1110 represents, e.g., one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include the Industry Standard Architecture (ISA), the Micro Channel Architecture (MCA), the Enhanced ISA (EISA), the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI).
As examples, one or more special purpose processors (e.g., adjunct processors) can be separate from but coupled to one or more general purpose processors and/or can be embedded within one or more general purpose processors. May variations are possible.
Memory 1106 may include, for instance, a cache 1112, such as a shared cache, which may be coupled to local caches 1114 of processors 1104 and/or to adjunct processor(s) 1121, via, for instance, one or more buses 1111. Further, memory 1106 may include one or more programs or applications 1116, at least one operating system 1118, one or more computer readable program instructions 1120 and one or more interference libraries 1122. Computer readable program instructions 1120 and inference libraries 1122 can be configured to carry out, or facilitate, functions of embodiments of aspects of the invention.
By way of example, in one embodiment, memory 1106 (e.g., at least a hardware system area of memory 1106) is coupled to one or more inference accelerators 1121 via one or more buses 1111, and in one or more embodiments.
Computer system 1102 can communicate via, e.g., I/O interfaces 1108 with one or more external devices 1130, such as a user terminal, a tape drive, a pointing device, a display, and one or more data storage devices 1134, etc. A data storage device 1134 can store one or more programs 1136, one or more computer readable program instructions 1138, and/or data, etc. The computer readable program instructions can be configured to carry out functions of embodiments of aspects of the invention.
Computer system 1102 can also communicate via, e.g., I/O interfaces 1108 with network interface 1132, which enables computer system 1102 to communicate with one or more networks, such as a local area network (LAN), a general wide area network (WAN), and/or a public network (e.g., the Internet), providing communication with other computing devices or systems.
Computer system 1102 can include and/or be coupled to removable/non-removable, volatile/non-volatile computer system storage media. For example, it can include and/or be coupled to a non-removable, non-volatile magnetic media (typically called a “hard drive”), a magnetic disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a “floppy disk”), and/or an optical disk drive for reading from or writing to a removable, non-volatile optical disk, such as a CD-ROM, DVD-ROM or other optical media. It should be understood that other hardware and/or software components could be used in conjunction with computer system 1102. Examples, include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.
Computer system 1102 can be operational with numerous other general-purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations that may be suitable for use with computer system 1102 include, but are not limited to, personal computer (PC) systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems or devices, and the like.
Another embodiment of a computing environment to incorporate and use one or more aspects of the present invention is described with reference to
Native central processing unit 1212 includes one or more native registers 1220, such as one or more general purpose registers and/or one or more special purpose registers used during processing within the environment. These registers include information that represents the state of the environment at any particular point in time.
Moreover, native central processing unit 1212 executes instructions and code that are stored in memory 1214. In one particular example, the central processing unit executes emulator code 1222 stored in memory 1214. This code enables the computing environment configured in one architecture to emulate another architecture. For instance, emulator code 1222 allows machines based on architectures other than the z/Architecture hardware architecture, such as PowerPC processors, HP Superdome servers or others, to emulate the z/Architecture hardware architecture and to execute software and instructions developed based on the z/Architecture hardware architecture.
Further details relating to emulator code 1222 are described with reference to
Further, emulator code 1222 includes an emulation control routine 1240 to cause the native instructions to be executed. Emulation control routine 1240 may cause native CPU 1212 to execute a routine of native instructions that emulate one or more previously obtained guest instructions and, at the conclusion of such execution, return control to the instruction fetch routine to emulate the obtaining of the next guest instruction or a group of guest instructions. Execution of the native instructions 1236 may include loading data into a register from memory 1214; storing data back to memory from a register; or performing some type of arithmetic or logic operation, as determined by the translation routine.
Each routine is, for instance, implemented in software, which is stored in memory and executed by native central processing unit 1212. In other examples, one or more of the routines or operations are implemented in firmware, hardware, software or some combination thereof. The registers of the emulated processor may be emulated using registers 1220 of the native CPU or by using locations in memory 1214. In embodiments, guest instructions 1230, native instructions 1236 and emulator code 1222 may reside in the same memory or may be disbursed among different memory devices.
Further, in one embodiment, computing environment 1210 includes one or more inference accelerators 1215 coupled to memory 1214. The one or more accelerators are defined in one architecture and are configured to emulate another architecture. For example, an accelerator obtains guest commands of the architecture being emulated, translates the guest commands into native commands of the one architecture and executes the native commands.
The computing environments described above are only examples of computing environments that can be used. Other environments, including but not limited to, non-partitioned environments, partitioned environments, cloud environments and/or emulated environments, may be used; embodiments are not limited to any one environment. Although various examples of computing environments are described herein, one or more aspects of the present invention may be used with many types of environments. The computing environments provided herein are only examples.
Each computing environment is capable of being configured to include one or more aspects of the present invention. For instance, each may be configured for an inference acceleration facility, in accordance with one or more aspects of the present invention.
Although various embodiments are described herein, many variations and other embodiments are possible without departing from a spirit of aspects of the present invention. It should be noted that, unless otherwise inconsistent, each aspect or feature described herein, and variants thereof, may be combinable with any other aspect or feature.
One or more aspects may relate to cloud computing.
It is to be understood that although this disclosure includes a detailed description on cloud computing, implementation of the teachings recited herein are not limited to a cloud computing environment. Rather, embodiments of the present invention are capable of being implemented in conjunction with any other type of computing environment now known or later developed.
Cloud computing is a model of service delivery for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) that can be rapidly provisioned and released with minimal management effort or interaction with a provider of the service. This cloud model may include at least five characteristics, at least three service models, and at least four deployment models.
Characteristics are as follows:
On-demand self-service: a cloud consumer can unilaterally provision computing capabilities, such as server time and network storage, as needed automatically without requiring human interaction with the service's provider.
Broad network access: capabilities are available over a network and accessed through standard mechanisms that promote use by heterogeneous thin or thick client platforms (e.g., mobile phones, laptops, and PDAs).
Resource pooling: the provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, with different physical and virtual resources dynamically assigned and reassigned according to demand. There is a sense of location independence in that the consumer generally has no control or knowledge over the exact location of the provided resources but may be able to specify location at a higher level of abstraction (e.g., country, state, or datacenter).
Rapid elasticity: capabilities can be rapidly and elastically provisioned, in some cases automatically, to quickly scale out and rapidly released to quickly scale in. To the consumer, the capabilities available for provisioning often appear to be unlimited and can be purchased in any quantity at any time.
Measured service: cloud systems automatically control and optimize resource use by leveraging a metering capability at some level of abstraction appropriate to the type of service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency for both the provider and consumer of the utilized service.
Service Models are as follows:
Software as a Service (SaaS): the capability provided to the consumer is to use the provider's applications running on a cloud infrastructure. The applications are accessible from various client devices through a thin client interface such as a web browser (e.g., web-based e-mail). The consumer does not manage or control the underlying cloud infrastructure including network, servers, operating systems, storage, or even individual application capabilities, with the possible exception of limited user-specific application configuration settings.
Platform as a Service (PaaS): the capability provided to the consumer is to deploy onto the cloud infrastructure consumer-created or acquired applications created using programming languages and tools supported by the provider. The consumer does not manage or control the underlying cloud infrastructure including networks, servers, operating systems, or storage, but has control over the deployed applications and possibly application hosting environment configurations.
Infrastructure as a Service (IaaS): the capability provided to the consumer is to provision processing, storage, networks, and other fundamental computing resources where the consumer is able to deploy and run arbitrary software, which can include operating systems and applications. The consumer does not manage or control the underlying cloud infrastructure but has control over operating systems, storage, deployed applications, and possibly limited control of select networking components (e.g., host firewalls).
Deployment Models are as follows:
Private cloud: the cloud infrastructure is operated solely for an organization. It may be managed by the organization or a third party and may exist on-premises or off-premises.
Community cloud: the cloud infrastructure is shared by several organizations and supports a specific community that has shared concerns (e.g., mission, security requirements, policy, and compliance considerations). It may be managed by the organizations or a third party and may exist on-premises or off-premises.
Public cloud: the cloud infrastructure is made available to the general public or a large industry group and is owned by an organization selling cloud services.
Hybrid cloud: the cloud infrastructure is a composition of two or more clouds (private, community, or public) that remain unique entities but are bound together by standardized or proprietary technology that enables data and application portability (e.g., cloud bursting for load-balancing between clouds).
A cloud computing environment is service oriented with a focus on statelessness, low coupling, modularity, and semantic interoperability. At the heart of cloud computing is an infrastructure that includes a network of interconnected nodes.
Referring now to
Referring now to
Hardware and software layer 60 includes hardware and software components. Examples of hardware components include: mainframes 61; RISC (Reduced Instruction Set Computer) architecture based servers 62; servers 63; blade servers 64; storage devices 65; and networks and networking components 66. In some embodiments, software components include network application server software 67 and database software 68.
Virtualization layer 70 provides an abstraction layer from which the following examples of virtual entities may be provided: virtual servers 71; virtual storage 72; virtual networks 73, including virtual private networks; virtual applications and operating systems 74; and virtual clients 75.
In one example, management layer 80 may provide the functions described below. Resource provisioning 81 provides dynamic procurement of computing resources and other resources that are utilized to perform tasks within the cloud computing environment. Metering and Pricing 82 provide cost tracking as resources are utilized within the cloud computing environment, and billing or invoicing for consumption of these resources. In one example, these resources may include application software licenses. Security provides identity verification for cloud consumers and tasks, as well as protection for data and other resources. User portal 83 provides access to the cloud computing environment for consumers and system administrators. Service level management 84 provides cloud computing resource allocation and management such that required service levels are met. Service Level Agreement (SLA) planning and fulfillment 85 provide pre-arrangement for, and procurement of, cloud computing resources for which a future requirement is anticipated in accordance with an SLA.
Workloads layer 90 provides examples of functionality for which the cloud computing environment may be utilized. Examples of workloads and functions which may be provided from this layer include: mapping and navigation 91; software development and lifecycle management 92; virtual classroom education delivery 93; data analytics processing 94; transaction processing 95; and inference accelerator and related processing 96.
The present invention may be a system, a method, and/or a computer program product at any possible technical detail level of integration. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.
The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.
Aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.
These computer readable program instructions can be provided to a processor of a computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions can also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be accomplished as one step, executed concurrently, substantially concurrently, in a partially or wholly temporally overlapping manner, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.
In addition to the above, one or more aspects may be provided, offered, deployed, managed, serviced, etc. by a service provider who offers management of customer environments. For instance, the service provider can create, maintain, support, etc. computer code and/or a computer infrastructure that performs one or more aspects for one or more customers. In return, the service provider may receive payment from the customer under a subscription and/or fee agreement, as examples. Additionally or alternatively, the service provider may receive payment from the sale of advertising content to one or more third parties.
In one aspect, an application may be deployed for performing one or more embodiments. As one example, the deploying of an application comprises providing computer infrastructure operable to perform one or more embodiments.
As a further aspect, a computing infrastructure may be deployed comprising integrating computer readable code into a computing system, in which the code in combination with the computing system is capable of performing one or more embodiments.
As yet a further aspect, a process for integrating computing infrastructure comprising integrating computer readable code into a computer system may be provided. The computer system comprises a computer readable medium, in which the computer medium comprises one or more embodiments. The code in combination with the computer system is capable of performing one or more embodiments.
Although various embodiments are described above, these are only examples. For example, computing environments of other architectures can be used to incorporate and use one or more embodiments. Further, different instructions, commands or operations may be used. Additionally, different types of indications or tags may be specified. Many variations are possible.
Various aspects are described herein. Further, many variations are possible without departing from a spirit of aspects of the present invention. It should be noted that, unless otherwise inconsistent, each aspect or feature described herein, and variants thereof, may be combinable with any other aspect or feature.
Further, other types of computing environments can benefit and be used. As an example, a data processing system suitable for storing and/or executing program code is usable that includes at least two processors coupled directly or indirectly to memory elements through a system bus. The memory elements include, for instance, local memory employed during actual execution of the program code, bulk storage, and cache memory which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
Input/output or I/O devices (including, but not limited to, keyboards, displays, pointing devices, DASD, tape, CDs, DVDs, thumb drives and other memory media, etc.) can be coupled to the system either directly or through intervening I/O controllers. Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modems, and Ethernet cards are just a few of the available types of network adapters.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” and/or “comprising”, when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below, if any, are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of one or more embodiments has been presented for purposes of illustration and description but is not intended to be exhaustive or limited to in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiment was chosen and described in order to best explain various aspects and the practical application, and to enable others of ordinary skill in the art to understand various embodiments with various modifications as are suited to the particular use contemplated.
