Document Actions
Dynamic Adaption and Steering
Application Mapping, Dynamic Adaptation and Steering
As computer systems grow in size and complexity, tool support is needed
to facilitate the efficient mapping of large-scale applications onto
these systems. Today, most applications are mapped to a set of
resources at program launch and then run to completion using these
resources. However, large-scale systems built from commodity components
are prone to failure and long-running applications for such systems
must sense and respond to component failure.
Intelligent mapping and performance steering offer an opportunity to adjust a running program for more efficient execution and to adapt to changing resource availability (e.g., due to component failures or resource sharing). A challenge is to develop strategies that enable applications running on ASC-scale systems to monitor their own behavior and reactively adjust their behavior to optimize performance according to one or more metrics. For this purpose, performance analysis tools must provide robust performance observation capabilities at all levels of the system and the ability to map low-level behavior to high-level program constructs. Our goal is to develop tools and approaches that can help applications achieve high performance even when system components fail or applications are subject to other system constraints – managing the challenge of large scale and integration with multiple subsystems. This work will explicitly target the LANL Clustermatic infrastructure.
Our goal is to develop tools and approaches that can help applications achieve high performance even when system components fail or applications are subject to other system constraints. Strategies for automatic performance steering based on performance and fault models offer the potential to enable long-running programs to repeatedly adjust themselves to changes in the execution environment – perhaps to opportunistically acquire more resources as they become available, to rebalance load, or adapt to component failures. Moreover, measurement of environmental conditions on nodes promises to allow users and schedulers to balance checkpoint frequency and partition allocation based on failure likelihood.
In addition, validated performance “contracts” among applications, systems, and users that combine temporal and behavioral reasoning from performance predictions, previous executions, and compile-time analyses are one promising approach. This work continues to explore the use of performance contracts to guide the monitoring of application and resource behavior; contracts will include dynamic performance signatures and techniques for locally (per process) and globally (per application and per system) evaluating observed behavior relative to that expected.
Intelligent mapping and performance steering offer an opportunity to adjust a running program for more efficient execution and to adapt to changing resource availability (e.g., due to component failures or resource sharing). A challenge is to develop strategies that enable applications running on ASC-scale systems to monitor their own behavior and reactively adjust their behavior to optimize performance according to one or more metrics. For this purpose, performance analysis tools must provide robust performance observation capabilities at all levels of the system and the ability to map low-level behavior to high-level program constructs. Our goal is to develop tools and approaches that can help applications achieve high performance even when system components fail or applications are subject to other system constraints – managing the challenge of large scale and integration with multiple subsystems. This work will explicitly target the LANL Clustermatic infrastructure.
Our goal is to develop tools and approaches that can help applications achieve high performance even when system components fail or applications are subject to other system constraints. Strategies for automatic performance steering based on performance and fault models offer the potential to enable long-running programs to repeatedly adjust themselves to changes in the execution environment – perhaps to opportunistically acquire more resources as they become available, to rebalance load, or adapt to component failures. Moreover, measurement of environmental conditions on nodes promises to allow users and schedulers to balance checkpoint frequency and partition allocation based on failure likelihood.
In addition, validated performance “contracts” among applications, systems, and users that combine temporal and behavioral reasoning from performance predictions, previous executions, and compile-time analyses are one promising approach. This work continues to explore the use of performance contracts to guide the monitoring of application and resource behavior; contracts will include dynamic performance signatures and techniques for locally (per process) and globally (per application and per system) evaluating observed behavior relative to that expected.
