Network-aware Service Placement in Community Network Clouds Mennan - PowerPoint PPT Presentation

Network-aware Service Placement in Community Network Clouds Mennan Selimi mselimi@ac.upc.edu Advisors: Prof. Felix Freitag (UPC), Prof. Luis Veíga (IST/INESC-ID) EMJD-DC Summer Event 2 nd of June, 2016

Activity Network - started in 2004 - open, free and neutral community network - extremely dynamic and diverse - Guifi.net foundation: - non-profit - non partisan Growth of Guifi.net in - without conflict of interest terms of operational nodes Scaling - growing exponentially at triple digit every year - more than 30K nodes ( 31090 nodes, May 2016 ) - 50.000 + km of links Goal - internet for everyone ! Evolution of the total inbound and outbound Guifi.net traffic over the last 2 years

Community Network Clouds - The vision of cloud-based services in community networks - We are talking about specific community cloud: - built in community networks - hosted in community-owned computing resources - providing services of local interest - a P2P cloud made of individual computers and 
 home gateways Pros Cons - no single entity that owns or controls - can’t provide the QoS of Google, Amazon - low-power devices, unreliable network - to join / leave no need for permission or consent - small components, energy-efficient, well distributed

Cloudy: Community networking cloud in a box User CLOUDY Web interface API CLOUDY Service Layer Console - Debian GNU/Linux distribution aimed to foster 
 Storage Streaming Network the transition and adoption of the 
 PeerStreamer SNP Service DNS Service Tahoe-LAFS Syncthing WebDAV Proxy3 GVoD Community Network cloud environment VoIP - can be installed in VMs or “bare metal” Serf Service Service - service discovery platform Discovery Announcement Avahi - decentralization BASP Network Coordination - user-friendliness and experience Community Network CLOUDY - free, open source software (FOSS) http://cloudy.community/ M.Selimi et al., Cloud Services in Guifi.net. Computer Networks (2015)

Measurements taken: October 2015 - March 2016 1.0 Number of nodes (90) 100 UPCc6-ab 0.9 75 UPC 50 1.5 0.8 25 0 0.7 0 25 50 75 0.6 Bidirectional links (291/663) Presence (%) 1.0 y (km) ECDF 100 0.5 75 50 0.4 25 0 0.3 0.5 0 100 200 300 0.2 Unidirectinal links (372/663) 100 GSmVictoria-RK71 0.1 75 50 0.0 0.0 25 0 0 1 2 3 4 10 - 3 10 - 2 10 - 1 10 0 10 1 10 2 10 3 0 100 200 300 x (km) Link traffic in the busy hour [kbps] ( log 10 scale) Index QMP Topology Node and Link presence Link traffic in the busy hour ECDF GSgV-nsl-b828/GSgranVia255nl-c493 GSgV-nsl-b828/GSgranVia255nl-c493 1.0 27.5 1.5 25.0 1.0 0.5 22.5 0.8 Link throughput [Mbps] Link traffic [Mbps] GSgranVia255-db37/GScallao3Rd1-9090 GSgranVia255-db37/GScallao3Rd1-9090 1.6 0.6 25 1.2 ECDF 20 0.8 0.4 15 0.4 10 UPCc6-ab/UPC-ETSEIB-NS-7094 UPCc6-ab/UPC-ETSEIB-NS-7094 1.25 24 0.2 1.00 22 0.75 0.50 20 0.25 0.0 18 0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24 1 10 100 Hour of the day Link throughput [Mbps] ( log 10 scale) Hour of the day Traffic in three busiest links Throughput ECDF Throughput in three busiest links Monitoring page: http://dsg.ac.upc.edu/qmpsu/index.php

Topology UPCc6-ab UPC 1.5 - growth is unplanned - nodes often at non-optimal locations 1.0 y (km) - high diversity in the quality of links - unreachable nodes due to electric cuts - re-tuning radios by members 0.5 - well connected and adaptive GSmVictoria-RK71 - mesh topology 0.0 0 1 2 3 4 x (km) QMP Topology

1.0 0.9 0.8 0.7 0.6 ECDF 0.5 0.4 Usage 0.3 0.2 - highly skewed traffic pattern 0.1 0.0 - average traffic observed 70 kbps 10 - 3 10 - 2 10 - 1 10 0 10 1 10 2 10 3 Link traffic in the busy hour [kbps] ( log 10 scale) - busy hours between 22h-23h, 1.2Mbps Link traffic in the busy hour ECDF GSgV-nsl-b828/GSgranVia255nl-c493 1.5 1.0 0.5 Link traffic [Mbps] GSgranVia255-db37/GScallao3Rd1-9090 1.6 1.2 0.8 0.4 UPCc6-ab/UPC-ETSEIB-NS-7094 1.25 1.00 0.75 0.50 0.25 0 2 4 6 8 10 12 14 16 18 20 22 24 Hour of the day Traffic in three busiest links

1.0 0.8 0.6 ECDF 0.4 Network performance 0.2 0.0 - highly skewed throughput distribution 1 10 100 Link throughput [Mbps] ( log 10 scale) - average throughput observed 21.8 Mbps Throughput ECDF - throughput slightly affected by the traffic GSgV-nsl-b828/GSgranVia255nl-c493 - link asymmetry: 25% of links have 
 27.5 25.0 deviation higher than 30% 22.5 Link throughput [Mbps] GSgranVia255-db37/GScallao3Rd1-9090 25 20 15 10 UPCc6-ab/UPC-ETSEIB-NS-7094 24 22 20 18 0 2 4 6 8 10 12 14 16 18 20 22 24 Hour of the day Throughput in three busiest links

Services in the network User-focused services Proxy server (Internet access) 53.50 % In total 950 11.08% Web pages Network-focused services VoIP / audio / video / chat / IM 9.33% different 
 Data storage server 7.97% Network graph server 32.94% Radio / TV stations 3.50% DNS Server 35.48% services P2P server 3.50% NTP Server 17.20% 2.91% Linux mirrors Logs Server 6.45% Webcam 2.33% Bandwidth measurement 0.71% 1.94% Tunnel-based Internet access LDAP server 0.53% Mail server 1.16% Wake on LAN 0.35% 1.16% Weather station Games server 0.97% Bandwidth-hungry 0.38% CVS repository services 0.38% Server virtualisation (VPS)

Topology Usage - growth is unplanned - highly skewed traffic pattern - nodes often at non-optimal locations - average traffic observed 70 kbps - high diversity in the quality of links - busy hours between 22h-23h, 1.2Mbps - unreachable nodes due to electric cuts - re-tuning radios by members Given a community network cloud infrastructure, - well connected and adaptive what is an effective and low-complexity service placement solution that maximizes end-to-end Proxy server (Internet access) 53.50 % Network performance In total 950 performance (e.g., bandwidth) ? 11.08% Web pages VoIP / audio / video / chat / IM 9.33% different 
 Data storage server 7.97% - highly skewed bandwidth distribution Radio / TV stations 3.50% services - average throughput observed 21.8 Mbps P2P server 3.50% 2.91% Linux mirrors - throughput slightly affected by the traffic Webcam 2.33% 1.94% Tunnel-based Internet access Bandwidth-hungry - link asymmetry: 25% of links have 
 Mail server 1.16% services in 1.16% Weather station deviation higher than 30% Games server 0.97% QMP 0.38% CVS repository 0.38% Server virtualisation (VPS)

Bandwidth-aware Service Placement (BASP) - Phase 1 : use k-means algorithm to cluster nodes based S 2 S 1 on their geo-location - Phase 2 : find the node with maximal bandwidth within the cluster - the cluster head f ij - bandwidth of the path, from node i to j partition k of clusters: S = S 1 , S 2 , S 3 … S k S 3 - Phase 3 : reassign the nodes to the selected cluster heads

Bandwidth-aware Service Placement (BASP) - low complexity service placement heuristic - BASP consistently outperforms the currently service 
 placement in guifi.net by 35% bandwidth gain - as the number of services increases, the gain tends to increase accordingly

Video-streaming Service in Cloudy - PeerStreamer : mesh-based, video streaming platform, based on chunk swarming - The source “seeds” the overlay with N copies of each chunk - Immune to churn problems - Highly dynamic: topology is constantly updated - 30 real community network nodes 
 distributed in the city of Barcelona - Intel Jetway and Minix NEO Z64 
 devices running Cloudy Source node running in a Docker container - Camera attached to the source node 
 ( 512 kbps bitrate , 30 frame-per-second ) - 2 hours of live video-streaming

Video-streaming Service in Cloudy - video chunk loss decreased up to 3 pp (percentage point) in the peers side

Distributed Storage Service in Cloudy Introducer - Tahoe-LAFS: open source, secure distributed storage Storage Client - Provider-independent security Storage - Introducer, storage and client nodes Client Storage Client Storage - 30 real community network nodes 
 distributed in the city of Barcelona - Intel Jetway and Minix NEO Z64 
 devices running Cloudy Storage node running in a Docker container - client reading times 16% improvement Different workloads: 2 - 64 MB

References Towards Network-Aware Service Placement in Community Network Micro-Clouds M.Selimi, D.Vega, F. Freitag, L.Veiga. In Proceedings of the 22nd International European Conference on Parallel and Distributed Computing ( EuroPar 2016 ) Cloud Services in Guifi.net Community Network M.Selimi, A. Khan, E. Dimos, F. Freitag, R. Pueyes. Computer Networks , 93(2), 2015 Bandwidth-aware Service Placement in Community Network Micro-Clouds M. Selimi, Ll. Cerd à , L. Wang, A. Sathiaseelan, L. Veiga, F. Freitag. IEEE Conference on Local Computer Networks ( LCN 2016 ) ( in submission )

Network resilience Barcelona Topology Graph Lluçanes Topology Graph, Guifi.net guifi.net

Ongoing work - Service Resilience Web access : - DNS server - Proxy (gateway) Robustness of the web access service: - node failures Ratio of nodes in Barcelona that do not have web access (no DNS or Proxy access) when top-ranked nodes are removed. In collaboration with: Arjuna Sathiaseelan (University of Cambridge) and Davide Vega (University of Bologna)