Abstract: In one exemplary embodiment, a computer-implemented method includes receiving, at a remote direct memory access (RDMA) device, a plurality of RDMA requests referencing a plurality of virtual pages. Data transfers are scheduled for the plurality of virtual pages, wherein the scheduling occurs at the RDMA device. The number of the virtual pages that are currently pinned is limited for the RDMA requests based on a predetermined pinned page limit.

Abstract: In one exemplary embodiment, a computer-implemented method includes receiving, at a remote direct memory access (RDMA) device, a plurality of RDMA requests referencing a plurality of virtual pages. Data transfers are scheduled for the plurality of virtual pages, wherein the scheduling occurs at the RDMA device. The number of the virtual pages that are currently pinned is limited for the RDMA requests based on a predetermined pinned page limit.

Abstract: Predicting acceleration in a hybrid system may comprise determining a number of cross system calls in a first host-accelerator computer architecture running a workload. Host machine overhead and accelerator overhead in the first host-accelerator computer architecture associated with each of the cross system calls may be determined. Communication delay associated with each of the cross system calls in the first host-accelerator computer architecture running a workload may be determined. An application response time may be predicted for a candidate application to be run in a second host-accelerator computer architecture, based at least on the determined host machine overhead, the accelerator overhead, and the communication delay associated with each of the cross system calls in the first host-accelerator computer architecture running a workload.

Abstract: Predicting acceleration in a hybrid system may comprise determining a number of cross system calls in a first host-accelerator computer architecture running a workload. Host machine overhead and accelerator overhead in the first host-accelerator computer architecture associated with each of the cross system calls may be determined. Communication delay associated with each of the cross system calls in the first host-accelerator computer architecture running a workload may be determined. An application response time may be predicted for a candidate application to be run in a second host-accelerator computer architecture, based at least on the determined host machine overhead, the accelerator overhead, and the communication delay associated with each of the cross system calls in the first host-accelerator computer architecture running a workload.

Abstract: In one exemplary embodiment, a computer-implemented method includes receiving, at a remote direct memory access (RDMA) device, a plurality of RDMA requests referencing a plurality of virtual pages. Data transfers are scheduled for the plurality of virtual pages, wherein the scheduling occurs at the RDMA device. The number of the virtual pages that are currently pinned is limited for the RDMA requests based on a predetermined pinned page limit.

Abstract: In one exemplary embodiment, a computer-implemented method includes receiving, at a remote direct memory access (RDMA) device, a plurality of RDMA requests referencing a plurality of virtual pages. Data transfers are scheduled for the plurality of virtual pages, wherein the scheduling occurs at the RDMA device. The number of the virtual pages that are currently pinned is limited for the RDMA requests based on a predetermined pinned page limit.

Abstract: Inducing perturbation by varying a supply amount of the resource type in the system and measuring performance of the software entity at multiple variation levels of the supply amount of the resource type in the system. A model may be built that characterizes a relationship between the measured performance and the variation levels. The model may be applied to detect the resource bottleneck. The model may be also applied for capacity planning.

Abstract: Inducing perturbation by varying a supply amount of the resource type in the system and measuring performance of the software entity at multiple variation levels of the supply amount of the resource type in the system. A model may be built that characterizes a relationship between the measured performance and the variation levels. The model may be applied to detect the resource bottleneck. The model may be also applied for capacity planning.

Abstract: Methods and arrangements for task scheduling. At least one job is assimilated from at least one node, each job comprising at least a map phase and a reduce phase, each of the map and reduce phases comprising at least one task. Progress of a map phase of at least one job is compared with progress of a reduce phase of at least one job. Launching of a task of a reduce phase of at least one job is scheduled in response to progress of the reduce phase of at least one job being less than progress of the map phase of at least one job.

Abstract: Methods and arrangements for task scheduling. At least one job is assimilated from at least one node, each job comprising at least a map phase and a reduce phase, each of the map and reduce phases comprising at least one task. Progress of a map phase of at least one job is compared with progress of a reduce phase of at least one job. Launching of a task of a reduce phase of at least one job is scheduled in response to progress of the reduce phase of at least one job being less than progress of the map phase of at least one job.

Abstract: Affinity-based preferential call technique, in one aspect, may improve performance of distributed applications in a hybrid system having heterogeneous platforms. A segment of code in a program being executed on a processor may be intercepted or trapped in runtime. A platform is selected in the hybrid system for executing said segment of code, the platform determined to run the segment of code with best efficiency among a plurality of platforms in the hybrid system. The segment of code is dynamically executed on the selected platform determined to run the segment of code with best efficiency.

Abstract: A system and method may be utilized to identify concurrency levels of processing stages in a distributed system, identify common resources and bottlenecks in the distributed system using the identified concurrency levels, and allocate resources in the distributed system using the identified concurrency levels.

Abstract: A system and method may be utilized to identify concurrency levels of processing stages in a distributed system, identify common resources and bottlenecks in the distributed system using the identified concurrency levels, and allocate resources in the distributed system using the identified concurrency levels.

Abstract: Methods and arrangements for task scheduling. A job is accepted, the job comprising a plurality of phases, each of the phases comprising at least one task. For each of a plurality of slots, a fetching cost associated with receipt of one or more of the tasks is determined. The slots are grouped into a plurality of sets. A pair of thresholds is determined for each of the sets, the thresholds being associated with the determined fetching costs and comprising upper and lower numerical bounds for guiding receipt of one or more of the tasks. Other variants and embodiments are broadly contemplated herein.

Abstract: Affinity-based preferential call technique, in one aspect, may improve performance of distributed applications in a hybrid system having heterogeneous platforms. A segment of code in a program being executed on a processor may be intercepted or trapped in runtime. A platform is selected in the hybrid system for executing said segment of code, the platform determined to run the segment of code with best efficiency among a plurality of platforms in the hybrid system. The segment of code is dynamically executed on the selected platform determined to run the segment of code with best efficiency.

Abstract: Techniques for monitoring a set of one or more event counters of application execution are provided. The techniques include constructing a virtual performance monitoring counter (VPMC) layer as a unified abstraction of a physical performance monitoring counter (PMC) architecture, and incorporating one or more programming interfaces (PIs) in connection with the virtual performance monitoring counter, wherein the one or more programming interfaces facilitate simultaneous access and data monitoring across a set of one or more event counters.

Abstract: In an exemplary aspect, method, apparatus, and program products are disclosed suitable for clock offset determination. One method includes performing a number of exchanges of at least single bytes with another network node, where values of the single bytes are different for the exchanges. The method also includes capturing and storing timestamps for each of the number of exchanges performed on the network node. A second method includes capturing and saving arrival timestamps for each of a number of timing messages in a set of timing messages received from another network node. This second method also includes sending the timestamps to at least the another node in response to completion of the set of timing messages.

Abstract: A system and method may be utilized to identify concurrency levels of processing stages in a distributed system, identify common resources and bottlenecks in the distributed system using the identified concurrency levels, and allocate resources in the distributed system using the identified concurrency levels.

Abstract: A system and method may be utilized to identify concurrency levels of processing stages in a distributed system, identify common resources and bottlenecks in the distributed system using the identified concurrency levels, and allocate resources in the distributed system using the identified concurrency levels.

Abstract: Methods and arrangements for task scheduling. At least one job is assimilated from at least one node, each job comprising at least a map phase and a reduce phase, each of the map and reduce phases comprising at least one task. Progress of a map phase of at least one job is compared with progress of a reduce phase of at least one job. Launching of a task of a reduce phase of at least one job is scheduled in response to progress of the reduce phase of at least one job being less than progress of the map phase of at least one job.