Amdahl's Law in the Datacenter Era: A Market for Fair Processor Allocation

Zahedi, Seyed Majid; Llull, Qiuyun; Lee, Benjamin C.

doi:10.1109/hpca.2018.00011

Cited by 8 publications

(2 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Game-theoretic notions of fairness can incentivize strategic users when sharing hardware. 6,12,19,20 Markets and price theory can allocate and manage heterogeneous servers. 8,9,17 Demand response models can handle power emergencies.…”

Section: Resultsmentioning

confidence: 99%

Distributed strategies for computational sprints

2019

Self Cite

View full text Add to dashboard Cite

Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe a sprinting architecture in which many, independent chip multiprocessors share a power supply and sprints are constrained by the chips' thermal limits and the rack's power limits. Moreover, we present the computational sprinting game, a multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based on application phases and system conditions. The game produces an equilibrium that improves task throughput for data analytics workloads by 4-6× over prior greedy heuristics and performs within 90% of an upper bound on throughput from a globally optimized policy.

show abstract

Section: Resultsmentioning

confidence: 99%

Distributed strategies for computational sprints

2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…Figure B compares the evaluated speed‐up curve and the estimated Amdahl's Law for the theoretical speed‐up of the system. The Amdahl's Law is given by

S_{latency} false(s false) = \frac{1}{false(1 - p false) + \frac{p}{s}}

, where s is the number of processing cores that are running the task, p is the parallel portion of the task, and S latency is the theoretical speed‐up in latency. In order to estimate the parallel portion of the task, we fit the Amdahl's Law to the experimental curve.…”

Section: Catraca Evaluationmentioning

confidence: 99%

Toward a monitoring and threat detection system based on stream processing as a virtual network function for big data

Lopez

Mattos

Duarte

et al. 2019

Concurrency and Computation

View full text Add to dashboard Cite

The late detection of security threats causes a significant increase in the risk of irreparable damages and restricts any defense attempt. In this paper, we propose a sCAlable TRAffic Classifier and Analyzer (CATRACA). CATRACA works as an efficient online Intrusion Detection and Prevention System implemented as a Virtualized Network Function. CATRACA is based on Apache Spark, a Big Data Streaming processing system, and it is deployed over the Open Platform for Network Functions Virtualization (OPNFV), providing an accurate real-time threat-detection service. The system presents a friendly graphical interface that provides real-time visualization of the traffic and the attacks that occur in the network. Our prototype can differentiate normal traffic from denial of service (DoS) attacks and vulnerability probes over 95% accuracy under three different datasets. Moreover, CATRACA handles streaming data under concept drift detection with more than 85% of accuracy. KEYWORDSbig data, network traffic classification, stream processing, threat detection, virtual network function INTRODUCTIONThe Internet is facing constant changes, from the diversity of the user, the complexity of its application, until the heterogeneity of the information producers. 1 As a consequence, traffic monitoring, a critical task in maintaining the stability, reliability, and security of computer networks, is facing new challenges. 2 Current network monitoring tools are inadequate for current speed and management needs of large network domains.To ensure network security, new systems must be designed since current security systems such as Security Information and Event Management (SIEM) are inadequate. While 82% of security threats occur in minutes, an intrusion can take up to 8 months to be detected. 3 It is essential that the detection time is the least possible so that intrusion prevention can be effective. 4Security incidents have increased their complexity, and simple analysis and filtering of packets are no longer sufficient. Attackers try to hide malicious traffic from the security tools by forging the source IP and dynamically changing TCP port. In this context, a promising alternative for classifying network traffic and detect threats is to apply Machine Learning (ML) techniques. These techniques are suitable for big data, with more samples to train the classifier, as methods have higher effectiveness. 5 With a large number of features, however, ML techniques perform results with high latency due to computational resource consumption. This high latency is a disadvantage for applications that use machine learning for real-time classification. For example, network monitoring applications must analyze data and detect threats as quickly as possible. In this context, real-time stream processing allows the immediate analysis of different types of data and consequently benefits traffic monitoring for security threat detection. Open source distributed processing platforms such as Apache Storm, 6 Apache Flink, 7 and Apache Spark 8 process big data w...

show abstract