Introduction to the Grapevine Distributed System

1
Introduction to theGrapevine Distributed System

• CS-4513Distributed Computing Systems

2
Grapevine

• A distributed system for e-mail and other
  distributed activities
• Implemented at Xerox PARC in 1978-80
• Deployed throughout company
• Productized in Xerox 8000 series office systems
• Predates most Internet mail systems
• Embodies many principles of distributed computing
  systems

3
Grapevine (references)

• Birrell, et. al., Grapevine An Exercise in
  Distributed Computing, Communications of ACM,
  vol. 25, 4, April 1982, pp 260-274. (pdf)
• Schroeder, et. al., Experience with Grapevine
  The Growth of a Distributed System, ACM
  Transactions on Computer Systems, vol. 2, 1, Feb
  1984, pp 3-23. (pdf)

4
Goals

• Delivery mechanism for large, dispersed e-mail
  system
• Previous e-mail system were time-sharing based
• (now) Familiar model
• Senders
• Recipients
• Distribution lists
• Buffered delivery
• Independent clients (e-mail agents)

5
Design Goals

• No assumptions about
• Message content
• Client correctness
• Message delivery guarantee
• If a message was accepted, it would be delivered
  to recipients inbox
• or returned with error
• Failure of server ? unavailability of service
• Decentralized administration

6
Key Components

• Registration (i.e. naming) Data Base
• Distributed
• Replicated
• Recursively implemented
• Message Delivery System
• Transports, buffers, delivers
• Expands distribution lists
• Distributed
• Replicated

7
Issues Exposed by Grapevine

• Reliable Transport and Storage
• Synchronizing and Ordering of Operations
• Fault-tolerance
• Consistency and Replication
• Security and Authentication

8
Registration Database

• RName, value pairs
• Two types of entry
• Individual
• Group
• Group entry
• value list of RNames i.e., the group members
• May be distribution list, resource list, access
  control list, etc.
• Individual entry
• May be names of human users, servers, etc.
• value authenticator, ordered inbox list,
  connect site,

9
Organization

• Registry
• Subset of the registration database entries
• Organizational, geographic, or other convenient
  partition
• RName two-part name Name.R
• R is a registry
• Name is a unique name within that particular
  registry
• (Easily expandable to three part names or more)

10
Functions Provided by Grapevine Service

• Delivery services
• Accept message
• sender, password, recipients, message-body ? ok
• Message polling
• individual ? empty, non-empty
• Retrieve messages
• name, password ? sequence of messages ? ok

11
Functions Provided by Grapevine Service
(continued)

• Registry services
• Authenticate
• individual, password ? authentic, bogus
• Membership
• name, group ? member, non-member
• Resource location
• group ? members
• individual ? connect site
• individual ? ordered list of inbox sites

12
Grapevine Server

• Contains both Registration and Message servers
• Independent of each other
• Communicate only by internet protocols
• Registration server
• Contains replicas of one or more Registries
• Can accept change request for any of its
  registries
• Propagates changes to other replicas of Registry
• Message server
• Accepts any message for delivery
• Stores inboxes for some individuals
• (Individuals typically have more than one inbox)

13
Grapevine User Package

• Library code for Grapevine client programs
• E-mail clients
• Other kinds of distributed applications
• Handles all naming and addressing issues
• Clients never need to know name or address of any
  Grapevine server
• Servers optimized based on User Package

14
Grapevine Operation

• User P.Q sends message to user X.Y
• User X.Y wants authenticated access to server E

15
Message Delivery

• Find Message service
• Present sender, password, recipients to service
• Authenticates sender
• Validates recipients
• (Assume valid if recipient registry not
  available)
• Construct envelope (aka property list)
• Sender Name, Return-to, recipient list, postmark
• Postmark server internet address, server time
  stamp
• Accept message-body
• Store envelope message body in reliable storage
• Acknowledge receipt of message

16
Message Delivery (continued) Transport system

• Recursively enumerate groups in destination list
• Find inbox server contact for each recipient
• Sort by server, transmit one copy to each server
  with list of recipient names
• Duplicate elimination done at receiving server
• E.g., an individual a member of more than one
  group
• Queue if all servers down or inaccessible
• Receiving client polls all its inboxes
• Downloads from each
• Message deleted from inbox only after receipt is
  acknowledged

17
Replication in Message Delivery

• Any server can accept a message for delivery
• Individuals typically have more than one inbox
• However,
• Crashes can delay messages in transit.

18
Registration Database

• gv registry
• Replicated in all registration servers
• Groups in gv
• Names of registries
• Members are RNames of registration servers
  holding replicas of the group registry
• I.e.,
• reg is a Registry iff there is a group reg.gv
• Server holds a replica of reg iff its name is in
  reg.gv

19
Registration Database (continued)

• gv.gv is a group in gv registry
• Its members are RNames of all registration
  servers
• Adding a registration server ? adding a member to
  gv.gv
• Adding a new registry r ? adding a new group
  entry r.gv to gv registry
• Members of r.gv are a subset of gv.gv

20
Registration Database (continued)

• Used to locate resources
• Service is a group in the database
• Members are RNames of servers
• Servers are individuals in the database
• connect site is internet address of server
• To find foo.bar
• Contact local Grapevine server, ask for bar.gv
• Enumerate members, contact one of them, and ask
  for foo.bar

21
Finding the first Grapevine Server

• Ask another name service (e.g., DNS) for
  GrapevineRServer
• Returns a list of addresses
• List changes very infrequently
• or
• Broadcast to well-known socket on LAN
• Local Grapevine servers listen on socket
• Usually quicker

22
Summary

• Names in Grapevine
• People
• Services
• Servers
• Anything else you want them to be
• Registry maps names to
• Groups (i.e., a list of other names)
• Individuals (something with an address and/or
  other properties)

23
Summary (continued)

• Highly replicated
• Registries exist in multiple servers
• Individuals have more than one mailbox
• Mail can be delivered to anywhere for any
  recipient
• Registry used to manage itself!
• Distributed System issues to be addressed
• Synchronizing and Ordering of Operations
• Fault-tolerance
• Consistency and Replication
• Security and Authentication

24
Questions?