Upcoming SlideShare
×

# Pcgrid presentation qos p2p grid

680 views
594 views

Published on

0 Likes
Statistics
Notes
• Full Name
Comment goes here.

Are you sure you want to Yes No
• Be the first to comment

• Be the first to like this

Views
Total views
680
On SlideShare
0
From Embeds
0
Number of Embeds
40
Actions
Shares
0
7
0
Likes
0
Embeds 0
No embeds

No notes for slide
• When peers exchange resources, the balance from the provider is increased and from the consumer is decreased
• ### Pcgrid presentation qos p2p grid

1. 1. Predicting the Quality of Service of a Peer-to-Peer Desktop Grid<br />Marcus Carvalho<br />OurGrid Project<br />Federal University of Campina Grande, Brazil<br />
2. 2. 2<br />P2P Desktop Grid<br />Predicting the QoS of a P2P Grid<br />Network of Favours:<br /><ul><li>Prioritise local jobs
3. 3. Prioritise collaborators</li></ul>Uncertainties on the<br />Quality of Service<br />Incentive mechanism to motivate resource sharing<br /><ul><li>Volatile resources
4. 4. Competitive environment</li></li></ul><li>3<br />P2P Desktop Grid<br />Predicting the QoS of a P2P Grid<br />
5. 5. Prediction of P2P Grid’s Quality of Service<br />How many resources will be available for a peer in a P2P Desktop Grid at future instants of time?<br />4<br />Problem<br />Predicting the QoS of a P2P Grid<br />We propose a prediction model based <br />on system behaviour and grid information<br />
6. 6. Network of Favours<br />A = 1<br />C = 0<br />B = 2<br />C = 4<br />A = 0<br />C = 0<br />Predicting the QoS of a P2P Grid<br />B = 1<br />C = 2<br />Peer A<br />Peer B<br />Resources proportionally shared<br />A = 0<br />B = 0<br />A = 2<br />B = 0<br />Peer C<br />Task Queue<br />Balances<br />Task<br />Resource<br />5<br />
7. 7. Network of Favours<br />A = 0<br />C = 2<br />B = 0<br />C = 0<br />A = 0<br />C = 0<br />B = 0<br />C = 2<br />Peer A<br />Peer B<br />Resources equally shared<br />A = 0<br />B = 0<br />Peer C<br />Task Queue<br />Balances<br />Task<br />Resource<br />6<br />Predicting the QoS of a P2P Grid<br />
8. 8. Predicting the QoS of a P2P Grid<br />7<br />Prediction model<br />Resources equally shared<br />Resources proportionally shared<br />Proportion of local peer’s balance for peer d<br />Estimation of resources obtained from a peer d attime t =<br />We need to estimate the functions’ values for a prediction target time tp in the future<br /> <br />  myBalanced(t)othersBalancedt+myBalanced(t) ∙  idleResourcesdt idleResourcesd(t)consumers (t)<br /> <br />If any consumer has positive balance<br />tc<br />otherwise<br />tp2<br />tp3<br />tp1<br />
9. 9. We assume that information at the current time tc is available<br />Estimating local peer’s balance update<br />myBalancedtc is known<br />myBalancedt=myBalancedt−1−obtaineddestimatet−1<br />for  t≥tc<br />We estimate the other functions as constant<br />idleResourcesdt=idleResourcesd(tc)<br />othersBalancedt=othersBalanced(tc)<br />consumerst=consumers(tc)<br /> <br />Predicting the QoS of a P2P Grid<br />8<br />Prediction model<br />tp2<br />tp3<br />tp1<br />tc<br />
10. 10. Obtained Ratio<br />OR= obtainedrequested<br />Estimated Ratio<br />ER=min⁡(  estimatedrequested ,100% )<br />Prediction Error<br />𝜉 = ER − OR<br /> <br />Predicting the QoS of a P2P Grid<br />9<br />Evaluation - Metrics<br />
11. 11. Workload<br />NorduGrid trace (Grid Workload Archive)<br />6 months<br />40 sites (peers)<br />Prediction target times (tp)<br />All consumption windows of each peer<br />Δtp = {0, 2, 4, 6, ..., te−tc} minutes<br /> <br />Predicting the QoS of a P2P Grid<br />10<br />Evaluation - Scenarios<br />tc<br />te<br />Δtp3<br />Δtp1<br />tp3<br />tp1<br />tp2<br />Δtp2<br />
12. 12. Resources per peer r = {20, 40, 60, 80}<br />Same # resources for all peers<br />OR (obtainedrequested) variation over time:<br /> <br />Predicting the QoS of a P2P Grid<br />11<br />Evaluation - Scenarios<br />resources per peer  r = 20<br /> <br />resources per peer  r =40<br /> <br />resources per peer  r =60<br /> <br />resources per peer  r =80<br /> <br />
13. 13. Predicting the QoS of a P2P Grid<br />12<br />Prediction Results<br />resources per peer  r = 20<br /> <br />resources per peer  r  =40<br /> <br />resources per peer  r =60<br /> <br />resources per peer r =80<br /> <br />
14. 14. Predicting the QoS of a P2P Grid<br />13<br />Prediction Results<br />
15. 15. P2P Desktop Grids have volatile resources and a competitive environment <br />Uncertainty on the Quality of Service<br />It is possible to have good estimations on the number of resources available for a peer in the grid<br />Predictions based on system behaviour and grid information<br />Predicting the QoS of a P2P Grid<br />14<br />Conclusion<br />
16. 16. Prediction based on partial and inaccurate grid information<br />Explicitly model the idleness of resources<br />Evaluation with different workloads<br />Predicting the QoS of a P2P Grid<br />15<br />Future work<br />
17. 17. Predicting the QoS of a P2P Grid<br />16<br />Thank you!<br />Questions<br />?<br />Marcus Carvalho<br />marcus@lsd.ufcg.edu.br<br />http://www.ourgrid.org<br />