7DS: Node Cooperation in Mostly Disconnected Networks Henning Schulzrinne (joint work with Arezu Moghadan, Maria Papadopouli, Suman Srinivasan and Andy Yuen) Dept. of Computer Science Columbia University
Dec 22, 2015
7DS: Node Cooperation in Mostly Disconnected Networks
Henning Schulzrinne(joint work with Arezu Moghadan, Maria Papadopouli, Suman
Srinivasan and Andy Yuen)
Dept. of Computer ScienceColumbia University
Problems with Wide Area Wireless
• 802.11 currently hard to deploy across city or large area• 2.5G/3G wireless networks not suitable for large data volumes
– cost and network design• Problem: How can mobile devices / gadgets get and generate
information?• Examples:
– PDAs and smartphones• digital content such as news, maps, tourist information,
entertainment– digital still and video cameras MB of data
Solution: 7DS!
Concept
• Use local peer-to-peer wireless networks to exchange information– Peers can get information they do not have
from another peer• Model core Internet data applications
– Internet mobile: web access (retrieval)– Mobile Internet: email (messaging)
• Once wireless 7DS network set up, it could be used for any purpose– Getting web pages from peers– Sending e-mails– File transfer to selected client
IllustrationIn the absence of the Internet, nodes can exchange information amongst themselves
Internet
Design
• Peer-to-peer network set up using zeroconf– Protocol enables devices to get IP address
and communicate without a DHCP / other server
• Proxy server serves content– If connected to Internet, functions normally– If not connected, connects to peers to get
information
Design
• Search engine– Allows self / peers to search within the
device’s database for necessary information• Multicast querying system
– Allows peers to query systems in peer-to-peer network
– Searches can be for files or keywords• Transport System
– To forward e-mail
Connection & Proxy Server
• Connection set up using zeroconf protocol– Similar to AppleTalk, Microsoft NETBIOS, Novell
IPX
– Uses link-local addressing, multicast DNS, DNS service discovery
• Proxy server– Peer’s user client uses localhost proxy server by
default
– Detects Internet connection availability
– If Internet, normal network operation
– No Internet, peer-to-peer data exchange
Search Engine
• Provides ability to query self for results
• Searches the cache index using Swish-e library
• Presents results in any of three formats: HTML, XML and plain text
• Similar in concept to Google Desktop
Query multicast engine• Used to actually exchange
information among peers• Requesting peer broadcasts
a query to the network• Responding peers reply if
they have information– Send encoded string with
list of matching items• Requesting peer retrieves
suitable information
Analytical work on 7DS • Email upload application
– How to remove message replicas upon email delivery?
– Impact of a feedback channel?• P2P file exchange
– Throughput related to content popularity model– How to improve dissemination for unpopular content?
• Push based vs. Pull based data dissemination– Bloom filter based vs. query based strategy– Optimal design parameters for Bloom filter based
algorithm
Email Delivery App• How to remove redundant message replicas upon
email delivery?– Time-based (TB): purge message upon TTL
expiration TB(TTL)– Hop-based (HB): purge message when
dissemination tree reaches specific depth and breadth HB(breadth,depth)
– Feedback (FB): receive notification upon message delivery
• 4 Schemes: TB, HB, TB/FB, HB/FB• Storage cost (number of message replicas)
Arrival to AP 17 min on avg
TB(170) HB(2,6) HB(3,4) HB(4,3) HB(7,2)
No Feedback 3169 114 102 75 50
Feedback(FB) 54 41 32 29 18
Email Delivery App (Cont ‘d)• Hop-based scheme is
superior, with smaller expected cost and variance
• Expedited (100sec) and reliable message delivery (Pd=1) possible at small cost – 50 replicas for HB(7,2)
• With feedback HB(7,2) has a cost of 18 replicas
Stopping time: time when all message replicas are purgedMessage delivery time: time when email message is deliveredFB and NFB schemes have same message delivery time statisticallyStopping time and message delivery time is the same for FB schemes