The DataCentric Revolution in Networking

1
The Data-Centric Revolution in Networking?

• Scott Shenker
• International Computer Science Institute
• U. C. Berkeley

Liberally stealing the insight and work of
others, particularly Hari Balakrishnan, Deborah
Estrin, Ramesh Govindan, Joe Hellerstein, and Ion
Stoica
2
Two Communities Apart

• Networking (Internet) researchers
• dont know and dont care about databases
• Vast gap between communities
• much more overlap with other systems communities
• But data-centrism has narrowed the gap
• metaphors and algorithms
• This talk will tell that story in reverse order
• Internet, then sensornets

3
Our Central Mission
Get data from here to there
from here to there
4
Host-centric Protocols

• Protocols defined in terms of IP addresses
• Unicast IP address host
• Multicast IP address set of hosts
• Destination address is given to protocol
• Protocol delivers data from one host to another
• unicast conceptually trivial
• multicast address is logical, not physical

5
Host-centric Applications

• Classic applications destination is intrinsic
• telnet target machine
• FTP location of files
• electronic mail email address turns into mail
  server
• multimedia conferencing machines of participants
• Destination is specified by user (not network)
• Usually specified by hostname not address
• DNS translates names into addresses

6
Domain Name System (DNS)

• DNS is built around recursive delegation
• Top level domains (TLDs) .com, .net, .edu, etc.
• TLDs delegate authority to subdomains
• berkeley.edu
• Subdomains can further delegate
• cs.berkeley.edu
• Hierarchy fits host administrative structure
• Local decentralized control
• Crucial to efficient hostname resolution

7
Network Research in Early 90s

• Consumed by a few obsessions
• Quality of service for streaming media
• Multicast
• Congestion control
• But nobody questioned host-centricity
• assumed to be the only way to build Internet

8
Surprise 1 The web catches on!

• But we dont....

9
The Web

• Web URLs have host-name/path format
• Essentially the same information as FTP
• Early web
• browsers basically a GUI for FTP
• URLs were easily transmitted pointers
• Early web was host-centric
• and largely ignored (but used) by net researchers

10
Modern Web

• URLs often function as names of data
• users think of www.cnn.com as data, not a host
• Fact that www.cnn.com is a hostname is irrelevant
• Users want data, not access to particular host
• The web is now data-centric

11
Data-centric App in Host-centric World

• Data still associated with host names (URLs)
• administrative structure of data same as hosts
• weak point in current web
• Key enabler search engines
• Searchable databases map keywords to URLs
• Allowed users to find desired data
• Networkers focused on technical problems
• HTTP, persistence (URNs), replication (CDNs), ...

12
We Missed the Point!

• We thought
• web was an aberration
• search engines were a sufficient hack
• No networker (except Jacobson) articulated that
• web had gone from host-centric to data-centric
• it was a harbinger of future applications

13
Surprise 2 Stolen Music is Popular!

• And we finally get the message...

14
The P2P Filesharing Phenomena

• Napster Fastest growing Internet application
• Music sharing is intrinsically data-centric
• data never associated with hosts
• Centralized searchable database
• listed IP addresses where content could be found
• analogous to GoogleDNS in the web
• Legal problems forced decentralization
• Led to Gnutella and other distributed programs

15
Gnutella-style File Sharing

• Gnutella nodes form an overlay network
• each node has a few neighbors in a virtual
  network
• virtual link node knows others IP address
• do app-level networking on this graph

16
Gnutella-style Searching

• Keyword queries are flooded (within scope)
• query is processed locally at each node
• all nodes having hits respond to source
• many variations on this theme (freenet, etc.)
• Clearly not scalable
• P2P traffic now sizable fraction of overall load
• We finally realize that we need a scalable way to
  find data for data-centric applications

17
Is there life outside the Internet?

• Yes, and we should have been listening!

18
Sensornets (predating P2P)

• Vision
• Many sensing devices with radio and processor
• Enable fine-grained measurements over large areas
• Huge potential impact on science, and society
• Technical challenges
• untethered power consumption must be limited
• unattended robust and self-configuring
• wireless ad hoc networking

19
Conceptual Challenge

• Sensornets are inherently data-centric
• Users know what data they want, not where it is
• Estrin, Govindan, Heidemann (2000, etc.)
• Centralized database infeasible
• vast amount of data, constantly being updated
• small fraction of data will ever be queried
• sending to single site expends too much energy

20
Flood-then-Aggregate

• General class of methods
• Flood query to all nodes (or in region)
• Nodes with data matching query respond
• Responses are aggregated as appropriate
• Examples
• Directed diffusion reinforce based on data
• TAG tree for flood and return-path aggregation
• Etc....

21
Scaling Problems

• This approach suffers as
• systems get bigger
• queries more frequent and more specific
• For current deployments, not an issue
• systems are small, queries primitive
• But if technology progresses as hoped
• want to get relevant data without flooding
• similar to situation in Internet

22
Is Data-centric Flooding Necessary?

• The initial decentralized data-centric designs
  (in both Internet and sensornets) used flooding
• unscalable and unsustainable
• Since data-centrism is here to stay, we cant
  ignore this problem
• We had to broaden our research charter

23
Our Revised Mission
Get data from here to there
Get data from here to there
24
A DNS for Data?

• Can we map data names into addresses?
• a data-centric DNS, distributed and scalable
• doesnt alter net protocols, but aids data
  location
• not just about stolen music, but a general
  facility
• A formidable challenge
• Data does not have a clear administrative
  hierarchy
• Likely need to support a flat namespace
• Can one do this scalably?
• Data-centrism requires scalable flat lookups

25
Distributed Hash Tables (DHTs)

• The latest networking fad....

Presented from the Internet perspective but
applies to sensornets as well
26
An Internet-scale Distributed Index

• Interface put(key,object), get(key)
• DHTs form a structured overlay network
• nodes choose particular neighbors
• all objects have keys, usually hash(name)
• each node responsible for range of keys
• puts/gets routed to appropriate node

27
Example Design Chord
?
?
?

• Node and object keys
• random location around a circle
• Neighbors
• nodes 2-i around the circle
• found by routing to desired key
• Routing greedy
• pick nbr closest to destination
• Storage own interval
• node owns key range betweenher key and previous
  nodes key

?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
Ownership range
?
?
28
Key Properties

• Large aggregate capacity O(n) storage/bwidth
• Scalable
• O(log n) routing hops and state
• O((log n)2) update costs for node join/leaves
• Robust self-configuring and resilient to
  failures
• Nonproperty strict guarantees when failures

29
Our Version of Data Independence

• DHT interface allows us to get data by name
• We no longer care where data is
• A radical transition in databases
• perhaps it will be one in networking as well
• Apologies to Joe Hellerstein...
• see latest SIGMOD Record for his article

30
Caveat!

• DHTs are a work-in-progress
• A flurry of research activity on
• security
• replication
• proximity
• real operational experience
• .....
• For rest of talk, we put these worries aside...

31
Why Not Centralized Solutions?

• Ugh! (and infeasible for sensornets)
• Fault tolerance avoid single point of failure
• Economic
• DNS donated machines, scales organically
• Centralized solutions require business model
• Issue still open....but irrelevant to
  data-centrism
• need to support interface
• DHTs allow us to choose between cent. and decent.

32
Multiple Roles for DHTs

• Application-specific
• rolled into P2P application, run on peers
• General-purpose service
• run on managed nodes
• Intrinsic part of Internet architecture
• run on managed nodes

33
Multiple Roles for DHTs

• Application-specific
• rolled into P2P app, run on peers
• General-purpose service
• run on managed nodes
• Intrinsic part of Internet architecture
• run on managed infrastructure nodes

34
Some Applications using DHTs

• Partial list
• File sharing
• Storage repositories and file systems
• Backup systems
• Event notification systems
• Electronic mail
• App-layer multicast and streaming media
• .....
• Useful substrate for many (not all) large
  distributed applications because HTs are useful

35
Multiple Roles for DHTs

• Application-specific
• rolled into P2P app, run on peers
• General-purpose service
• run on managed nodes
• Intrinsic part of Internet architecture
• run on managed infrastructure nodes

36
Internet-scale Query Processing

• Superficial motivation
• Joins can be implemented with hash tables so...
• Distributed joins can be implemented with DHTs
• Scaling latency O(log n) while computation O(n)
• PIER (talk later today in session A9!)
• joins, aggregation, recursive and continuous
  queries
• Intended targets
• data in the wild (filesharing, net monitoring,
  etc.)
• schema provided by standardized protocols
• no need for ACID semantics

37
More Complex Queries

• Range search
• using prefix hash table
• no need to walk tree
• Keyword search
• engineering the boolean approach
• Active research on DHT-based distributed data
  structures for search (net and db communities)

38
Multiple Roles for DHTs

• Application-specific
• rolled into P2P app, run on peers
• General-purpose service
• run on managed nodes
• Intrinsic part of Internet architecture
• run on managed infrastructure nodes

39
Cleaning Up the Architecture

• Making URNs a reality
• webNG based on flat and opaque DHT keys
• enables persistence and eliminates branding
• Host identifiers versus routing information
• IP addresses currently (and stupidly) serve as
  both
• DHT key host id, resolves to routing address
• Architectural challenge for basic protocols

40
Subverting the Architecture

• Use DHT for forwarding, not just lookup!
• e.g., Internet Indirection Infrastructure (i3)
• similar in spirit to multicast (logical
  addressing)
• transcends current naming/addressing structures
• Make overlay the real network layer
• turn IP into a link layer technology
• Leverages, not limited by, current infrastructure
• New network layer is still simple, but not IP

41
New Generation of Networking?

• Current Internet relies on hierarchies to scale
• DNS naming, IP addressing, etc.
• Hierarchies limit flexibility
• addresses and names have to fit given structure
• need to care where data/machines are
• Scalable flat lookup avoids hierarchy
• network would be structure independent
• Less of a distinction between hosts and data

42
Do DHTs Apply to Sensornets?

• Can we build them?
• Do they help?

43
Finding Sensornet Data w/o Flooding

• Extract high-level features or events
• Temperature spikes, toxins, animal sightings
• Name these events
• Store/Access events with DHT-like structure
• Can later get detailed data from specific nodes
• Call this data-centric storage (DCS)
• Good for frequent specific queries
• Not good for long-running or aggregate queries
• But how do you build a sensornet DHT?

44
Geographic Routing

• Nodes know own and neighbors positions
• Packets routed to geographic destination
• Greedy forwarding, when possible
• If greedy fails at a void, use the right hand
  rule to navigate around the void

B
(x,y)
A
45
Geographic Hash Table (GHT)

• Keys hashed to random coordinates
• Likely no node exists at that location!
• Forwarding ends at node closest to destination
• Closest node stores the data

A
(x,y)
46
Additional Algorithms

• Caching and replication
• Cache around perimeter, replicate independently
• Structured replication (SR)
• Hierarchical decomposition of key space
• Tree of mirror images

Mirror Images
Hash(event)
47
More Complex Queries

• Using GHTSR (which has spatial structure)
• Range searches in space and value
• Wavelet analysis
• New data structures
• Higher-dimensional range searches
• Active research in distributed data structures
  for sensornet queries (in net and db communities)

48
We are finally on our way to the land of data
independence...

• We ask for your guidance....

49
Areas of Common Interest

• Algorithmic
• distributed data structures for search
• Metaphoric
• thinking about data independence