Robust and flexible power-proportional storage

Amur, Hrishikesh; Cipar, James; Gupta, Varun; Ganger, Gregory R.; Kozuch, Michael A.; Schwan, Karsten

doi:10.1145/1807128.1807164

Cited by 157 publications

(151 citation statements)

References 15 publications

Supporting

Mentioning

148

Contrasting

Unclassified

Order By: Relevance

“…For instance, it is based on a simple definition of an augmenting cycle (analog of an augmenting path), leading to an augmenting blossom (which models the triangle cluster of Cornuejols and Pulleyblank); we show a maximum cardinality matching with the greatest number of augmenting blossoms gives a maximum cardinality cardinality triangle-free 2-matching (Lemma 4(i)). 4 Our proof shows that the scheduling problem is solved correctly, independently of [16,17,5]. We also extend the ideas to get similar results for arbitrary job lengths.…”

Section: The Case B = 2 and 2-matchingsmentioning

confidence: 55%

A Model for Minimizing Active Processor Time

Chang

Gabow

Khuller

2012

Algorithms – ESA 2012

View full text Add to dashboard Cite

We introduce the following elementary scheduling problem. We are given a collection of n jobs, where each job J i has an integer length ℓ i as well as a set T i of time intervals in which it can be feasibly scheduled. Given a parameter B, the processor can schedule up to B jobs at a timeslot t so long as it is "active" at t. The goal is to schedule all the jobs in the fewest number of active timeslots. The machine consumes a fixed amount of energy per active timeslot, regardless of the number of jobs scheduled in that slot (as long as the number of jobs is non-zero). In other words, subject to ℓ i units of each job i being scheduled in its feasible region and at each slot at most B jobs being scheduled, we are interested in minimizing the total time during which the machine is active. We present a linear time algorithm for the case where jobs are unit length and each T i is a single interval. For general T i , we show that the problem is N P -complete even for B = 3. However when B = 2, we show that it can be efficiently solved. In addition, we consider a version of the problem where jobs have arbitrary lengths and can be preempted at any point in time. For general B, the problem can be solved by linear programming. For B = 2, the problem amounts to finding a triangle-free 2-matching on a special graph. We extend the algorithm of Babenko et. al.[5] to handle our variant, and also to handle non-unit length jobs. This yields an O( √ Lm) time algorithm to solve the preemptive scheduling problem for B = 2, where L = i ℓ i . We also show that for B = 2 and unit length jobs, the optimal non-preemptive schedule has ≤ 4/3 times the active time of the optimal preemptive schedule; this bound extends to several versions of the problem when jobs have arbitrary length.

show abstract

Section: The Case B = 2 and 2-matchingsmentioning

confidence: 55%

A Model for Minimizing Active Processor Time

Chang

Gabow

Khuller

2012

Algorithms – ESA 2012

View full text Add to dashboard Cite

show abstract

“…Rabbit [1] provides an interesting data placement algorithm for energy proportionality in MapReduce clusters. The key idea is to place data items in a skewed way across the nodes in the cluster.…”

Section: ) Mapreduce Cluster Energy Managementmentioning

confidence: 99%

Energy Proportionality and Performance in Data Parallel Computing Clusters

Kim

Chou

Rotem

2011

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract-Energy consumption in datacenters has recently become a major concern due to the rising operational costs and scalability issues. Recent solutions to this problem propose the principle of energy proportionality, i.e., the amount of energy consumed by the server nodes must be proportional to the amount of work performed. For data parallelism and fault tolerance purposes, most common file systems used in MapReduce-type clusters maintain a set of replicas for each data block. A covering subset is a group of nodes that together contain at least one replica of the data blocks needed for performing computing tasks. In this work, we develop and analyze algorithms to maintain energy proportionality by discovering a covering subset that minimizes energy consumption while placing the remaining nodes in lowpower standby mode. Our algorithms can also discover covering subset in heterogeneous computing environments. In order to allow more data parallelism, we generalize our algorithms so that it can discover k-covering subset, i.e., a set of nodes that contain at least k replicas of the data blocks. Our experimental results show that we can achieve substantial energy saving without significant performance loss in diverse cluster configurations and working environments.

show abstract

“…Chase et al discuss how to turn servers on and off based on demand in datacenters [9] and more recent work [12,15,5] specifically tackles the problem of designing a distributed file system to be power-proportional. We Since we did not have enough Asterix-II nodes to build a cluster, the total power for this case is projected.…”

Section: Related Workmentioning

confidence: 99%

“…In fact, a data layout policy based on a randomized approach, such as those used in the Hadoop Distributed File System (HDFS) [1] and GFS [10], severely restricts the number of nodes that can be turned off. In response, alternate data layout policies, such as those proposed in [5,15], attempt to bias data storage in a manner that allows performance to be traded off with the number of nodes turned off in a cluster. These policies maintain a primary copy of the data on a small subset of cluster nodes, which represents the lowest power setting.…”

Section: Introductionmentioning

confidence: 99%

Achieving Power-Efficiency in Clusters without Distributed File System Complexity

Amur

Schwan

2011

Computer Architecture

Self Cite

View full text Add to dashboard Cite

Power-efficient operation is a desirable property, particularly for large clusters housed in datacenters. Recent work has advocated turning off entire nodes to achieve power-proportionality, but this leads to problems with availability and fault tolerance because of the resulting limits imposed on the replication strategies used by the distributed file systems (DFS) employed in these environments, with counter-measures adding substantial complexity to DFS designs. To achieve power-efficiency for a cluster without impacting data availability and recovery from failures and maintain simplicity in DFS design, our solution exploits cluster nodes that have the ability to operate in at least two extreme system-level power states, characterized by minimum vs. maximum power consumption and performance. The paper describes a cluster built with power-efficient node prototypes and presents experimental evaluations to demonstrate power-efficiency.

show abstract

Robust and flexible power-proportional storage

Cited by 157 publications

References 15 publications

A Model for Minimizing Active Processor Time

A Model for Minimizing Active Processor Time

Energy Proportionality and Performance in Data Parallel Computing Clusters

Achieving Power-Efficiency in Clusters without Distributed File System Complexity

Contact Info

Product

Resources

About