ROSE: Cluster Resource Scheduling via Speculative Over-Subscription

Abstract

Abstract-A long-standing challenge in cluster scheduling is to achieve a high degree of utilization of heterogeneous resources in a cluster. In practice there exists a substantial disparity between perceived and actual resource utilization. A scheduler might regard a cluster as fully utilized if a large resource request queue is present, but the actual resource utilization of the cluster can be in fact very low. This disparity results in the formation of idle resources, leading to inefficient resource usage and incurring high operational costs and an inability to provision services. In this paper we present a new cluster scheduling system, ROSE, that is based on a multi-layered scheduling architecture with an ability to over-subscribe idle resources to accommodate unfulfilled resource requests. ROSE books idle resources in a speculative manner: instead of waiting for resource allocation to be confirmed by the centralized scheduler, it requests intelligently to launch tasks within machines according to their suitability to oversubscribe resources. A threshold control with timely task rescheduling ensures fully-utilized cluster resources without generating potential task stragglers. Experimental results show that ROSE can almost double the average CPU utilization, from 36.37% to 65.10%, compared with a centralized scheduling scheme, and reduce the workload makespan by 30.11%, with an 8.23% disk utilization improvement over other scheduling strategies.Index Terms-cluster scheduling, resource management, oversubscription

Perph: A Workload Co-location Agent with Online Performance Prediction and Resource Inference

“…The procedure allows for quick resource adjustment used in Perph and makes it free from RM temporarily. This is largely backed by our previous work [30]. To synchronize resource usage, NM will coordinate the oversubscribed and preempted resource slices with RM for timely updating.…”

“…In the future, we intend to federate individual agents and coordinate the model learning. We also plan to integrate Perph mechanism with our previous work on resource over-subscription [30] to supervise the workload co-location considering both LRA's runtime performance and batch job's throughput.…”

“…To make the procedure fair and avoid resource starvation, DRF [33], capacity scheduling [34] or fairness scheduling [35] are adopted in the resource sharing among multiple jobs. Decentralized approaches [10] [36] [30] are developed for clusters that expect a high throughput or high cluster utilization. They mainly optimize task placement to enable that very short, subsecond tasks can quickly access to idle resources.…”

Modeling and Analyzing Offloading Strategies of IoT Applications over Edge Computing and Joint Clouds

“…Unfortunately, if the forecast on the next following task is incorrect, the reservation becomes a kind of resource waste. Introducing overbooking technology [37] can cope with the problem and improve the tolerance of offloading architecture for inaccurate estimation.…”

TOPOSCH: Latency-Aware Scheduling Based on Critical Path Analysis on Shared YARN Clusters

“…Current cluster managers [5] [6] [7][8] [9] are designed for short-running tasks within batch jobs, whose performance is minimally affected when launching additional tasks. The central resource manager (RM) is application-agnostic and completely unaware of runtime QoS requirements of interactive and latency-sensitive applications; RM is only responsible for resource allocation among jobs but leaves all application-specific logic to application managers.…”