Iterative graph algorithms often compute intermediate values and update them as computation progresses. Updated output values are used as inputs for computations in current or subsequent iterations; hence the number of iterations required for values to converge can potentially reduce if the newest values are asynchronously made available to other updates computed in the same iteration. In a multi-threaded shared memory system, the immediate propagation of updated values can cause memory contention that may offset the benefit of propagating updates sooner. In some cases, the benefit of a smaller number of iterations may be diminished by each iteration taking longer. Our key idea is to combine the low memory contention that synchronous approaches have with the faster information sharing of asynchronous approaches. Our hybrid approach buffers updates from threads locally before committing them to the global store to control how often threads may cause conflicts for others while still sharing data within one iteration and hence speeding convergence. On a 112-thread CPU system, our hybrid approach attains up to 4.5% - 19.4% speedup over an asynchronous approach for Pagerank and up to 1.9% - 17% speedup over asynchronous Bellman Ford SSSP. Further, our hybrid approach attains 2.56x better performance than the synchronous approach. Finally, we provide insights as to why delaying updates is not helpful on certain graphs where connectivity is clustered on the main diagonal of the adjacency matrix.
This report contains two related sets of results with different assumptions on synchrony. The first part is about iterative algorithms in synchronous systems. Following our previous work on synchronous iterative approximate Byzantine consensus (IABC) algorithms, we provide a more intuitive tight necessary and sufficient condition for the existence of such algorithms in synchronous networks1. We believe this condition and the previous results also hold in partially asynchronous algorithmic model. In the second part of the report, we explore the problem in asynchronous networks. While the tradit…
Facebook AI Research（脸书人工智能研究院）；University College London（伦敦大学学院）；University of Oxford（牛津大学）；McGill University（麦吉尔大学）；MILA（蒙特利尔学习算法研究所）
Effective network congestion control strategies are key to keeping the Internet (or any large computer network) operational. Network congestion control has been dominated by hand-crafted heuristics for decades. Recently, ReinforcementLearning (RL) has emerged as an alternative to automatically optimize such control strategies. Research so far has primarily considered RL interfaces which block the sender while an agent considers its next action. This is largely an artifact of building on top of frameworks designed for RL in games (e.g. OpenAI Gym). However, this does not translate to real-world…
This document describes a new consensus algorithm which is asynchronous and uses gossip based message dissemination between nodes. The current version of the algorithm does not cover the case of a node failure or significantly delayed response. This is the subject of further research of the algorithm. An outline of a new design for trust-less payment system is given in appendices.
Myriad of graph-based algorithms in machine learning and data mining require parsing relational data iteratively. These algorithms are implemented in a large-scale distributed environment in order to scale to massive data sets. To accelerate these large-scale graph-based iterative computations, we propose delta-based accumulative iterative computation (DAIC). Different from traditional iterative computations, which iteratively update the result based on the result from the previous iteration, DAIC updates the result by accumulating the "changes" between iterations. By DAIC, we can process only…
Federated learning (FL) is a new machine learning framework which trains a joint model across a large amount of decentralized computing devices. Existing methods, e.g., Federated Averaging (FedAvg), are able to provide an optimization guarantee by synchronously training the joint model, but usually suffer from stragglers, i.e., IoT devices with low computing power or communication bandwidth, especially on heterogeneous optimization problems. To mitigate the influence of stragglers, this paper presents a novel FL algorithm, namely Hybrid Federated Learning (HFL), to achieve a learning balance i…
The advent of asynchronous iterative scheme gives high efficiency to numerical computations. However, it is generally difficult to handle the problems of resource management and convergence detection. This paper uses JACK2, an asynchronous communication kernel library for iterative algorithms, to implement both classical and asynchronous parareal algorithms, especially the latter. We illustrate the measures whereby one can tackle the problems above elegantly for the time-dependent case. Finally, experiments are presented to prove the availability and efficiency of such application.
We consider straggler-resilient learning. In many previous works, e.g., in the coded computing literature, straggling is modeled as random delays that are independent and identically distributed between workers. However, in many practical scenarios, a given worker may straggle over an extended period of time. We propose a latency model that captures this behavior and is substantiated by traces collected on Microsoft Azure, Amazon Web Services (AWS), and a small local cluster. Building on this model, we propose DSAG, a mixed synchronous-asynchronous iterative optimization method, based on the s…
University of Science and Technology of China（中国科学技术大学）；Peking University（北京大学）；Microsoft Research（微软研究院）；Academy of Mathematics and Systems Science, Chinese Academy of Sciences（中国科学院数学与系统科学研究院）
With the fast development of deep learning, it has become common to learn big neural networks using massive training data. Asynchronous Stochastic Gradient Descent (ASGD) is widely adopted to fulfill this task for its efficiency, which is, however, known to suffer from the problem of delayed gradients. That is, when a local worker adds its gradient to the global model, the global model may have been updated by other workers and this gradient becomes "delayed". We propose a novel technology to compensate this delay, so as to make the optimization behavior of ASGD closer to that of sequential SG…
Volunteer computing grids offer super-computing levels of computing power at the relatively low cost of operating a server. In previous work, the authors have shown that it is possible to take traditionally iterative evolutionary algorithms and execute them on volunteer computing grids by performing them asynchronously. The asynchronous implementations dramatically increase scalability and decrease the time taken to converge to a solution. Iterative and asynchronous optimization algorithms implemented using MPI on clusters and supercomputers, and BOINC on volunteer computing grids have been pa…
In this paper, we address the problem of designing a distributed application meant to run both classical and asynchronous iterations. MPI libraries are very popular and widely used in the scientific community, however asynchronous iterative methods raise non-negligible difficulties about the efficient management of communication requests and buffers. Moreover, a convergence detection issue is introduced, which requires the implementation of one of the various state-of-the-art termination methods, which are not necessarily highly reliable for most computational environments. We propose here an …
LAAS-CNRS（LAAS-法国国家科学研究中心）；Université de Toulouse（图卢兹大学）；CNRS（法国国家科学研究中心）
We describe several features of parallel or distributed asynchronous iterative algorithms such as unbounded delays, possible out of order messages or flexible communication. We concentrate on the concept of macroiteration sequence which was introduced in order to study the convergence or termination of asynchronous iterations. A survey of asynchronous iterations for convex optimization problems is also presented. Finally, a new result of convergence for parallel or distributed asynchronous iterative algorithms with flexible communication for convex optimization problems and machine learning is…
Distributed Stream Processing Systems (DSPSs) are among the currently most emerging topics in data management, with applications ranging from real-time event monitoring to processing complex dataflow programs and big data analytics. The major market players in this domain are clearly represented by Apache Spark and Flink, which provide a variety of frontend APIs for SQL, statistical inference, machine learning, stream processing, and many others. Yet rather few details are reported on the integration of these engines into the underlying High-Performance Computing (HPC) infrastructure and the c…
The need for scalable numerical solutions has motivated the development of asynchronous parallel algorithms, where a set of nodes run in parallel with little or no synchronization, thus computing with delayed information. This paper studies the convergence of the asynchronous parallel algorithm ARock under potentially unbounded delays. ARock is a general asynchronous algorithm that has many applications. It parallelizes fixed-point iterations by letting a set of nodes randomly choose solution coordinates and update them in an asynchronous parallel fashion. ARock takes some recent asynchronous …
We give a simple distributed algorithm for computing adjacency matrix eigenvectors for the communication graph in an asynchronous gossip model. We show how to use this algorithm to give state-of-the-art asynchronous community detection algorithms when the communication graph is drawn from the well-studied stochastic block model. Our methods also apply to a natural alternative model of randomized communication, where nodes within a community communicate more frequently than nodes in different communities. Our analysis simplifies and generalizes prior work by forging a connection between asynchr…
Asynchronous iterations are more and more investigated for both scaling and fault-resilience purpose on high performance computing platforms. While so far, they have been exclusively applied within space domain decomposition frameworks, this paper advocates a novel application direction targeting time-decomposed time-parallel approaches. Specifically, an asynchronous iterative model is derived from the Parareal scheme, for which convergence and speedup analysis are then conducted. It turned out that Parareal and async-Parareal feature very close convergence conditions, asymptotically equivalen…
This papers studies multi-agent (convex and \emph{nonconvex}) optimization over static digraphs. We propose a general distributed \emph{asynchronous} algorithmic framework whereby i) agents can update their local variables as well as communicate with their neighbors at any time, without any form of coordination; and ii) they can perform their local computations using (possibly) delayed, out-of-sync information from the other agents. Delays need not be known to the agent or obey any specific profile, and can also be time-varying (but bounded). The algorithm builds on a tracking mechanism that i…
In distributed ML applications, shared parameters are usually replicated among computing nodes to minimize network overhead. Therefore, proper consistency model must be carefully chosen to ensure algorithm's correctness and provide high throughput. Existing consistency models used in general-purpose databases and modern distributed ML systems are either too loose to guarantee correctness of the ML algorithms or too strict and thus fail to fully exploit the computing power of the underlying distributed system. Many ML algorithms fall into the category of \emph{iterative convergent algorithms} w…
This paper considers the problem of asynchronous distributed multi-agent optimization on server-based system architecture. In this problem, each agent has a local cost, and the goal for the agents is to collectively find a minimum of their aggregate cost. A standard algorithm to solve this problem is the iterative distributed gradient-descent (DGD) method being implemented collaboratively by the server and the agents. In the synchronous setting, the algorithm proceeds from one iteration to the next only after all the agents complete their expected communication with the server. However, such s…
Hybrid Edge-HPC Systems for Low-Latency Data-Driven Inference
University of California Santa Barbara（加州大学圣塔芭芭拉分校）；University of Notre Dame（圣母大学）；University of Nebraska-Lincoln（内布拉斯加大学林肯分校）；Princeton Plasma Physics Laboratory（普林斯顿等离子体物理实验室）；Rutgers-New Brunswick（罗格斯大学新不伦瑞克校区）
