Managing energy and server resources in hosting centers

J. ChaseDarrell C. AndersonP. N. ThakarAmin VahdatRonald P. Doyle

article2001SOSP1,512 citations

Proposes an economic resource management architecture for hosting centers that dynamically allocates servers and routes traffic based on service performance bids, cutting server energy consumption by nearly 30% while honoring service-level agreements.

Listen

Shared Internet hosting centers face severe challenges from rapid traffic fluctuations, high electricity costs, and operational risks such as cooling failures or power disruptions. Because standard servers consume more than 60% of their peak power even when idle, traditional static overprovisioning for peak demand wastes significant capital and operating expenditures. Modern facilities require operating policies that can dynamically balance computing capacity, service quality, and energy consumption in real time.

The main objective of the article is to design, implement, and evaluate Muse, an adaptive resource management architecture that treats energy and computing resources within a hosting center as variable commodities managed through an economic framework.

The authors designed a hosting center operating system that combines reconfigurable network switches, continuous performance monitors, and a centralized policy controller termed the executive. The system applies a greedy optimization algorithm called Maximize Service Revenue and Profit alongside a smoothing filter called flop-flip to continuously adjust server allotments. To validate this approach, the authors built a physical testbed using FreeBSD servers running Apache and evaluated it under synthetic bursty Web workloads and a 43-million-request trace from IBM's commercial website.

The evaluation revealed several key findings. First, energy-conscious server management reduced energy consumption by 29% on the representative IBM Web workload, with simulated projections indicating savings up to 38% for enterprise loads and up to 78% for highly variable workloads such as the 1998 World Cup trace. Second, power measurements across multiple server architectures confirmed that idle servers draw substantial baseload power, making whole-server deactivation and power transitions far more effective than processor-only frequency adjustments. Third, during induced resource constraints and power "browndown" events, the economic allocation mechanism reliably prioritized high-value services and preserved revenue according to contract terms while degrading lower-value services gracefully.

These findings demonstrate that hosting centers can transition away from costly worst-case overprovisioning to a dynamic model that lowers operational expenses, reduces cooling infrastructure strain, and enables differentiated service level agreements based on performance and price trade-offs. Additionally, the ability to selectively scale down power consumption mitigates thermal risks and allows centers to operate safely on backup power during grid failures.

Based on these results, hosting facility operators should implement energy-aware load redirection and dynamic server power cycling for large server clusters. Organizations should also structure flexible service level agreements that incorporate utility- and penalty-based terms to monetize service quality tradeoffs. Before broad enterprise deployment, administrators should conduct pilot testing to determine appropriate load smoothing thresholds and establish warm low-power machine states that minimize wake-up delays.

The findings are supported by solid physical and trace-driven experiments, though readers should note certain limitations. The prototype primarily manages central processing unit resources rather than complex combinations of storage and memory bandwidth, and the architecture relies on a centralized controller that could become a single point of failure. Nevertheless, the evidence provides high confidence that whole-server energy management yields immediate and meaningful cost reductions in scaled hosting environments.

Duke University IBM computer hardware economics operating systems energy management resource management server clusters

Chase et al (2001).pdf

No sufficiently relevant recommendations were found.

Paper: Power provisioning for a warehouse-sized computer, Xiaobo Fan et al. (2007). This paper builds on dynamic hosting center energy management by analyzing aggregate power provisioning and cluster-wide capacity limits across warehouse-scale datacenter environments.
Paper: Live migration of virtual machines, Christopher J. Clark et al. (2005). This work advances dynamic cluster resource and server management by introducing live virtual machine migration to seamlessly rebalance workloads across physical hosts without service disruption.
Paper: Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center, Benjamin Hindman et al. (2011). Mesos extends dynamic cluster resource allocation concepts to multi-tenant datacenters by proposing a fine-grained two-level resource offer framework for diverse computing frameworks.
Paper: Dark silicon and the end of multicore scaling, Hadi Esmaeilzadeh et al. (2011). This article extends the focus on energy as a primary computing constraint by analyzing device-level power limitations and dark silicon bottlenecks in multicore architectures.

Cover for Managing energy and server resources in hosting centers

Abstract

Internet hosting centers serve multiple service sites from a common hardware base. This paper presents the design and implementation of an architecture for resource management in a hosting center operating system, with an emphasis on energy as a driving resource management issue for large server clusters. The goals are to provision server resources for co-hosted services in a way that automatically adapts to offered load, improve the energy efficiency of server clusters by dynamically resizing the active server set, and respond to power supply disruptions or thermal events by degrading service in accordance with negotiated Service Level Agreements (SLAs).

Our system is based on an economic approach to managing shared server resources, in which services “bid” for resources as a function of delivered performance. The system continuously monitors load and plans resource allotments by estimating the value of their effects on service performance. A greedy resource allocation algorithm adjusts resource prices to balance supply and demand, allocating resources to their most efficient use. A reconfigurable server switching infrastructure directs request traffic to the servers assigned to each service. Experimental results from a prototype confirm that the system adapts to offered load and resource availability, and can reduce server energy usage by 29% or more for a typical Web workload.

Topics

computer hardware economics operating systems service level agreements application servers web servers energy conservation

Keywords

energy management resource management server clusters power management energy consumption resource provisioning load balancing thermal management representative Web workloads

1. INTRODUCTION
2. MOTIVATION
3. OVERVIEW OF MUSE
3.1 Services and Servers
3.2 Redirecting Switches
3.3 Adaptive Resource Provisioning
3.4 Energy-Conscious Provisioning
4. THE RESOURCE ECONOMY
4.1 Bids and Penalties
4.2 MSRP Resource Allocation
4.3 Estimating Performance Effects
4.4 Feedback and Stability
4.5 Pricing
4.6 Multiple Resources
5. PROTOTYPE
5.1 Monitoring and Estimation
5.2 The Executive
5.3 The Request Redirector
6. EXPERIMENTAL RESULTS
6.1 Experimental Setup
6.2 Allocation Under Constraint
6.3 Browndown
6.4 Varying Load and Power
7. RELATED WORK
8. CONCLUSION
Acknowledgements
9. REFERENCES

Knowls

Knowl 1 — Maximize Service Revenue and Profit (MSRP) Resource Allocation Algorithm
algorithm
The Maximize Service Revenue and Profit (MSRP) algorithm is an incremental, greedy gradient-climbing allocation algorithm executed at periodic epochs (e.g., every 10 seconds) or upon trigger events (such as load shifts or capacity alterations). MSRP allocates discrete resource units (such as $1\%$ increments of a server CPU) among $N$ hosted services to maximize aggregate center profit while balancing resource supply and marginal value.

The algorithm relies on two monotonic helper functions for each service $i$ : $\text{grow}(i, \text{target})$ , which increments resource allotment $\mu_i$ until marginal value $\text{price}_i(\mu_i) \approx \text{target}$ or capacity is exhausted, and $\text{shrink}(i, \text{target})$ , which decrements $\mu_i$ until $\text{price}_i(\mu_i) \approx \text{target}$ . Assuming concave utility functions (monotonically non-increasing marginal utility), MSRP converges to a unique Pareto-optimal allocation in at worst $O(N + \mu N)$ time, where $\mu$ is the number of reallocated resource units and $N$ is the number of services.
Input: Set of hosted services $S = \{1, \dots, N\}$ , maximum capacity $\mu_{\max}$ , unit resource cost $cost(t)$ , current allocation vector $\{\mu_i\}$
Output: Updated allocation vector $\{\mu_i\}$
Phase 1: Reclaim negative-return resources
for each $i \in S$ do
    if $price_i(\mu_i) < cost(t)$ then
         $\mu_i \leftarrow \text{shrink}(i, cost(t))$
    end if
end for
Phase 2: Allot resources to profitable services
$\mu \leftarrow \sum_{k \in S} \mu_k$
while $\mu < \mu_{\max}$ and $\max_{k \in S} price_k(\mu_k) \ge cost(t)$ do
     $i \leftarrow \arg\max_{k \in S} price_k(\mu_k)$
     $j \leftarrow \arg\max_{k \in S \setminus \{i\}} price_k(\mu_k)$
    target $\leftarrow \max(price_j(\mu_j), cost(t))$
     $\mu_i \leftarrow \text{grow}(i, \text{target})$
     $\mu \leftarrow \sum_{k \in S} \mu_k$
end while
Phase 3: Reclaim overcommitted resources (e.g., following browndown)
$\mu \leftarrow \sum_{k \in S} \mu_k$
while $\mu > \mu_{\max}$ do
     $i \leftarrow \arg\min_{k \in S : \mu_k > 0} price_k(\mu_k)$
     $j \leftarrow \arg\min_{k \in S \setminus \{i\} : \mu_k > 0} price_k(\mu_k)$
     $\mu_i \leftarrow \text{shrink}(i, price_j(\mu_j))$
     $\mu \leftarrow \sum_{k \in S} \mu_k$
end while
Phase 4: Equilibrate marginal prices across services
while $\exists i, j \in S$ such that $price_i(\mu_i) < price_j(\mu_j)$ do
    target $\leftarrow \frac{price_i(\mu_i) + price_j(\mu_j)}{2}$
     $\mu_i \leftarrow \text{shrink}(i, \text{target})$
     $\mu_j \leftarrow \text{grow}(j, \text{target})$
end while
return $\{\mu_i\}$
Knowl 2 — Economic Profit Maximization Formulation for Shared Hosting
equation
In a shared hosting facility, the resource manager determines discrete resource allotments $\mu_i$ across $N$ co-hosted services to maximize aggregate hosting profit per unit time at time $t$ :

$\text{profit}(t) = \sum_{i=1}^{N} \Big( U_i(t, \mu_i) - \mu_i \cdot \text{cost}(t) \Big)$

subject to the aggregate capacity constraint:

$\sum_{i=1}^{N} \mu_i \le \mu_{\max}$

where:
- $\mu_i \ge 0$ is the discrete number of resource units (e.g., $1\%$ CPU slices) allocated to service $i$ .
- $\mu_{\max}$ is the total discrete capacity available in the center at time $t$ , which may decrease during failures or power browndown events.
- $\text{cost}(t)$ is the variable cost per resource unit per unit time (reflecting energy prices and hardware wear), incurred only when the resource unit is actively allocated.
- $U_i(t, \mu_i)$ is the utility function for customer $i$ , defined as:
$U_i(t, \mu_i) = \text{bid}_i\big(\lambda_i(t, \mu_i)\big) - \text{penalty}_i(t, \mu_i)$

where $\lambda_i(t, \mu_i)$ is the delivered request throughput of service $i$ given allotment $\mu_i$ under current offered load, $\text{bid}_i$ represents revenue generated per unit throughput (e.g., in dollars per hits per minute), and $\text{penalty}_i$ represents monetary penalties defined by Service Level Agreements (SLAs) for underprovisioning (e.g., when utilization exceeds a target threshold while allotment is below a contracted reservation $r_i$ ).

The marginal price offered by customer $i$ for its last allocated resource unit is:

$\text{price}_i(\mu_i) = U_i(t, \mu_i + 1) - U_i(t, \mu_i) \approx \frac{\partial U_i(t, \mu_i)}{\partial \mu_i}$

Assuming each $U_i$ is concave (the utility gradient is positive and monotonically non-increasing), the resource optimization problem has a unique global maximum with no local maxima.
Knowl 3 — Muse Hosting Center Operating System Architecture
model/method
Muse is an operating system architecture for hosting centers that dynamically provisions server and energy resources across co-hosted services. The architecture consists of four interrelated components:
1. Generic Server Appliances: A pool of interchangeable, stateless server nodes running host operating systems equipped with kernel resource principals (e.g., Resource Containers) that enforce performance-isolated proportional CPU allocations.
2. Reconfigurable Network Switching Fabric: Host-based or hardware Layer-4 redirecting switches that intercept client TCP connections to virtual service endpoints, modify packets via Network Address Translation (NAT) and incremental checksum updates, and distribute requests across an active server set registered for each service.
3. Load Monitoring and Estimation Modules: Kernel monitoring extensions on nodes and switches that track per-service CPU utilization $\rho_i$ , request arrival and completion rates $\lambda_i$ (measured in TCP accept and FIN-ACK rates), and TCP accept queue lengths $q_i$ .
4. The Executive: A centralized policy controller that periodically evaluates monitoring data, executes an economic allocation algorithm, reconfigures switch active sets, modifies node resource container allocations via remote commands, and modulates server power states using Advanced Power Management (APM) and Wake-on-LAN to turn idle machines off and recruit standby machines as load varies.
Knowl 4 — Performance and Resource Demand Estimation via Target Utilization
model/method
To predict the throughput $\lambda_i(t, \mu_i)$ and utilization $\rho_i$ resulting from changing resource allotment $\mu_i$ to service $i$ , Muse continuously evaluates smoothed runtime metrics against a target utilization threshold $\rho_{\text{target}} \in (0, 1)$ (typically set between $0.5$ and $0.8$ ):
- Resource Reclaim (Overprovisioned State, $\rho_i < \rho_{\text{target}}$ ): Because utilization varies linearly with throughput below saturation, the executive can reclaim $\mu_i(\rho_{\text{target}} - \rho_i)$ resource units from service $i$ without reducing request throughput $\lambda_i$ or revenue.
- Resource Addition (Saturated State, $\rho_i > \rho_{\text{target}}$ ): Mean per-request service demand is estimated empirically as $d_i = \frac{\rho_i \mu_i}{\lambda_i}$ . The system predicts that allocating additional resource units will increase throughput linearly with rate $1/d_i$ until utilization decreases to $\rho_{\text{target}}$ . In the absence of SLA penalties, the marginal price per resource unit is estimated as:
$\text{price}_i(\mu_i) = \frac{\Delta \text{bid}_i / \Delta \lambda_i}{d_i} = \frac{\Delta \text{bid}_i / \Delta \lambda_i}{\rho_i \mu_i / \lambda_i}$

If the system overshoots past the saturation knee, $\rho_i$ drops below $\rho_{\text{target}}$ in the subsequent epoch, causing the excess allotment to be reclaimed.
- I/O Bottleneck Detection: When $\rho_i < \rho_{\text{target}}$ but smoothed TCP accept queue length $q_i$ exceeds an overload threshold, the monitor infers that service $i$ is I/O-bound rather than CPU-bound. The system responds by dynamically lowering $\rho_{\text{target}}$ for service $i$ to allot a larger node share and avoid I/O contention, gradually restoring $\rho_{\text{target}}$ to its default as queue lengths subside.
Knowl 5 — Flop-Flip Signal Smoothing Filter
algorithm
The flop-flip filter is an adaptive signal smoothing algorithm designed for noisy, bursty performance signals (e.g., server CPU utilization and request throughput) in feedback-controlled resource management systems. Unlike standard Exponentially Weighted Moving Average (EWMA) or agile dual-gain filters that continuously fluctuate, the flop-flip filter maintains a flat, stable estimate during transient oscillations and steps to a new level only when a persistent load shift occurs.
Input: New raw observation $O_t$ , sliding window duration $W$ (e.g., 30 s), step threshold $\delta$ (e.g., 1 standard deviation $\sigma_t$ of window observations)
Output: Smoothed estimate $E_t$
Add $O_t$ to window observation history $W_t$
Purge observations older than $t - W$ from $W_t$
$\overline{O}_t \leftarrow \frac{1}{|W_t|} \sum_{\tau \in W_t} O_\tau$
if $t = 0$ then
     $E_t \leftarrow O_t$
else if $|E_{t-1} - \overline{O}_t| \le \delta$ then
     $E_t \leftarrow E_{t-1}$
else
     $E_t \leftarrow \overline{O}_t$
end if
return $E_t$

Knowl 6 — Server Power Draw and Idle Power Inefficiency

data/table

Power draw measurements across various commodity server hardware platforms and operating systems show that idle servers consume more than $60\%$ of their peak power consumption, even when the operating system executes the CPU idle halt loop between interrupts. This fixed power overhead is caused by internal power supply transformers maintaining charged capacity. While active power consumption scales roughly linearly from idle baseline to maximum load with CPU utilization, reducing active server count and transitioning surplus nodes into low-power hibernation states ( $2.5\text{--}5.5\text{ W}$ ) eliminates the idle power supply penalty.

Architecture	Machine	Disks	Operating System	Power Draw (Watts)
				Boot	Max	Idle	Hibernate
PIII 866MHz	SuperMicro 370-DER	1	FreeBSD 4.0	136	120	93	—
PIII 866MHz	SuperMicro 370-DER	1	Windows 2000	134	120	98	5.5
PIII 450MHz	ASUS P2BLS	1	FreeBSD 4.0	66	55	35	4
PIII Xeon 733MHz	PowerEdge 4400	8	FreeBSD 4.0	278	270	225	—
PIII 500MHz	PowerEdge 2400	3	FreeBSD 4.0	130	128	95	2.5
PIII 500MHz	PowerEdge 2400	3	Windows 2000	127	120	98	2.5
PIII 500MHz	PowerEdge 2400	3	Solaris 2.7	129	124	127	2.5

In active systems, the CPU is the dominant variable power consumer (drawing up to 38 W for a 600 MHz Intel Pentium-III), while memory and networking draw negligible power and disks consume between 50 and 250 W per terabyte.

Knowl 7 — Browndown in Hosting Centers
definition
Browndown is a managed partial failure operating state in a data center or hosting cluster in which the aggregate available computing capacity $\mu_{\max}$ is intentionally and dynamically reduced to restrict electrical power consumption and heat dissipation. Browndown is initiated in response to thermal emergencies (such as cooling infrastructure failures or extreme ambient temperatures), electrical supply disruptions requiring prolonged operation on limited backup batteries or generators, or power utility constraints. Under browndown, the hosting operating system adaptively redistributes reduced capacity across hosted services according to Service Level Agreements and utility functions, gracefully degrading response times or request throughput rather than risking uncoordinated thermal shutdown or physical hardware damage.
Knowl 8 — Energy-Conscious Provisioning Savings under Web Workloads
empirical result
Muse was evaluated on a cluster of five 450 MHz Pentium-III servers and a dedicated Layer-4 redirector running a 10-hour trace replay of the full-week February 2001 IBM web server log (43 million requests) accelerated at 16x speedup:
- Baseline Static Provisioning: With all five servers continuously powered, total server power draw fluctuated between 190 W and 240 W proportionally to request load, consuming a total of $2.38\text{ KWh}$ over the 10-hour run.
- Muse Energy-Conscious Provisioning: With target utilization $\rho_{\text{target}} = 0.5$ , Muse adaptively activated and hibernated servers to match traffic volume, consuming $1.69\text{ KWh}$ —a $29\%$ net energy reduction relative to static provisioning.
- Cluster Scale Projections: In larger clusters where server capacity can be dynamically provisioned in smaller relative increments (reducing server quantization error), projected server energy savings reach up to $38\%$ on the IBM trace and up to $78\%$ on the 1998 World Cup trace.
Knowl 9 — Competitive Resource Allocation and Utility-Based Browndown Degradation
empirical result
In experimental evaluations on a 3-server cluster hosting two Web services ( $s_0$ receiving steady load and $s_1$ subjected to periodic surges peaking at $2\times$ the load of $s_0$ , with allocation granularity of $1\%$ per server and $\mu_{\max} = 300$ ):
- Prioritized Rationing Under Contention: When total demand exceeds 3 servers, giving $s_1$ a higher per-hit bid causes the executive to throttle $s_0$ 's allotment from 1.0 to 0.4 servers during peaks, granting $s_1$ 2.6 servers. When the bid priority is reversed, $s_0$ retains its full 1.0 server allotment while $s_1$ receives the remaining 2.0 servers.
- Browndown Resource Shifting: When cluster capacity is dropped from 3 servers to 2 servers ( $\mu_{\max} = 200$ ) during a browndown event, Muse initially degrades $s_1$ . However, when $s_1$ 's load surges while heavily saturated, the executive reallocates capacity away from $s_0$ to $s_1$ despite $s_0$ having a higher per-request bid. Because $s_1$ is heavily backlogged, each added resource unit satisfies a larger absolute number of completed requests on $s_1$ , yielding higher aggregate system utility.
Knowl 10 — Limitations of the Muse Prototype and Economic Framework
limitation
The Muse hosting architecture and economic resource allocation framework have four primary stated limitations:
1. Absence of Dynamic Microeconomic Competition: Customer utility functions are static and sealed; clients do not observe real-time congestion price signals or dynamically adjust bids to compete for resources or switch providers.
2. Single-Resource Management: The prototype manages only CPU resource allocations via Resource Containers. Memory, disk bandwidth, and network bandwidth are not explicitly co-scheduled as complementary goods, relying instead on an indirect heuristic that lowers CPU target utilization $\rho_{\text{target}}$ when I/O request queues build up.
3. Server Power State Transition Lag: Cold-booting and waking an off-power server via Wake-on-LAN requires approximately one minute, necessitating damping of power transitions to match slow, diurnal load curves rather than reacting to short-term, transient traffic spikes.
4. Centralized Executive Bottleneck: The policy executive operates as a centralized user-level daemon, representing a potential scalability bottleneck and single point of failure (although servers and switches continue running autonomously if the executive crashes).

Coverage note — No substantial contributed material was omitted. All key architectural mechanisms, mathematical formulations, algorithms (MSRP, Flop-Flip), empirical energy savings, competitive allocation dynamics, power profiles, and limitations are represented.

References

1.Tarek F. Abdelzaher and Chenyang Lu. Modeling and Performance Control of Internet Servers. In 39th IEEE Conference on Decision and Control, December 2000.
2.Tarek F. Abdelzaher, Kang G. Shin, and Nina Bhatti. Performance Guarantees for Web Server End-Systems: A Control-Theoretical Approach. IEEE Transactions on Parallel and Distributed Systems, June 2001.
3.Jeffrey Aman, Catherine K. Eilert, David Emmes, Peter Yocom, and Donna Dillenberger. Adaptive Algorithms for Managing a Distributed Data Processing Workload. IBM Systems Journal, 36(2), 1997.
4.Darrell C. Anderson, Jeffrey S. Chase, and Amin M. Vahdat. Interposed Request Routing for Scalable Network Storage. In Proceedings of the Fourth Symposium on Operating System Design and Implementation (OSDI), October 2000.
5.Karen Appleby, Sameh Fakhouri, Liana Fong, Germán Goldszmidt, Michael Kalantar, Srirama Krishnakumar, Donald Pazel, John Pershing, and Benny Rochwerger. Oceano - SLA Based Management of a Computing Utility. In Proceedings of the 7th IFIP/IEEE International Symposium on Integrated Network Management, May 2001.
6.Martin Arlitt and Tai Jin. Workload Characterization of the 1998 World Cup Web Site. Technical Report HPL-1999-35R1, HP Laboratories, September 1999. The trace is available from the Internet Traffic Archive at ita.ee.lbl.gov.
7.Armando Fox and Steven D. Gribble and Yatin Chawathe and Eric A. Brewer and Paul Gauthier. Cluster-based scalable network services. In Proceedings of the Sixteenth ACM Symposium on Operating System Principles (SOSP), pages 78-91, October 1997.
8.Mohit Aron. Differentiated and Predictable Quality of Service in Web Server Systems. PhD thesis, Department of Computer Science, Rice University, October 2000.
9.Mohit Aron, Peter Druschel, and Willy Zwaenepoel. Cluster Reserves: A Mechanism for Resource Management in Cluster-based Network Servers. In Proceedings of the Joint International Conference on Measurement and Modeling of Computer Systems (ACM SIGMETRICS 2000), pages 90-101, June 2000.
10.Gaurav Banga, Peter Druschel, and Jeffrey C. Mogul. Resource Containers: A New Facility for Resource Management in Server Systems. In Proceedings of the Third Symposium on Operating Systems Design and Implementation (OSDI), February 1999.
11.Paul Barford and Mark E. Crovella. Generating Representative Web Workloads for Network and Server Performance Evaluation. In Proceedings of the ACM Conference on Measurement and Modeling of Computer Systems (SIGMETRICS '98), pages 151-160, June 1998.
12.Pat Bohrer, Elmootazbellah N. Elnozahy, Tom Keller, Michael Kistler, Charles Lefurgy, and Ray Rajamony. The Case for Power Management in Web Servers. In Power-Aware Computing (Robert Graybill and Rami Melhem, editors). Kluwer/Plenum series in Computer Science, to appear, January 2002.
13.David Brooks and Margaret Martonosi. Dynamic Thermal Management for High-Performance Microprocessors. In Proceedings of the Seventh International Symposium on High-Performance Computer Architecture (HPCA-7), January 2001.
14.Jeffrey S. Chase, Henry M. Levy, Michael J. Feeley, and Edward D. Lazowska. Sharing and Protection in a Single Address Space Operating System. ACM Transactions on Computer Systems, 12(4), November 1994.
15.Fred Douglis, P. Krishnan, and Brian Bershad. Adaptive Disk Spin-down Policies for Mobile Computers. In 2nd USENIX Symposium on Mobile and Location-Independent Computing, April 1995. Monterey CA.
16.Donald F. Ferguson, Christos Nikolaou, Jakka Sairamesh, and Yechiam Yemini. Economic Models for Allocating Resources in Computer Systems. In Market-Based Control: A Paradigm for Distributed Resource Allocation (Scott H. Clearwater, editor). World Scientific, 1996.
17.Jason Flinn and Mahadev Satyanarayanan. Energy-aware Adaptation for Mobile Applications. In Proceedings of the Seventeenth Symposium on Operating Systems Principles (SOSP), pages 48-63, December 1999.
18.Syam Gadde. The Proxycizer Web Proxy Tool Suite. http://www.cs.duke.edu/ari/Proxycizer/.
19.Toshihide Ibaraki and Naoki Katoh, editors. Resource Allocation Problems: Algorithmic Approaches. MIT Press, Cambridge, MA, 1988.
20.Arun Iyengar, Jim Challenger, Daniel Dias, and Paul Dantzig. High-performance Web Site Design Techniques. IEEE Internet Computing, 4(2):17-26, March 2000.
21.Van Jacobson. Congestion Avoidance and Control. ACM Computer Communication Review: Proceedings of the SIGCOMM Symposium, 18(4):314-329, August 1988.
22.Sugih Jamin, Peter B. Danzig, Scott J. Shenker, and Lixia Zhang. A Measurement-based Admission Control Algorithm for Integrated Services Packet Networks. IEEE/ACM Transactions on Networking, 5(1):56-70, February 1997.
23.Sugih Jamin, Scott J. Shenker, and Peter B. Danzig. Comparison of Measurement-based Admission Control Algorithms for Controlled-Load Service. In Proceedings of IEEE Infocom 1997, April 1997.
24.Michael B. Jones, Daniela Rosu, and Marcel-Catalin Rosu. CPU Reservations and Time Constraints: Efficient, Predictable Scheduling of Independent Activities. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, pages 198-211, October 1997.
25.Minkyong Kim and Brian Noble. Mobile Network Estimation. In Proceedings of the Seventh Annual Conference on Mobile Computing and Networking, July 2001.
26.Alvin R. Lebeck, Xiaobo Fan, Heng Zeng, and Carla S. Ellis. Power-Aware Page Allocation. In Proceedings of the Ninth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS IX), 2000.
27.Kelvin Li and Sugih Jamin. A Measurement-Based Admission-Controlled Web Server. In Proceedings of IEEE Infocom 2000, March 2000.
28.Lee W. McKnight and Joseph P. Bailey, editors. Internet Economics. MIT Press, Cambridge, MA, 1997.
29.Jennifer D. Mitchell-Jackson. Energy Needs in an Internet Economy: A Closer Look at Data Centers. Master's thesis, Energy and Resources Group, University of California at Berkeley, July 2001.
30.Rolf Neugebauer and Derek McAuley. Energy is Just Another Resource: Energy Accounting and Energy Pricing in the Nemesis OS. In Proceedings of the Eighth IEEE Workshop on Hot Topics in Operating Systems HotOS-VIII, pages 59-64, May 2001.
31.Brian D. Noble, M. Satyanarayanan, Dushyanth Narayanan, James Eric Tilton, Jason Flinn, and Kevin R. Walker. Agile Application-Aware Adaptation for Mobility. In Proceedings of the Sixteenth ACM Symposium on Operating Systems Principles (SOSP), pages 276-287, Saint Malo, France, October 1997.
32.Vivek S. Pai, Mohit Aron, Gaurav Banga, Michael Svendsen, Peter Druschel, Willy Zwaenepoel, and Erich Nahum. Locality-Aware Request Distribution in Cluster-based Network Servers. In Proceedings of the Eighth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS VIII), October 1998.
33.Eduardo Pinheiro, Ricardo Bianchini, Enrique V. Carrera, and Taliver Heath. Load Balancing and Unbalancing for Power and Performance in Cluster-Based Systems. Technical Report DCS-TR-440, Department of Computer Science, Rutgers University, May 2001.
34.John Reumann, Ashish Mehra, Kang G. Shin, and Dilip Kandlur. Virtual Services: A New Abstraction for Server Consolidation. In Proceedings of the USENIX 2000 Technical Conference, June 2000.
35.Timothy Roscoe and Prashant Shenoy. New Resource Control Issues in Shared Clusters. In Proceedings of the Eight International Workshop on Interactive Distributed Multimedia Systems (IDMS'01), September 2001.
36.Yasushi Saito, Brian N. Bershad, and Henry M. Levy. Manageability, Availability and Performance in Porcupine: A Highly Scalable Cluster-Based Mail Service. In Proceedings of the 17th ACM Symposium on Operating Systems Principles (SOSP), pages 1-15, Kiawah Island, December 1999.
37.Barry C. Smith, John F. Leimkuhler, and Ross M. Darrow. Yield Management at American Airlines. Interfaces, 22(1), January 1992.
38.David C. Steere, Ashvin Goel, Joshua Gruenberg, Dylan McNamee, Calton Pu, and Jonathan Walpole. A Feedback-driven Proportion Allocator for Real-Rate Scheduling. In Proceedings of the Third Symposium on Operating Systems Design and Implementation (OSDI), pages 145-158, February 1999.
39.Michael Stonebraker, Paul M. Aoki, Witold Litwin, Avi Pfeffer, Adam Sah, Jeff Sidell, Carl Staelin, and Andrew Yu. Mariposa: A Wide-Area Distributed Database System. VLDB Journal: Very Large Data Bases, 5(1):48-63, 1996.
40.David G. Sullivan and Margo I. Seltzer. Isolation with Flexibility: A Resource Management Framework for Central Servers. In Proceedings of the 2000 USENIX Annual Technical Conference, pages 337-350, June 2000.
41.Kevin Thompson, Gregory J. Miller, and Rick Wilder. Wide-Area Internet Traffic Patterns and Characteristics. In IEEE Network, November 1997.
42.Amin Vahdat, Thomas Anderson, Michael Dahlin, Eshwar Belani, David Culler, Paul Eastham, and Chad Yoshikawa. WebOS: Operating System Services for Wide-Area Applications. In Proceedings of the Seventh IEEE Symposium on High Performance Distributed Computing (HPDC), Chicago, Illinois, July 1998.
43.Amin Vahdat, Alvin R. Lebeck, and Carla S. Ellis. Every Joule is Precious: The Case for Revisiting Operating System Design for Energy Efficiency. In Proceedings of the 9th ACM SIGOPS European Workshop, September 2000.
44.Ben Verghese, Anoop Gupta, and Mendel Rosenblum. Performance Isolation: Sharing and Isolation in Shared Memory Multiprocessors. In Proceedings of the 8th International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), October 1998.
45.Carl A. Waldspurger, Tad Hogg, Bernardo A. Huberman, Jeffrey O. Kephart, and W. Scott Stornetta. Spawn: A Distributed Computational Economy. IEEE Transactions on Software Engineering, 18(2):103-117, February 1992.
46.Carl A. Waldspurger and William E. Weihl. Lottery Scheduling: Flexible Proportional-Share Resource Management. In Proceedings of the First Symposium on Operating Systems Design and Implementation (OSDI), pages 1-11, November 1994.
47.Huican Zhu, Hong Tang, and Tao Yang. Demand-driven Service Differentiation in Cluster-Based Network Servers. In Proceedings of IEEE Infocom 2001, April 2001.

Citation

MLA

Chase, J. S., et al. “Managing Energy and Server Resources in Hosting Centers”. Proceedings of the Eighteenth ACM Symposium on Operating Systems Principles, 2001, pp. 103–16, https://doi.org/10.1145/502034.502045.

APA

Chase, J. S., Anderson, D. C., Thakar, P. N., Vahdat, A. M., & Doyle, R. P. (2001). Managing energy and server resources in hosting centers. Proceedings of the Eighteenth ACM Symposium on Operating Systems Principles, 103–116. https://doi.org/10.1145/502034.502045

Chicago

Chase, J. S., D. C. Anderson, P. N. Thakar, A. M. Vahdat, and R. P. Doyle. 2001. “Managing Energy and Server Resources in Hosting Centers”. Proceedings of the Eighteenth ACM Symposium on Operating Systems Principles, 103–16. https://doi.org/10.1145/502034.502045.

Harvard

Chase, J.S. et al. (2001) “Managing energy and server resources in hosting centers”, Proceedings of the eighteenth ACM symposium on Operating systems principles. ACM, pp. 103–116. Available at: https://doi.org/10.1145/502034.502045.

Vancouver

1. Chase JS, Anderson DC, Thakar PN, Vahdat AM, Doyle RP (2001) Managing energy and server resources in hosting centers. In: Proceedings of the eighteenth ACM symposium on Operating systems principles. ACM, pp 103–116

BibTeX

@inproceedings{Chase_2001, series={SOSP01}, title={Managing energy and server resources in hosting centers}, url={http://dx.doi.org/10.1145/502034.502045}, DOI={10.1145/502034.502045}, booktitle={Proceedings of the eighteenth ACM symposium on Operating systems principles}, publisher={ACM}, author={Chase, Jeffrey S. and Anderson, Darrell C. and Thakar, Prachi N. and Vahdat, Amin M. and Doyle, Ronald P.}, year={2001}, month=Oct, pages={103–116}, collection={SOSP01} }

Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Managing energy and server resources in hosting centers

Abstract

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Maximize Service Revenue and Profit (MSRP) Resource Allocation Algorithm

Knowl 2 — Economic Profit Maximization Formulation for Shared Hosting

Knowl 3 — Muse Hosting Center Operating System Architecture

Knowl 4 — Performance and Resource Demand Estimation via Target Utilization

Knowl 5 — Flop-Flip Signal Smoothing Filter

Knowl 6 — Server Power Draw and Idle Power Inefficiency

Knowl 7 — Browndown in Hosting Centers

Knowl 8 — Energy-Conscious Provisioning Savings under Web Workloads

Knowl 9 — Competitive Resource Allocation and Utility-Based Browndown Degradation

Knowl 10 — Limitations of the Muse Prototype and Economic Framework

References

Citation

BibTeX

Access the Paper