Talk at the Workshop on Wireless Information Systems at the Conference ICEIS Ordering in Mobile Networks Using Integrated Sequencers Sven Bittner, 13 April 2004 Institute of Computer Science Freie Universität Berlin

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 2/17 Motivation Distributed applications (ISs, replication, CSCW) use group communications (e.g. multicast)Distributed applications (ISs, replication, CSCW) use group communications (e.g. multicast) Systems expanded to mobile environmentsSystems expanded to mobile environments Wired networks:Wired networks: –Delivery trees only for delivery –Sequencers for ordering Wireless networks: vector clocks and matricesWireless networks: vector clocks and matrices –Bad scalability  Enhance and extend wired algorithms to suit mobile environments

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 3/17 Roadmap System ModelSystem Model AlgorithmsAlgorithms –Ordering –Dynamics –Mobility Performance EvaluationPerformance Evaluation Conclusions & OutlookConclusions & Outlook

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 4/17 Roadmap System ModelSystem Model AlgorithmsAlgorithms –Ordering –Dynamics –Mobility Performance EvaluationPerformance Evaluation Conclusions & OutlookConclusions & Outlook

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 5/17 System Model (1) Mobile Support Stations (MSS, S = {S 1,S 2,…,S n }) Mobile Support Stations (MSS, S = {S 1,S 2,…,S n }) – Fixed wired network (reliable and FIFO) – Acyclic graph in application layer (no restriction in case of link errors) (no restriction in case of link errors) Mobile Hosts (MH, H = {h 1,h 2,…,h x }) Mobile Hosts (MH, H = {h 1,h 2,…,h x }) – Connected to 1 MSS (local MSS S (h x )) – All communications via local MSS – Connections reliable and FIFO System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 6/17 System Model (2) Multicast Groups (MG, G = {g 1,g 2,…,g x })Multicast Groups (MG, G = {g 1,g 2,…,g x }) –Closed groups: only send to registered groups G (h x ) Sequencers (located on MSSs)Sequencers (located on MSSs) –One per MG –integrated components  no extra cost –Direction sufficient to know for MSSs DeliveryDelivery –MSSs know MGs from local MHs –MSSs exchange information about MH’s MGs  Direction of MHs of MGs sufficient System Model Algorithms Performance Evaluation Conclusions gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx Sequencer of g x gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx gxgxgxgx MHs reg. for g x

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 7/17 Roadmap System ModelSystem Model AlgorithmsAlgorithms –Ordering –Dynamics –Mobility Performance EvaluationPerformance Evaluation Conclusions & OutlookConclusions & Outlook

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 8/17 Algorithm – Causal Order Between multiple groupsBetween multiple groups Side-effect from tree-based overlay and FIFOSide-effect from tree-based overlay and FIFO System Model Algorithms Performance Evaluation Conclusions mxmxmxmx mxmxmxmx mymymymy mymymymy mx  mymx  mymx  mymx  my m x arrived (and is delivered) before m y in all nodes

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 9/17 Algorithm – Total Order Within each groupWithin each group –Send unmarked message to sequencer S x via MSSs –Sequencer sends marked message to all neighbors –Deliver marked message to MHs System Model Algorithms Performance Evaluation Conclusions Sequencer of g x mxmxmxmx mxmxmxmx mx√mx√mx√mx√ mx√mx√mx√mx√ mx√mx√mx√mx√ mx√mx√mx√mx√ mx√mx√mx√mx√ mx√mx√mx√mx√ mx√mx√mx√mx√ MHs reg. for g x

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 10/17 Algorithm – Causal & Total Order ProblemProblem –Destruction of multiple group causal order by different sequencers –Delivery of causally related messages depends on the path from the sequencer SolutionSolution –Local MSSs delay messages until last (causally preceding) arrived at sender  Causal order between multiple groups and total order within each group System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 11/17 Algorithm – Dynamics & Mobility Sequencer movement (from S x to S y )Sequencer movement (from S x to S y ) –Claim: located in central network position –Count messages from neighbors –Moving is transparent to other nodes Dynamic groups (offline/online)Dynamic groups (offline/online) –Register/deregister at local MSS –Update forwarding information Handoff (h x moves from S x to S y )Handoff (h x moves from S x to S y ) System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 12/17 Roadmap System ModelSystem Model AlgorithmsAlgorithms –Ordering –Dynamics –Mobility Performance EvaluationPerformance Evaluation Conclusions & OutlookConclusions & Outlook

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 13/17 Performance Evaluation Evaluate performance of network of MSSEvaluate performance of network of MSS –Represented by time to deliver a multicast Symmetric network, density parameter ρ - no. of subordinate MSSs (ρ=3, 4 levels)Symmetric network, density parameter ρ - no. of subordinate MSSs (ρ=3, 4 levels) Simulation using OMNeT++Simulation using OMNeT++ Uniform distribution of MHsUniform distribution of MHs 1 multicast group1 multicast group System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 14/17 Performance Evaluation – MSSs/MHs Sequencer in central network position, ρ=3Sequencer in central network position, ρ=3 –Significant decrease –More MSSs  less delivery time –67 to 136 times faster –Linear increase (constant extra-work per MH) –12 to 19 times slower System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 15/17 Performance Evaluation – Sequencer 10 level of MSSs, ρ=210 level of MSSs, ρ=2 (all) – MHs at all MSSs (outer) – MHs on outer level (inner) - MHs on 7 inner levels –Central position  best results Up to 21% performance increaseUp to 21% performance increase –Best: (outer) 4 times as much MHs as in (inner), but faster delivery4 times as much MHs as in (inner), but faster delivery Reason: inner MSSs not involved in deliveryReason: inner MSSs not involved in delivery –Worst: (inner) 8 times less MHs as in (all), but same delivery time8 times less MHs as in (all), but same delivery time Reason: inner MSSs always same loadReason: inner MSSs always same load System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 16/17 Roadmap System ModelSystem Model AlgorithmsAlgorithms –Ordering –Dynamics –Mobility Performance EvaluationPerformance Evaluation Conclusions & OutlookConclusions & Outlook

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 17/17 Conclusions & Outlook System modelSystem model –Integrated sequencers, acyclic overlay network Algorithms for ordering, sequencer movement, changing groups and disconnections, handoffsAlgorithms for ordering, sequencer movement, changing groups and disconnections, handoffs –Claim: sequencer in central position EvaluationEvaluation –Best results in central position Future workFuture work –Evaluate handoffs and movement itself System Model Algorithms Performance Evaluation Conclusions

Sven Bittner - Ordering in Mobile Networks Using Integrated Sequencers 18/17 Thank you for your attention! Contact:Sven Bittner, Contact:Sven Bittner,