Computing at Massive Scale: Scalability and Dependability Challenges

Abstract

-Large-scale Cloud systems and big data analytics frameworks are now widely used for practical services and applications. However, with the increase of data volume, the heterogeneity of workloads and resources, together with the dynamicity of massive user requests, the uncertainties and complexity of resource scheduling and service provisioning increase dramatically, often resulting in poor resource utilization, vulnerable system dependability, and negative user-perceived performance impacts. In this paper, we present our understanding of the current and future challenges in this particular area, and discuss both existing and potential solutions to the problems, especially those concerned with system efficiency, scalability and dependability. We first introduce a data-driven analysis methodology for characterizing the resource and workload patterns and tracing performance bottlenecks in a massive-scale distributed computing environment. We then examine and analyze several fundamental challenges and the solutions we are developing to tackle them, including for example, incremental resource scheduling and incremental messaging communication, decentralized scheduling, rapid system failover, and request handling parallelism etc. We integrate these techniques together with the data analytics methodology to drive a novel big data and service engineering to facilitate the system optimization, tuning and verification. We aim to develop and offer innovative methods and mechanisms for future computing platforms that will provide strong support for new big data and IoE (Internet of Everything) applications.

Intelligent Resource Scheduling at Scale: A Machine Learning Perspective

“…With big data processing demands soaring and service decoupling, there is a general manifestation that heterogeneous workloads (in terms of execution durations, resource patterns, etc.) run and operate in the data center cluster [39]. Herein, we make coarse comparisons shown in Table 1.…”

Massive-Scale Automation in Cyber-Physical Systems: Vision & Challenges

“…Specifically these data sets are too big to store on a single machine and so must be distributed. Already the growth of data is exponential [54] and increasing data collection and further cloud services will only accelerate this further [55]. Very quickly this could lead to a situation where we are no longer able to process the vast amount of data being collected.…”

Reliable Computing Service in Massive-Scale Systems through Rapid Low-Cost Failover

“…Faults may occur simultaneously and in any aspect of system operations ranging from application to hardware, and may have a wide variety of causes, including insufficient memory (OOM), overweight system utilization, performance interference, network congestion, server faults (e.g. disk, middleware software), and applications crash or hanging etc [8].…”

“…Dependability is a key concern for Cloud resource managers due to increasingly common failures which are now the norm rather than the exception caused by the enlarged system scale and complexity [6] [7] [8], different workload characteristics, and plethora of faults types that can activate. Failures within a resource manager have the potential to cause significant economic consequences to Cloud providers due to loss of service to consumers [9], and could affect services provisioned to millions globally in the event of Manuscript received Jun 10, 2015; revised Dec 31, 2015; accepted Mar 3, 2016. catastrophic failures.…”