Main Page

From IDGF-SP Financial and energetic analysis of volunteer computing and desktop grids
Revision as of 19:39, 23 March 2015 by Hanappe (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

This wiki was created to gather information about the cost and energy consumption of volunteer computing networks and desktop grids (DG).


Contents

Evaluating the performance of Desktop Grids

Benchmarks

Different benchmarks are used to compare systems. Here are the most used ones:

Name Used by
Drystone/Whetstone BOINC
(High-Performance) LINPACK (HPL) Top500, Green500
SPEC2006 / CFP2006 Many hardware vendors (see the results)

Some rough comparison might be possible between Whetstone and CFP2006, because they both test the floating-point performance ofsingle machines. The performance of the network has a big impact, however, on the LINPACK benchmark that is used to classify the computing clusters in the Top500 list.

This web page estimates the equivalence between Whetstone and CFP2006 as: CFP2006 = 0.001085 * Whetstone. However, a quick test on my machine (Whetstone measured by BOINC: 2552, estimated CFP2006: 30) shows that this relation does not hold.


Debian provides a package to run the HPC Challenge benchmark, which includes LINPACK. Installing and running it is easy:

 apt-get install hpcc
 mpirun -np 2 hpcc


Evaluating the energy consumption of Desktop Grids

Facts and figures

Models

  • Sony CSL: comparison of cost and energy consumption between volunteer computing networks and computing clusters: Sony CSL's spreadsheet

Software

References

  • Bernhard Schott, Ad Emmen: "Green Desktop-Grids: Scientific Impact, Carbon Footprint, Power Usage Efficiency." Scalable Computing: Practice and Experience 01/2011, vol. 12. (File:718-352-1-PB.pdf)
  • Brown D. and Reams C.: Toward Energy-efficient Computing, ACM Queue, Vol. 8, No. 2, February 2010 Online, PDF: File:P30-brown.pdf
  • Profiling energy consumption of VMs for green cloud computing Qingwen Chen, Paola Grosso, Karel van der Veldt, Cees de Laat, Rutger Hofman and Henri Bal. In: International Conference on Cloud and Green Computing (CGC2011), Sydney December 2011. PDF: File:CGC2011-submission.pdf


Solutions to reduce the energy consumption of Desktop Grids

The Low-Energy BOINC project

References

  • Bernhard Schott, Ad Emmen: "Green Desktop-Grids: Scientific Impact, Carbon Footprint, Power Usage Efficiency." Scalable Computing: Practice and Experience 01/2011, vol. 12. (File:718-352-1-PB.pdf)
  • Brown D. and Reams C.: Toward Energy-efficient Computing, ACM Queue, Vol. 8, No. 2, February 2010 Online, PDF: File:P30-brown.pdf
  • Hanappe, P. Fine-grained CPU Throttling to Reduce the Energy Footprint of Volunteer Computing. Sony Computer Science Laboratory Paris, January 2012. File:Hanappe-12a.pdf


Financial evaluation of Desktop Grids

Facts and figures

In this section we gather concrete facts and figures from institutions that have set up DG, clusters, or cloud computing solutions.

Models

This section lists the various models to estimate the energy consumption and the costs of DG, clusters, cloud computing, etc.

  • e-Fiscal
  • Bernhard Schott: draft for DG evaluation for different (older) desktop and server machines for different ambient climate types: Desktop Grid Calc 101
  • Sony CSL: comparison of cost and energy consumption between volunteer computing networks and computing clusters: Sony CSL's spreadsheet


References

  • David Anderson, "Volunteer Computing: the Ultimate Cloud". Crossroads - Plugging Into the Cloud Crossroads Homepage archive, Volume 16 Issue 3, March 2010, Pages 7-10, ACM New York, NY, USA. doi>10.1145/1734160.1734164 (slides: File:Anderson BOINC.pdf)
  • Kondo, D., e.a., " Cost-Benefit Analysis of Cloud Computing versus Desktop Grids". Computing (2009). File:Kondo hcw09.pdf



Summary

Type Model or data Year Benchmark Number of CPUs Performance Power consumption (MW) Yearly energy consumption (GWh/year) Yearly cost (million Euro/year) Performance/Watt Performance/Euro Notes
Desktop grid Sony CSL 2011 CFP2006 2200 to 64000 600000 0.3 to 1.3 2.9 to 12.2 0.19 to 0.48 0.46 to 2 CFP2006/Watt 1.25 to 3 CFP2006/Euro 1
Data center Sony CSL 2011 CFP2006 1100 to 2600 600000 0.3 to 0.5 2.9 to 4.4 1.2 to 3.4 1.2 to 2 CFP2006/Watt 0.17 to 0.5 CFP2006/Euro
Data center James Hamilton 2010 CFP2006 46000 2254000 11.6 101 32 0.19 CFP2006/Watt 0.07 CFP2006/Euro 2, 3
Cloud (AWS) File:Anderson BOINC.pdf 2010 ? 5000 10 TeraFLOPS 3 4
Data center File:Anderson BOINC.pdf 2010 ? 10 TeraFLOPS 1.14
Volunteer computing File:Anderson BOINC.pdf 2010 ? 10 TeraFLOPS 0.1
Data center Green500 2013 LINPACK 128 CPUs + 128 GPUs 3208.83 MFLOPS/W Green500 #1
Data center Green500 2013 LINPACK 42.33 MFLOPS/W Green500 #500
Desktop grid RaspberryPi 2013 Whetstone 1 270 2x10-6 (2 W) 17.5x10-6 (17.5kWh) 12.3 Euro total 135 Whetstones/W 22 Whetstones/Euro 5
Desktop grid Vaio VPC Z2 Intel Core i5-2410M (see) 2013 Whetstone 2 4270 48x10-6 (2 W) 42x10-5 (420 kWh) 333 Euro total 89 Whetstones/W 7.66 Whetstones/W 6, 7

Notes:

  1. The volunteers contribute between between 35 and 245 Euro/year/volunteer in electricity costs.
  2. $42M was equivalent to 32M Euro, in August 2013.
  3. The performance characteristics of the servers was not specified by James Hamilton. We assumed a top of the line Intel CPU in 2010, such as the Intel Xeon X5677. It has a CFP2006 benchmark of 49 [2]. The total performance is estimated at 46000 x 49 = 2254000 units on the CFP2006 benchmark scale.
  4. $3.94M = +- 3M Euro
  5. 31 euro + 6 euro for power adaptor. Renewed every 3 years: 37/3=12.3 euro/yr.
  6. 1000 euro, renewed every 3 years: 333 euro/yr.
  7. The Whetstone figure was taken from [3], 2 threads, Core2 Duo 2.4 GHz 32 Bit SP.


.