Tolerating Process Imbalance in Hierarchical Collective Operations
نویسندگان
چکیده
This work studies the temporal imbalance that processes experience when operating on a modern multicore cluster, and develops hierarchical algorithms that tolerate this imbalance. A micro-benchmark is used to investigate the nature of process imbalance with perfectly balanced workloads, and understand the nature of interversus intra-node imbalance. These insights are then used to develop imbalance tolerant reduction and broadcast algorithms, that minimize the synchronization delay observed by early arriving processes. Preliminary results show speed-ups reaching 80x for reduce and 14x for broadcast over a binomial tree algorithm in the presence of high, but not unreasonable, imbalance.
منابع مشابه
Accelerating MPI collective communications through hierarchical algorithms with flexible inter-node communication and imbalance awareness
This work investigates collective communication algorithms on a shared memory system, and develops the universal hierarchical algorithm. This algorithm can pair arbitrary hierarchy unaware inter-node communication algorithms with shared memory intra-node communication. In addition to flexible inter-node communication, this algorithm works with all collectives, including those incompatible with ...
متن کاملMPI collectives at scale
Collective operations improve the performance and reduce code complexity of many applications parallelized with the messagepassing interface (MPI) paradigm. In this article, we will investigate the impact of load imbalance on the performance of collective operations and possibility for hiding parallel overhead caused by a collective communication pattern, by overlapping the communication with c...
متن کاملA New Method for Duplicate Detection Using Hierarchical Clustering of Records
Accuracy and validity of data are prerequisites of appropriate operations of any software system. Always there is possibility of occurring errors in data due to human and system faults. One of these errors is existence of duplicate records in data sources. Duplicate records refer to the same real world entity. There must be one of them in a data source, but for some reasons like aggregation of ...
متن کاملPower Imbalance in Collaboration Relationships
Power plays a significant role in many organizational theories such as resource dependency theory and transaction cost economics. It allows the strong companies to win more than others, or more broadly, to coerce others to do what they would not otherwise do. Power can seriously affect the confidence and commitment between parties. This paper aims to analyze the power concept in inter-organizat...
متن کاملAn Efficient Collective Communication Method for Grid Scale Networks
This research is to design a collection of effective collective operations for the Grid scale network environment. In the Grid, several network features should be specified and adapted into the algorithmic design of collective operations. For this, we characterize the Grid and design hierarchical latency optimal tree algorithm for the MPI (message passing interface) library. The experimental re...
متن کامل