Abstract: According to an aspect, a candidate token sequence including one or more word tokens is extracted from an unstructured domain glossary that includes entries associated with a domain. A look-up operation is performed to retrieve language data for each word token in the candidate token sequence and annotates each word token in the candidate token sequence found by the look-up operation with corresponding retrieved language data to form an annotated sequence. A pattern match of the annotated sequence is performed relative to a repository of patterns and identifies a best matching pattern from the repository of patterns to the annotated sequence based on matching criteria. The annotated sequence is refined with lexical information associated with the best matching pattern as a refined annotated sequence. The candidate token sequence and the refined annotated sequence are output to a domain-specific computational lexicon file.

Abstract: According to an aspect, a processing system of a question answering computer system determines a first set of relations between one or more pairs of terms in a question. The processing system also determines a second set of relations between one or more pairs of terms in a candidate passage including a candidate answer to the question. The processing system matches the first set of relations to the second set of relations. A plurality of scores is determined by the processing system based on the matching. The processing system aggregates the scores to produce an answer score indicative of a level of support that the candidate answer correctly answers the question.

Abstract: According to an aspect, a processing system of a question answering computer system determines a first set of relations between one or more pairs of terms in a question. The processing system also determines a second set of relations between one or more pairs of terms in a candidate passage including a candidate answer to the question. The processing system matches the first set of relations to the second set of relations. A plurality of scores is determined by the processing system based on the matching. The processing system aggregates the scores to produce an answer score indicative of a level of support that the candidate answer correctly answers the question.

Abstract: A method for executing kernels in a hybrid system includes running a program on a host computer and identifying in an instruction stream of the program a first instruction including a function of a target classification. The method includes generating a first kernel including the function and transmitting the first kernel to a client system to execute the first kernel based on identifying the first instruction as being of the target classification. The method also includes determining whether to store results of executing the first kernel in a read-only buffer of the client system based on determining whether a subsequent instruction of the target classification relies upon results of the first instruction.

Abstract: Embodiments of the present invention manage workloads in a high-throughput computing environment for a hybrid processing system. A set of high-throughput computing service level agreements (SLAs) is retrieved. The set of SLAs is associated with a hybrid processing system including a server system and a set of accelerator systems, where each system has a different architecture. A first set of data-parallel workload tasks scheduled on the server system and a second set of data-parallel workload tasks scheduled with the set of accelerator systems are identified. At least a portion of one of the first set of data-parallel workload tasks and the second set of data-parallel workload tasks is dynamically rescheduled on a second one of the server system and the set of accelerator systems. The dynamically rescheduling substantially satisfies the set of high-throughput computing SLAs.

Abstract: A system for managing communications to add a first Remote Direct Memory Access (RDMA) link between a TCP server and a TCP client, where the first RDMA link references first remote memory buffer (RMB) and a second RMB, and further based on a first remote direct memory access network interface card (RNIC) associated with the TCP server and a second RNIC associated with the TCP client. The system determines whether a third RNIC is enabled. The system adds a second RDMA link, responsive to a determination that the third RNIC is enabled. The system detects a failure in a failed RDMA link. The system reconfigures the first RDMA link to carry at least one TCP message of a connection formerly assigned to the failed RDMA link, responsive to detecting the failure. The system communicates at least one message of the at least one connection on the first RDMA link.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes managing workloads on a first processor with a first processor architecture by an agent process executing on a second processor with a second processor architecture. The method proceeds by activating redundant computation on the second processor by the agent process. The method continues by performing a same computation from a workload of the workloads at least twice. Finally, the method includes comparing results of the same computation. In this embodiment the first processor is coupled the second processor by a network, and the first processor architecture and second processor architecture are different architectures.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes managing workloads on a first processor with a first processor architecture by an agent process executing on a second processor with a second processor architecture. The method proceeds by activating redundant computation on the second processor by the agent process. The method continues by performing a same computation from a workload of the workloads at least twice. Finally, the method includes comparing results of the same computation. In this embodiment the first processor is coupled the second processor by a network, and the first processor architecture and second processor architecture are different architectures.

Abstract: A method for executing kernels in a hybrid system includes running a program on a host computer and identifying in an instruction stream of the program a first instruction including a function of a target classification. The method includes generating a first kernel including the function and transmitting the first kernel to a client system to execute the first kernel based on identifying the first instruction as being of the target classification. The method also includes determining whether to store results of executing the first kernel in a read-only buffer of the client system based on determining whether a subsequent instruction of the target classification relies upon results of the first instruction.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes performing a first data computation by a first set of processors, the first set of processors having a first computer processor architecture. The method continues by performing a second data computation by a second processor coupled to the first set of processors, the second processor having a second computer processor architecture, the first computer processor architecture being different than the second computer processor architecture. Finally, the method includes dynamically allocating computational resources of the first set of processors and the second processor based on at least one metric while the first set of processors and the second processor are in operation such that the accuracy and processing speed of the first data computation and the second data computation are optimized.

Abstract: Embodiments of the present invention provide high-throughput computing in a hybrid processing system. A set of high-throughput computing service level agreements (SLAs) is analyzed. The set of high-throughput computing SLAs are associated with a hybrid processing system. The hybrid processing system includes at least one server system that includes a first computing architecture and a set of accelerator systems each including a second computing architecture that is different from the first computing architecture. A first set of resources at the server system and a second set of resources at the set of accelerator systems are monitored. A set of data-parallel workload tasks is dynamically scheduled across at least one resource in the first set of resources and at least one resource in the second set of resources. The dynamic scheduling of the set of data-parallel workload tasks substantially satisfies the set of high-throughput computing SLAs.

Abstract: A method includes processing an application by a host including one or more processors and distributing at least a portion of the application to an accelerator for execution. The method includes instructing the accelerator to create a buffer on the accelerator and instructing the accelerator to execute the portion of the application, wherein the accelerator writes data to the buffer. The method also includes instructing the accelerator to transmit the data in the buffer to the host before the application requests the data in the buffer. The accelerator aggregates the data in the buffer before transmitting the data to the host based upon one or more runtime conditions in the host.

Abstract: A system for distributed function execution, the system includes a host in operable communication with an accelerator. The system is configured to perform a method including processing an application by the host and distributing at least a portion of the application to the accelerator for execution. The method also includes instructing the accelerator to create a buffer on the accelerator, instructing the accelerator to execute the portion of the application, wherein the accelerator writes data to the buffer and instructing the accelerator to transmit the data in the buffer to the host before the application requests the data in the buffer. The accelerator aggregates the data in the buffer before transmitting the data to the host based upon one or more runtime conditions in the host.

Abstract: A system for managing communications to add a first Remote Direct Memory Access (RDMA) link between a TCP server and a TCP client, where the first RDMA link references first remote memory buffer (RMB) and a second RMB, and further based on a first remote direct memory access network interface card (RNIC) associated with the TCP server and a second RNIC associated with the TCP client. The system determines whether a third RNIC is enabled. The system adds a second RDMA link, responsive to a determination that the third RNIC is enabled. The system detects a failure in a failed RDMA link. The system reconfigures the first RDMA link to carry at least one TCP message of a connection formerly assigned to the failed RDMA link, responsive to detecting the failure. The system communicates at least one message of the at least one connection on the first RDMA link.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes performing a first data computation by a first set of processors, the first set of processors having a first computer processor architecture. The method continues by performing a second data computation by a second processor coupled to the first set of processors, the second processor having a second computer processor architecture, the first computer processor architecture being different than the second computer processor architecture. Finally, the method includes dynamically allocating computational resources of the first set of processors and the second processor based on at least one metric while the first set of processors and the second processor are in operation such that the accuracy and processing speed of the first data computation and the second data computation are optimized.

Abstract: A system for distributed function execution, the system includes a host in operable communication with an accelerator. The system is configured to perform a method including processing an application by the host and distributing at least a portion of the application to the accelerator for execution. The method also includes instructing the accelerator to create a buffer on the accelerator, instructing the accelerator to execute the portion of the application, wherein the accelerator writes data to the buffer and instructing the accelerator to transmit the data in the buffer to the host before the application requests the data in the buffer. The accelerator aggregates the data in the buffer before transmitting the data to the host based upon one or more runtime conditions in the host.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes performing a first data computation by a first set of processors, the first set of processors having a first computer processor architecture. The method continues by performing a second data computation by a second processor coupled to the first set of processors, the second processor having a second computer processor architecture, the first computer processor architecture being different than the second computer processor architecture. Finally, the method includes dynamically allocating computational resources of the first set of processors and the second processor based on at least one metric while the first set of processors and the second processor are in operation such that the accuracy and processing speed of the first data computation and the second data computation are optimized.

Abstract: A method, system, and computer program product for maintaining reliability in a computer system. In an example embodiment, the method includes managing workloads on a first processor with a first processor architecture by an agent process executing on a second processor with a second processor architecture. The method proceeds by activating redundant computation on the second processor by the agent process. The method continues by performing a same computation from a workload of the workloads at least twice. Finally, the method includes comparing results of the same computation. In this embodiment the first processor is coupled the second processor by a network, and the first processor architecture and second processor architecture are different architectures.

Abstract: Embodiments of the present invention manage workloads in a high-throughput computing environment for a hybrid processing system. A set of high-throughput computing service level agreements (SLAs) is retrieved. The set of SLAs is associated with a hybrid processing system including a server system and a set of accelerator systems, where each system has a different architecture. A first set of data-parallel workload tasks scheduled on the server system and a second set of data-parallel workload tasks scheduled with the set of accelerator systems are identified. At least a portion of one of the first set of data-parallel workload tasks and the second set of data-parallel workload tasks is dynamically rescheduled on a second one of the server system and the set of accelerator systems. The dynamically rescheduling substantially satisfies the set of high-throughput computing SLAs.

Abstract: Embodiments of the present invention provide high-throughput computing in a hybrid processing system. A set of high-throughput computing service level agreements (SLAs) is analyzed. The set of high-throughput computing SLAs are associated with a hybrid processing system. The hybrid processing system includes at least one server system that includes a first computing architecture and a set of accelerator systems each including a second computing architecture that is different from the first computing architecture. A first set of resources at the server system and a second set of resources at the set of accelerator systems are monitored. A set of data-parallel workload tasks is dynamically scheduled across at least one resource in the first set of resources and at least one resource in the second set of resources. The dynamic scheduling of the set of data-parallel workload tasks substantially satisfies the set of high-throughput computing SLAs.