Power management of online data-intensive services

By David Meisner, Christopher M. Sadler, Luiz André Barroso, Wolf-Dietrich Weber, Thomas F. Wenisch. International Symposium on Computer Architecture, 2011

434 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/d5a807e1bd2b5ece92089af95085d3b720602ee2

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

Quasar: resource-efficient and QoS-aware cluster management (2014, background): “Underutilized servers contribute to capital expenses and, since they are not energy proportional [36, 42], to operational expenses as well.”
Ensuring Safety, Security, and Sustainability of Mission-Critical Cyber–Physical Systems (2012, background): “, transitioning to the sleep sate in idle times [53], [54], and active low-power modes, i.”
Paragon: QoS-aware scheduling for heterogeneous datacenters (2013, background): “Additionally, Paragon limits degradation to less than 10% for an additional 33% of applications and maintains performance degradation moderate (no cliffs in performance such as for NH in applications [1-1000]).”
Seer: Leveraging Big Data to Navigate the Complexity of Performance Debugging in Cloud Microservices (2019, background): “Specifically, even though the quality-of-service (QoS) requirements of the end-to-end application are similar for mi-croservices and monoliths, the tail latency required for each individual microservice is much stricter than for traditional cloud applications [34, 43–45, 61, 62, 64, 67, 70, 77].”
Interactive Analytical Processing in Big Data Systems: A Cross-Industry Study of MapReduce Workloads (2012, background): “A cross-workload comparison is thus far absent, and use cases beyond the technology industry have not been described.”
Shenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads (2019, methodology): “We assume a Poisson arrival process (empirically shown to be representative of Google’s datacenters [53]), exponentially distributed service times with a mean of 10 μs, and a latency limit of 100 μs at the 99.”
Dynamo: Facebook's Data Center-Wide Power Management System (2016, background): “There is a large body of previous work on server-level or local ensemble-level power management [4], [8], [9], [10], [26], [27], [28], [29], [30], [31], [32], [33].”
CPI2: CPU performance isolation for shared compute clusters (2013, background): “A typical web-search query involves thousands of machines working in parallel [6, 19, 29, 25], each one contributing some portion of the final result.”
Profiling a warehouse-scale computer (2015, background): “While WSCs were initially created with a “killer application” in mind [5], the model of “the datacenter is the computer” has since grown and current datacenters handle a rapidly increasing pool of applications.”
An Open-Source Benchmark Suite for Microservices and Their Hardware-Software Implications for Cloud & Edge Systems (2019, background): “Many of these applications are interactive , latency-critical services that must meet strict performance (through-put and tail latency), and availability constraints, while also handling frequent software updates [21, 28–34, 36, 44, 51, 61, 62, 65].”
Want to scale in centralized systems? Think P2P (2015, neutral): “However, accounting for dynamics requires periodic recomputations which turn out to be very costly [36, 38, 39].”
Data Center Energy Consumption Modeling: A Survey (2016, background): “Dataintensive applications can be categorized as online dataintensive [333] and offline data-intensive applications based on their type of operation.”
Heracles: Improving resource efficiency at scale (2015, background): “LC tasks operate with strict service level objectives (SLOs) on tail latency, and even small amounts of interference can cause significant SLO violations [51, 54, 39].”
AutoScale: Dynamic, Robust Capacity Management for Multi-Tier Data Centers (2012, methodology): “Our choice of SLA is motivated by recent studies [DeCandia et al. 2007; Krioukov et al. 2010; Meisner et al. 2011; Urgaonkar and Chandra 2005] that indicate that 95 percentile guarantees of hundreds of milliseconds are typical.”
Tradeoffs between power management and tail latency in warehouse-scale applications (2014, background): “Figure 1 also demonstrates the varying user demand in a Google production cluster in North America running content ad matching.”