Tyrex: Size-Based Resource Allocation in MapReduce Frameworks

Abstract

Abstract-Many large-scale data analytics infrastructures are employed for a wide variety of jobs, ranging from short interactive queries to large data analysis jobs that may take hours or even days to complete. As a consequence, data-processing frameworks like MapReduce may have workloads consisting of jobs with heavy-tailed processing requirements. With such workloads, short jobs may experience slowdowns that are an order of magnitude larger than large jobs do, while the users may expect slowdowns that are more in proportion with the job sizes. To address this problem of large job slowdown variability in MapReduce frameworks, we design a scheduling system called TYREX that is inspired by the well-known TAGS task assignment policy in distributed-server systems. In particular, TYREX partitions the resources of a MapReduce framework, allowing any job running in any partition to read data stored on any machine, imposes runtime limits in the partitions, and successively executes parts of jobs in a work-conserving way in these partitions until they can run to completion. We develop a statistical model for dynamically setting the runtime limits that achieves nearoptimal job slowdown performance, and we empirically evaluate TYREX on a cluster system with workloads consisting of both synthetic and real-world benchmarks. We find that TYREX cuts in half the job slowdown variability while preserving the median job slowdown when compared to state-of-the-art MapReduce schedulers such as FIFO and FAIR. Furthermore, TYREX reduces the job slowdown at the 95 th percentile by more than 50% when compared to FIFO and by 20-40% when compared to FAIR.

A Flexible Heuristic to Schedule Distributed Analytic Applications in Compute Clusters

“…Such schedulers are designed to support a specific "data-flow" programming model, but many of their design choices can also be used at a higher level. For example, Tyrex [20] and HFSP [53], [54] are a sample of size-based schedulers, which is a family of policies known to drastically improve turnaround times, as we also have verified with our experiments. Similarly, Quincy [17] and DRF [55] study max-min fair, task-level resource allocation, specifically working on multi-dimensional resources.…”

“…The context depicted above has driven a lot of research [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21] (see Section 7 for a detailed discussion) in the area of resource allocation and scheduling, both from academia and the industry. These efforts materialize in cluster management systems that offer simple mechanisms for users to request the deployment of the framework they need.…”

Flexible Scheduling of Distributed Analytic Applications

“…The context depicted above has driven a lot of research [8][9][10][11][12][13][14][15][16][17][18][19][20][21] in the area of resource allocation and scheduling, both from academia and the industry. These efforts materialize in cluster management systems that offer simple mechanisms for users to request the deployment of the framework they need.…”

Kairos

“…Tyrex [16] aims to avoid head-of-line blocking by partitioning the workload in classes depending on task runtimes, and by assigning different classes to disjoint partitions of worker nodes. Because runtimes are not known a priori, workload partitioning is achieved by initially assigning all tasks to partition 1, and then migrating a task from partition i to i + 1 when the task execution time has exceeded a threshold t i .…”