QoS-Aware Resource Allocation for Mobile IoT Pub/Sub Systems Raphael Gomes 1 , Georgios Bouloukakis 2,3 , Fábio Costa 4 , Nikolaos Georgantas 3 , & Ricardo da Rocha 5 Seattle, USA, June 2018 International Conference on Internet of Things (ICIOT) 1 Instituto Federal de Goiás - Câmpus Goiânia, Brazil 2 Donald Bren School of Information and Computer Sciences, UC Irvine, USA 3 MiMove team, Inria Paris, France 4 Instituto de Informática, Universidade Federal de Goiás, Brazil 5 Instituto de Biotecnologia, Universidade Federal de Goiás, Catalão, Brazil
Motivation Traffic Information Management (TIM) system: TIM system Heterogeneous Dynamic International Conference on Internet of Things (ICIOT) - 2
Motivation publishers subscribers ON message broker connect broker 1 network discon. OFF voluntary discon. broker 2 broker 3 ON connect Message broker network can be allocated on Cloud: o Can we ensure specific end-to-end QoS between publishers and subscribers? o Does the Things’ intermittent connectivity affects the cost of resource allocation? International Conference on Internet of Things (ICIOT) - 3
Publish/Subscribe System home broker p 1 s 1 subscriptions partitioning b 1 b 4 b 7 p 2 s 2 b 2 b 5 b 8 p 3 s 3 home broker s 4 p 4 b 3 b 6 b 9 event matching publishers subscribers brokers event routing process 1 R. Baldoni et al., “Distributed event routing in publish/subscribe communication systems: - 4 a survey ,” DIS, Universita di Roma La Sapienza, Tech. Rep, 2005.
Resource Synthesis with Intermittent Connectivity How can we allocate message brokers of the pub/sub system ? b 5 b 2 Multi- Q 2 Q 1 dimension Knapsack problem b 7 Q 3 p → s QoS ? cost ($) ? Use of the WS-HEU heuristic to efficiently select the resource types. Additional message broker instances, if necessary. International Conference on Internet of Things (ICIOT) - 5
End-to-end QoS Estimation We rely on Queueing Theory to estimate end-to-end response times between publishers and subscribers. We use 3 different queueing models: M/M/1 Queue 1 M/M/c Queue 1 ON/OFF Queue 2 𝜈 𝜈 𝜈 End-to-end response time between p 2 , s 1 : s 1 ’s home broker 𝑐 5 𝑐 7 𝑞 2 𝑐 2 𝑡 1 𝜇 𝑝𝑣𝑢 𝑝𝑢ℎ 𝜇 𝑗𝑜 𝑝𝑢ℎ - 𝜇 𝑝𝑣𝑢 𝑝𝑢ℎ 𝜇 𝑗𝑜 𝑝𝑢ℎ 𝜇 𝑝𝑣𝑢 𝑝𝑢ℎ - 𝜇 𝑗𝑜 𝑝𝑢ℎ - - - - 𝑐 5 𝑐 5 𝑐 7 𝑐 7 𝑐 2 𝑐 2 𝜇 𝑗𝑜 𝜇 𝜇 𝑗𝑜 𝜇 𝑗𝑜 𝑐 2 𝑞 2 , 𝑡 1 𝜈 𝜈 𝜈 𝑐 5 𝑐 7 𝜈 𝜈 𝜈 1 D. Gross et al., Fundamentals of queueing theory, 2008. 2 G. Bouloukakis et al., ICC 2017, ICPE 2017. International Conference on Internet of Things (ICIOT) - 6
Experimental Evaluation QoS estimation synthesis result pub/sub parameters , …., Δ 𝑞 2 , 𝑡 1 Δ 𝑞 2 , 𝑡 1 b 2 b 5 b 7 𝑐 2 , 𝑤 1 𝑐 2 , 𝑤 b 1 b 4 b 7 …. |𝑊| p 1 s 1 𝑤 7 𝑤 1 𝑤 3 p 2 s 2 b 2 b 5 b 8 𝑤 7 𝑤 3 p 3 s 3 resource synthesis p 4 s 4 b 3 b 6 b 9 𝑤 3 resource parameters M-CloudSim 1 … 𝑤 1 𝑤 |𝑊| simulated environment end-to-end QoS s 1 p 1 requirement p 2 s 2 Δ thr p 3 s 3 1 https://github.com/raphaeldeaquino/mcloudsim - 7
Evaluation Setup & Results (1) 1. always connected event size (E) = 200 bytes VMs by Amazon, 2. intermittent connected λ = 0.5 to 23 events/sec Microsoft, Google μ = 33.2 CPU instructions Δ thr = 0.025 sec Subscriber always connected: International Conference on Internet of Things (ICIOT) - 8
Evaluation Results (2) Subscriber intermittent connected: T ON = 155 sec, T OFF = 96 sec. The response time is tightly coupled by the subscriber’s intermittent connectivity. By slightly describing Δ thr : 1. the cost increases significantly. 2. the response time remains at the same level. International Conference on Internet of Things (ICIOT) - 9
Next steps We allocate Cloud resources of IoT applications by considering end-to- end QoS requirements and the Things’ intermittent connectivity . Future work: o Introduction of additional QoS parameters: e.g., memory. o Estimation of end-to-end QoS between peers by using well known pub/sub routing algorithms. o Resource allocation of interoperability software artifacts. International Conference on Internet of Things (ICIOT) - 10
Thank you International Conference on Internet of Things (ICIOT) International Conference on Internet of Things (ICIOT) - 11
Recommend
More recommend