skip to main content

Title: Heterogeneous Computation Assignments in Coded Elastic Computing
We study the optimal design of a heterogeneous coded elastic computing (CEC) network where machines have varying relative computation speeds. CEC introduced by Yang et al. is a framework which mitigates the impact of elastic events, where machines join and leave the network. A set of data is distributed among storage constrained machines using a Maximum Distance Separable (MDS) code such that any subset of machines of a specific size can perform the desired computations. This design eliminates the need to re-distribute the data after each elastic event. In this work, we develop a process for an arbitrary heterogeneous computing network to minimize the overall computation time by defining an optimal computation load, or number of computations assigned to each machine. We then present an algorithm to define a specific computation assignment among the machines that makes use of the MDS code and meets the optimal computation load.  more » « less
Award ID(s):
1824558 1817154
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
2020 IEEE International Symposium on Information Theory (ISIT)
Page Range / eLocation ID:
168 to 173
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Our extensive real measurements over Amazon EC2 show that the virtual instances often have different computing speeds even if they share the same configurations. This motivates us to study heterogeneous Coded Storage Elastic Computing (CSEC) systems where machines, with different computing speeds, join and leave the network arbitrarily over different computing steps. In CSEC systems, a Maximum Distance Separable (MDS) code is used for coded storage such that the file placement does not have to be re-defined with each elastic event. Computation assignment algorithms are used to minimize the computation time given computation speeds of different machines. While previous studies of heterogeneous CSEC do not include stragglers - the slow machines during the computation, we develop a new framework in heterogeneous CSEC that introduces straggler tolerance. Based on this framework, we design a novel algorithm using our previously proposed approach for heterogeneous CSEC such that the system can handle any subset of stragglers of a specified size while minimizing the computation time. Furthermore, we establish a trade-off in computation time and straggler tolerance. Another major limitation of existing CSEC designs is the lack of practical evaluations using real applications. In this paper, we evaluate the performance of our designs on Amazon EC2 for applications of the power iteration and linear regression. Evaluation results show that the proposed heterogeneous CSEC algorithms outperform the state-of-the-art designs by more than 30%. 
    more » « less
  2. Elasticity is one important feature in modern cloud computing systems and can result in computation failure or significantly increase computing time. Such elasticity means that virtual machines over the cloud can be preempted under a short notice (e.g., hours or minutes) if a high-priority job appears; on the other hand, new virtual machines may become available over time to compensate the computing resources. Coded Storage Elastic Computing (CSEC) introduced by Yang et al. in 2018 is an effective and efficient approach to overcome the elasticity and it costs relatively less storage and computation load. However, one of the limitations of the CSEC is that it may only be applied to certain types of computations (e.g., linear) and may be challenging to be applied to more involved computations because the coded data storage and approximation are often needed. Hence, it may be preferred to use uncoded storage by directly copying data into the virtual machines. In addition, based on our own measurement, virtual machines on Amazon EC2 clusters often have heterogeneous computation speed even if they have exactly the same configurations (e.g., CPU, RAM, I/O cost). In this paper, we introduce a new optimization framework on Uncoded Storage Elastic Computing (USEC) systems with heterogeneous computing speed to minimize the overall computation time. Under this framework, we propose optimal solutions of USEC systems with or without straggler tolerance using different storage placements. Our proposed algorithms are evaluated using power iteration applications on Amazon EC2. 
    more » « less
  3. In cloud computing systems, elastic events and stragglers increase the uncertainty of the system, leading to computation delays. Coded elastic computing (CEC) introduced by Yang et al. in 2018 is a framework which mitigates the impact of elastic events using Maximum Distance Separable (MDS) coded storage. It proposed a CEC scheme for both matrix-vector multiplication and general matrix-matrix multiplication applications. However, in these applications, the proposed CEC scheme cannot tolerate stragglers due to the limitations imposed by MDS codes. In this paper we propose a new elastic computing scheme using uncoded storage and Lagrange coded computing approaches. The proposed scheme can effectively mitigate the effects of both elasticity and stragglers. Moreover, it produces a lower complexity and smaller recovery threshold compared to existing coded storage based schemes. 
    more » « less
  4. Deep neural network (DNN) accelerators as an example of domain-specific architecture have demonstrated great success in DNN inference. However, the architecture acceleration for equally important DNN training has not yet been fully studied. With data forward, error backward and gradient calculation, DNN training is a more complicated process with higher computation and communication intensity. Because the recent research demonstrates a diminishing specialization return, namely, “accelerator wall”, we believe that a promising approach is to explore coarse-grained parallelism among multiple performance-bounded accelerators to support DNN training. Distributing computations on multiple heterogeneous accelerators to achieve high throughput and balanced execution, however, remaining challenging. We present ACCPAR, a principled and systematic method of determining the tensor partition among heterogeneous accelerator arrays. Compared to prior empirical or unsystematic methods, ACCPAR considers the complete tensor partition space and can reveal previously unknown new parallelism configurations. ACCPAR optimizes the performance based on a cost model that takes into account both computation and communication costs of a heterogeneous execution environment. Hence, our method can avoid the drawbacks of existing approaches that use communication as a proxy of the performance. The enhanced flexibility of tensor partitioning in ACCPAR allows the flexible ratio of computations to be distributed among accelerators with different performances. The proposed search algorithm is also applicable to the emerging multi-path patterns in modern DNNs such as ResNet. We simulate ACCPAR on a heterogeneous accelerator array composed of both TPU-v2 and TPU-v3 accelerators for the training of large-scale DNN models such as Alexnet, Vgg series and Resnet series. The average performance improvements of the state-of-the-art “one weird trick” (OWT) and HYPAR, and ACCPAR, normalized to the baseline data parallelism scheme where each accelerator replicates the model and processes different input data in parallel, are 2.98×, 3.78×, and 6.30×, respectively. 
    more » « less
  5. null (Ed.)
    We propose a flexible low complexity design (FLCD) of coded distributed computing (CDC) with empirical evaluation on Amazon Elastic Compute Cloud (Amazon EC2). CDC can expedite MapReduce like computation by trading increased map computations to reduce communication load and shuffle time. A main novelty of FLCD is to utilize the design freedom in defining map and reduce functions to develop asymptotic homogeneous systems to support varying intermediate values (IV) sizes under a general MapReduce framework. Compared to existing designs with constant IV sizes, FLCD offers greater flexibility in adapting to network parameters and significantly reduces the implementation complexity by requiring fewer input files and shuffle groups. The FLCD scheme is the first proposed low-complexity CDC design that can operate on a network with an arbitrary number of nodes and computation load. We perform empirical evaluations of the FLCD by executing the TeraSort algorithm on an Amazon EC2 cluster. This is the first time that theoretical predictions of the CDC shuffle time are validated by empirical evaluations. The evaluations demonstrate a 2.0 to 4.24 speedup compared to conventional uncoded MapReduce, a 12% to 52% reduction in total time, and a wider range of operating network parameters compared to existing CDC schemes. 
    more » « less