Low Latency Geo-distributed Data Analytics

Pu, Qifan; Ananthanarayanan, Ganesh; Bodík, Peter; Kandula, Srikanth; Akella, Aditya; Bahl, Paramvir; Stoica, Ion

doi:10.1145/2785956.2787505

Cited by 194 publications

(177 citation statements)

References 39 publications

Supporting

Mentioning

176

Contrasting

Order By: Relevance

“…For example, replicating the straggling task on another available node is a common approach to deal with stragglers (e.g., [4]), while partial data replication is also used to reduce the communication load in distributed computing (e.g., [5]). However, there have been recent results demonstrating that coding can play a transformational role for creating and exploiting computation redundancy to effectively alleviate the impact of system noise.…”

Section: Introductionmentioning

confidence: 99%

Coded computation over heterogeneous clusters

Reisizadeh

Prakash

Pedarsani

et al. 2017

2017 IEEE International Symposium on Information Theory (ISIT)

108

156

View full text Add to dashboard Cite

In large-scale distributed computing clusters, such as Amazon EC2, there are several types of "system noise" that can result in major degradation of performance: system failures, bottlenecks due to limited communication bandwidth, latency due to straggler nodes, etc. On the other hand, these systems enjoy abundance of redundancy -a vast number of computing nodes and large storage capacity. There have been recent results that demonstrate the impact of coding for efficient utilization of computation and storage redundancy to alleviate the effect of stragglers and communication bottlenecks in homogeneous clusters. In this paper, we focus on general heterogeneous distributed computing clusters consisting of a variety of computing machines with different capabilities. We propose a coding framework for speeding up distributed computing in heterogeneous clusters by trading redundancy for reducing the latency of computation. In particular, we propose Heterogeneous Coded Matrix Multiplication (HCMM) algorithm for performing distributed matrix multiplication over heterogeneous clusters that is provably asymptotically optimal for a broad class of processing time distributions. Moreover, we show that HCMM is unboundedly faster than uncoded schemes that partition the total work load among the workers. To demonstrate how the proposed HCMM scheme can be applied in practice, we provide numerical results demonstrating significant speedups of up to 90% and 35% for HCMM in comparison to the "uncoded" and "coded homogeneous" schemes, respectively. Furthermore, we carry out real experiments over Amazon EC2 clusters that corroborate our numerical studies, where HCMM is found to be up to 17% faster than the uncoded scheme. Additionally, our observation is that machines rarely become stragglers and when they do, they continue to exhibit slower performance for sometime. In our worst case experiments with artificial stragglers, HCMM provides speedups of up to 12× over the uncoded scheme. Furthermore, we provide a generalization of the problem of optimal load allocation for heterogeneous clusters to scenarios with budget constraints and develop a heuristic algorithm for efficient load allocation. In the end, we discuss about the decoding complexity and describe how LDPC codes can be combined with HCMM in order to control the complexity of decoding as the problem size increases.

show abstract

Section: Introductionmentioning

confidence: 99%

Coded computation over heterogeneous clusters

Reisizadeh

Prakash

Pedarsani

et al. 2017

2017 IEEE International Symposium on Information Theory (ISIT)

108

156

View full text Add to dashboard Cite

show abstract

“…This class of techniques builds Virtual Machines (VMs) for users to use computing resources across geo-distributed datacenters as a single logical virtual cluster. These techniques primarily ptimize the data placement [10] [12], the latency of the services [12] [13] [14] , the Quality of Service(QoS) [11] [15], the electricity cost [13] [16] across multiple datacenters.…”

Section: Related Workmentioning

confidence: 99%

Optimized Contract-Based Model for Resource Allocation in Federated Geo-Distributed Clouds

Palanisamy

2021

IEEE Trans. Serv. Comput.

View full text Add to dashboard Cite

Abstract-In the era of Big Data, with data growing massively in scale and velocity, cloud computing and its pay-as-you-go model continues to provide significant cost benefits and a seamless service delivery model for cloud consumers. The evolution of small-scale and large-scale geo-distributed datacenters operated and managed by individual Cloud Service Providers (CSPs) raises new challenges in terms of effective global resource sharing and management of autonomously-controlled individual datacenter resources towards a globally efficient resource allocation model. Earlier solutions for geo-distributed clouds have focused primarily on achieving global efficiency in resource sharing, that although tries to maximize the global resource allocation, results in significant inefficiencies in local resource allocation for individual datacenters and individual cloud provi ders leading to unfairness in their revenue and profit earned. In this paper, we propose a new contracts-based resource sharing model for federated geo-distributed clouds that allows CSPs to establish resource sharing contracts with individual datacenters apriori for defined time intervals during a 24 hour time period. Based on the established contracts, individual CSPs employ a contracts cost and duration aware job scheduling and provisioning algorithm that enables jobs to complete and meet their response time requirements while achieving both global resource allocation efficiency and local fairness in the profit earned. The proposed techniques are evaluated through extensive experiments using realistic workloads generated using the SHARCNET cluster trace. The experiments demonstrate the effectiveness, scalability and resource sharing fairness of the proposed model.

show abstract

“…As a result the massive data sets, expected to be involved in big data analytics scenarios, can be geo-distributed across such global cloud platforms. Providing analytics, thus, over such data sets is fraught with efficiency and scalability limitations -as has been well recognized, for example in the Iridium project [45].…”

Section: F Research Theme 5: Global-scale Geo-distributed Seamentioning

confidence: 99%

Towards Intelligent Distributed Data Systems for Scalable Efficient and Accurate Analytics

Triantafillou

2018

2018 IEEE 38th International Conference on Distributed Computing Systems (ICDCS)

View full text Add to dashboard Cite

Low Latency Geo-distributed Data Analytics

Cited by 194 publications

References 39 publications

Coded computation over heterogeneous clusters

Coded computation over heterogeneous clusters

Optimized Contract-Based Model for Resource Allocation in Federated Geo-Distributed Clouds

Towards Intelligent Distributed Data Systems for Scalable Efficient and Accurate Analytics

Contact Info

Product

Resources

About