P2P: Person-to-Person Decentralized Work

1
P2P Person-to-Person Decentralized Work

• Gregory Alan Bolcer
• http//www.endtech.com/
• info_at_endtech.com

2
Outline

• Philosophy
• Central Argument
• Case Studies
• Magi Approach
• Unique Applications

3
Philosophy Centralization

• Manage complexity
• Ensure security of information, control
• Performance and scalability
• Leverage economies of scale
• What do you have on your desktop? Single CPU or
  1000s of lightweight ones?
• Reflects Organizational Boundaries and cost of
  crossing organizational boundaries is prohibitive
• Tools technology geared toward centralized
  solutions.

4
Philosophy Pure Decentralized

• Once you commit, its difficult to retract
• Allows you to prevent forcible removal of
  information (good and bad)
• No central authority to fix errors, but Increased
  reliability
• reduced vulnerability to attack
• graceful degradation of errors
• Low cost of communication
• Many decentralized systems have centralized
  component

5
Cases Freenet, Gnutella, Napster

• Freenet
• No form of centralized control or administration.
  
• Virtually impossible to forcibly remove
  information - doing so distributed content more
  widely
• Authors and Readers may remain anonymous
• Demand-based routing and shared storage
• Difficult to determine where information is
  stored
• Content is encrypted and distributed
• Limitation of legal liability suitcase and
  closet
• First-come, Key-based access scheme
• Michael Jordan and Barbie problem
• Not controlled or dependent upon Domain Naming
  system no permanent Internet connection to
  publish

6
Cases Freenet, Gnutella, Napster

• Gnutella
• No Central Directory
• Access-based file availability
• Relies on Im Here.. static IP addresses
• Locational identification
• Napster
• Centralized User ID directory
• Centralized metadata cache
• Shared understanding of name and content

7
Philosophy Decentralized Work

• It is difficult to centralize enterprise data
• Knowing when to prevent access is extremely
  difficult
• 70 of enterprise data is not located in a
  centralized server or database,
• its on the desktops, laptops, palmtops, PDAs,
  smartphones, etc.
• Need to centrally scale the business logic with
  access to information in place
• It is an expensive operation to centralize data,
• Its constantly changing
• ERP Large Database systems have discovered this
• Human nature, I want to copy it and do it myself
• Human work not easily segmentable, overlap of
  work leads to social and political problems

8
Problems with Web for Decentralized Work

• No Event backchannel
• Web typically client-server
• Dynamic data generated on demand polling or
  refresh for new content
• Server cant initiate connection to client
• Routing and Naming difficult
• No named clients other than logging client
  environment
• Dynamic IP and dialups
• Not truly Ubiquitous

9
Magi Approach

• Goal of Magi to scale the Web beyond the desktop
  and server to every Internet-enabled device
  billions instead of millions
• Thin-server on every device to speak HTTP to
  every other device WebDAV for writable Web
• Similar to Freenet, Gnutella, Napster but doesnt
  reinvent the Web Not just sharing but access
• Apache HTTP server plugins including other p2p
  protocols
• Extensible Java protocol implementations
• XML-based access controls using user controlled
  Buddy lists
• Dynamic authentication controls

10
Magi Approach

• Access and control of users own information
• in place where its stored rather than moving
  information around the Internet
• Namespace controls access and authentication
• Claims to namespace map to devices, even if not
  active
• Keys represent users, not content
• GregLaptop (Gregs Laptop) refers to the same
  installation whether its at work behind a
  firewall, at home on a DSL connection, or on the
  road on a dialup connection
• URL Mapping http//endeavors.org/Magi/Dir/GregLa
  ptop

11
Magi Approach

• Decentralization of administrative control of
  authorization and access controls
• Users buddy lists allow permissions
• Default directories for default levels of access
  which include read, read/write, group, owner, and
  public based on HTTP and WebDAV methods
• Permissions can be locally defined or
  authenticated with a 3rd party server or service
• Locational awareness of information provides
  context for the information
• Centralization of Metadata is a cheaper operation
  than information collection allows knowledge
  about offline data

12
Magi Application GUI Java Win32
13
Magi Architecture
Magi Service
Apache HTTP Server
Xerces XML Engine Jserv Servlet Engine JSDK
2.x Java2 JRE Java SDK Plugins
MagiDAV
XML Buddy Lists and ACL Triggers
Apache Module Plugins
MagiSWAP
WAP Gateway
Dynamic DNS Services
MagiWAP
Magi GUI
Online Storage Sites
Other protocols Plugins
HTTP Event Service API
E-Service Manager
Smart Updates
Magi Work Folders Inbox Outbox Private Public
Shared Services
14
Magi Approach to Work

• Centralize Authentication to maintain control
  and access
• Centralize Metadata to allow tracking and
  status, even of offline information
• Peer to peer nature allows negotiated content
  and long duration tracking, status, and
  processing
• Every Magi server can also be a registration
  authority

15
Unique Applications E-commerce (Now)
E-Commerce Web Server
Electronic Wallet
Content

• Costly to put up e-commerce site
• Significant technological barriers to market
• Non-standard data and APIs
• Information controlled by vendor

Shopping Cart
Ship Track
Credit Card Settlement
16
Unique Applications Magi E-commerce
Lightweight Web Server
Product ID

• Decentralize Shopping Cart, E-wallet, and
  settlement services
• Putting up a product for sale only requires
  registering the product
• Anyone can become a Vendor
• Information Controlled by Purchaser/User

Magi
Electronic Wallet
Shopping Cart
Settlement E-Services
Ship Track
17
Unique Applications Home Networking
Lightweight Web Server
Service ID

• Remote Control of Networked services
• Secure sphere of networkable permissions
• Access and control own information

Magi
Magi
18
Unique Applications Streaming Data
Using a WAP phone as a remote control, a user can
download a media format (QT,MP3,AVI, etc) to his
desktop proxy and then stream the contents to any
Magi enabled device using current Web technology.
If the Magi server is on the phone, the local
IrDA or Bluetooth connection can provide the
local connection to any other IrDA or Bluetooth
device.
Magi
19
Conclusions

• Magi is a client side Web application server that
  provides an event destination
• Centralizing metadata and authentication allows
  maintaining of control and information about
  disconnected information
• Providing access rather than moving data allows
  work to remain decentralized

Http//www.endtech.com/ http//magi.endeavors.org/