Replica Placement Heuristics of Applicationlevel Multicast ChiaHsing Yu

Replica Placement Heuristics of Application-level Multicast Chia-Hsing Yu Jiahua He CSE of UCSD Project Presentation of CSE 222 A

Outline n n Multicast and RMX Model and Heuristics Simulation and Results Conclusion and Future Work 2021/9/12 Project Presentation of CSE 222 A 1

Application-level Multicast n Goal n n Distribute Contents to Many Clients Problem How to reduce the load of the central server? n How to reduce the response time of requests? Replication at different servers n 2021/9/12 Project Presentation of CSE 222 A 2

RMX: Reliable Multicast pro. Xy TCP SRM: Reliable IP Multicast 2021/9/12 Project Presentation of CSE 222 A 3

RMX n Semantic reliability information representation of information n Sender can lower the stream resolution if the network load is heavy 2021/9/12 Project Presentation of CSE 222 A 4

Existing Problems n n Only sources, no replicas No request, only recovery request Static RMXs in network Static configuration of data groups 2021/9/12 Project Presentation of CSE 222 A 5

Related works n Replication in unstructured P 2 P (Princeton) n n PAST(Microsoft and Rice) n n Nodes with similar id’s Ocean. Store (Berkeley) n n n Owner, Path, Random On or near the clients Focus on persistent storage with versions Chain (Cornell) n n 2021/9/12 Machines with replicas of a same file form a chain Focus on availability Project Presentation of CSE 222 A 6

Model and Heuristics n n n Fixed sources and dynamic replicas Streaming multicast on demand No replication n n Baseline Replication on path n n n 2021/9/12 FIFO LRU Color Project Presentation of CSE 222 A 7

Baseline n n Only sources, no replicas Learning bridge scheme to search n n Learn routing information from incoming data Soft state: periodically refresh Request suppression Ideal condition: no loss 2021/9/12 Project Presentation of CSE 222 A 8

FIFO and LRU n n n Replication on path Broadcast to search FIFO: n n Remove the oldest one if no space LRU: n n 2021/9/12 Order the files by last usage Remove the oldest one if no space Project Presentation of CSE 222 A 9

Color n Graph coloring n n Visiting Frequency n n More frequently visited, more possible to be visited Cost function: dist * freq n n n Neighbors with different colors (files) from mine Can get more different files from neighbors Remove the file with nearest replica dist: distance to the nearest replica freq: visiting frequency Upper bound of the cost if removed 2021/9/12 Project Presentation of CSE 222 A 10

Simulator n Event-driven Simulator New Event Min Heap Event Handler Earliest Event 2021/9/12 Project Presentation of CSE 222 A 11

Simulator(2) n Stream-level Simulation n n SIM_SEND_STREAM( bit rate, length ) Input n n 2021/9/12 Network Topology Host Resources Stream Sources User Requests Project Presentation of CSE 222 A 12

Experiment Configuration n Network Topology n n Host Resources n n n 1270 sources (average 10 per host) 500 Kbps, 8000 seconds each Randomly distributed User Request n n n 127 hosts (data groups) Hard disk size variable Stream Sources n n Binary Tree Randomly distributed Total number variable Experiment Span n 2021/9/12 100 hours Project Presentation of CSE 222 A 13

Experiment Configuration (2) n Variances n n n Number of requests: 211 ~ 218 Hard disk size: 8 G ~ 128 G Metrics n n n 2021/9/12 Client view average response time Server view load (number of streams per RMX) load balance (standard deviation of load) System view throughput Project Presentation of CSE 222 A 14

Client View Avg. Response Time vs. # of Requests About 30% improvement 2021/9/12 Project Presentation of CSE 222 A 15

Client View Avg. Response Time vs. Disk Size Disk size outperforms replication strategy 2021/9/12 Project Presentation of CSE 222 A 16

Server View Avg. # of Streams vs. # of Requests About 50% improvement 2021/9/12 Project Presentation of CSE 222 A 17

Server View S. D. # of Streams vs. # of Requests About 50% improvement 2021/9/12 Project Presentation of CSE 222 A 18

Server View Avg. # of Streams vs. Disk Size Disk size outperforms replication strategy 2021/9/12 Project Presentation of CSE 222 A 19

Server View S. D. # of Streams vs. Disk Size Disk size outperforms replication strategy 2021/9/12 Project Presentation of CSE 222 A 20

System View Throughput vs. # Requests About 25% improvement 2021/9/12 Project Presentation of CSE 222 A 21

System View Upper bound 25. 4398 Throughput vs. Disk Size 2021/9/12 Project Presentation of CSE 222 A 22

Contributions n n Implement and analyze Baseline, FIFO, LRU algs Propose and verify Color heuristics n n Avg. response time: up to 30% improvement Load: up to 50% improvement Load balance: up to 50% improvement Throughput: up to 25% improvement 2021/9/12 Project Presentation of CSE 222 A 23

Future Works n n n Biased requests Heterogeneous environment (hosts, links, streams) Random forward More sophisticated heuristics Experiment in real environment 2021/9/12 Project Presentation of CSE 222 A 24