WideArea Computation Seminar Uppsala University Dept. Information Science Kenji Taguchi

1
Wide-Area Computation (Seminar)Uppsala
UniversityDept. Information ScienceKenji
Taguchi
2
Some Remarks

• If you come across some unknown terms such as
• IBMs Personal Area Network,
• try to find it out on the web.

3
Luca Cardelli

• An Italian researcher at Microsoft Research
  Cambridge
• Famous for his works on
• programming languages,
• ML (Meta Language)
• Modular3
• Obliq
• Object-orientation
• A theory of Objects
• Mobile computing
• Ambients

4
About this paper

• An invited paper appeared at
• International Colloquium on Automata, Languages
  and Programming (ICALP)99
• Short version of Abstractions for Mobile
  Computation

5
Summary of the paper

• LAN to WAN
• New observables on WAN (the Web)
• Several Mobilities
• Modelling Wide Area Computation
• Ambients
• Wide Area Challenge A Conference Reviewing System

6
How to proceed?

• Can you suggest any issues you would like to
  pursue at this seminar?

7
Question 1

• What is the authors claim about objects in the
  past?

8
Answer 1-1

• The sea of objects

Interface
Object

• Not well-structured

9
Distributed Object-Oriented Programming (1)

• CORBA, DCOM
• CORBA (Common Object Request Broker Architecture)
  by OMG (Object Management Group)
• Open, vendor-independent architecture and
  infrastructure that computer applications use to
  work together over networks
• DCOM( Distributed Component Object Model)
• Microsoft versions CORBA

10
Distributed Object-Oriented Programming (2)
IDL (Interface Definitin Language)
Request
ORB (Object Request Broker)
The ORB will act as a central object which will
receive a request from objects and pass it to its
implementation.
11
Question 2

• What makes the wide area network different from
  the local area network?

12
Answer 2

• Virtual locations
• Physical locations
• Bandwidth fluctuations
• Failures

13
My Answers to Q 2

• Virtual locations
• domain names
• more than one domain name can be assigned to the
  same network
• Physical locations
• The speed of light is too slow for the global
  network
• One extreme example is a probe (and its software
  system) sent to Mars.
• Bandwidth fluctuations
• congested along the continent
• Failures
• server might be down, but we cant figure out
  whether the failure was caused by the server or
  the network itself

14
Question 3

• How can mobile computation overcome those
  difficulties?

15
Answer 3

• Virtual locations
• Physical locations
• Bandwidth fluctuations
• Failures

Local resource
Barrier
Trust mechanisms
Physically distant
16
Section 2- Three Mental Images -

• Local Area Network
• Wide Area Network
• Mobile Computing

17
Mental Image 1
Administrative Domain
Fire Wall
18
Mental Image 2
19
Mental Image 3
AF81
EU
US
CDG
SFO
NSA
20
Question4- Mental Image 1 -

• What features does the mental image 1
  characterise?
• Are they plausible enough?

21
Mental Image 1 (1)- Characteristics -

• Computers
• about the same power
• Network links
• about the same bandwidth and latency

22
Mental Image 1 (2)- Characteristics -

• Predictability
• Communication delays are bounded
• processor response time can be estimated
• link and process failures can be detected by
  time-outs and by pinging nodes

23
Mental Image 1 (3)- Characteristics -

• Well-administered
• Secure against attach

24
Question 5- Mental Image 1 -

• What makes the local area network different from
  the previous mainframes with dumb terminals?

25
Question 6- Mental Image 1 -

• What new technologies emerged from the local area
  network?

26
Answer 6 - Mental Image 1 -

• Remote Procedure-call
• Client-Server Architecture
• Distributed object-oriented programming

27
Involved Mobilities

• Control mobility
• control flows move to a remote host
• Data mobility
• data flows move to a remote host
• Link mobility
• e.g., proxy servers
• Object mobility
• e.g., load balancing between more than one
  processor
• Remote execution
• e.g., rsh command in Unix

28
Meaning of each figure - Mental Image 2 -

• Computers differ in
• power
• availability
• Network links differ in
• capacity
• reliability
• Physical distance has
• visible effects
• time zones

29
The Internet

• Basic Knowledge on the Internet
• What features characterise the Internet?
• The routing mechanism
• Global address

30
LAN to WAN - Mental Image 2 -

• File systems
• NFS (Network File System) to Global file system
• The Web is not a global file system
• The Web is not just a big LAN

31
Network File System

• The Network File System (NFS) is a distributed
  file system that allows users to access files and
  directories located on remote computers and treat
  those files and directories as if they were
  local.

32
Mental Image 3

• More abstract image of mobility
• Several boundaries with different properties
  exist over the network
• political boundaries
• etc

33
Question 7- General Mobility -

• How do mobile computing and mobile
  computation interact with each other?
• Do they have common problem and principle?
• Mobile Computing
• physical mobility
• mobile hardware
• Mobile Computation
• virtual mobility
• mobile software

34
Answer 7-1-1
Security Check (bytecode verification)
35
Answer 7-1-2
Security Check (virus protection)
36
Answer 7-2
E.g., X-windows
37
Answer 7-3
High band-width connection
Server
PC
Server
38
Essential Feature of Mobility

• Intermittent connectivity
• can be caused by
• physical movement as well as
• virtual movement.
• Accidental disconnection
• bad infrastructure, solar flares (from Cardellis
  slides)
• Intentional disconnection
• privacy, security, quiet (from Cardellis slides)

39
Question 8- Barriers and Action-at-a-Distance -

• Why are barriers so important?
• Why can we ignore the notion of
  action-at-a-distance computing?

40
Some History- Access Barriers -

• The solution in the past is to abstract them
  away.
• Physical memory boundaries
• virtual memory
• Address space boundaries
• network proxies
• Firewall boundaries
• secure tunnels
• agent sandboxing

41
What is tunnelling?

• Tunnelling can be thought of as the act of
  encapsulating ordinary (non-secure) IP packets
  inside of encrypted (secure) IP packets.

42
Sandboxing

• a sand-box, a restricted environment in which the
  executables can execute safely.
• In a sand-box, an executable can only perform
  functions that cannot harm the system

43
Good old days (1)

• The internet
• Good old days
• Now

44
Good old days (2)

• Mobile Agents
• Good old days
• Free access to remote hosts
• Now
• hostile agents
• hostile hosts

45
Hiding virtual locations

• A security infrastructure makes crossing borders
  transparent.
• Cryptography is a great help, but not almighty.
• Hostile or unfair servers (hosts) would cause
  this problem unsolvable.

46
Hiding physical locations

• Caching and replication might be practical
  solutions, but not satisfactory
• No help for long-distance real-time control and
  interaction.
• Its possible to make all delays uniform by
  introducing unacceptable delays!

47
Hiding bandwidth fluctuations

• QoS (Quality of Service) technique would be of
  some help, but again, not satisfactory.

48
Revealing failures

• A fundamental problem,
• Inherent to distributed system,
• We cannot achieve distributed consensus
• which detects the failure of processors or of
  network nodes or links.

49
WAN Postulates - Summary -

• Locations
• separate locations with different properties
• Barriers
• in order to preserve the properties of locations
• Mobility
• crossing the barriers.

50
- Modelling Wide Area Computation -

• Move from observables to modelling
• Barriers
• Locality
• Mobility
• Security
• Communication
• long-distance communication can only be achieved
  by a combination of local communication and
  movement across barriers.
• Action at a distance is forbidden

51
Some Approaches (1)

• Wide area network languages
• agent-based languages
• e.g., TeleScript
• place-based languages
• e.g., Linda
• Ambients
• a generalisation of both notions

52
Some Approaches (2)

• Agents (Telescript, etc)
• Move globally, co-ordinate locally.
• Spaces (Linda, etc)
• Communicate globally, co-ordinate locally.
• Dynamic hierarchies (Ambients, D-Join, etc),
• Various synchrony assumptions, wide spectrum
  between D-Join, Ambients, and Seal.
• Distributed transactions, Workflow, etc
• (Taken from Cardellis slide)

53
Linda

• Invented by D. Gelernter
• Sometimes called Co-ordination Language
• based on a simple shared memory architecture with
  operations for updates, remove and add operations
• A memory space is called a tuple space which
  consists of tuples
• An example of a tuple ( camera,
  shop, 300)
• Operations which manipulate tuples are in,
  read and out. They use matching to retrieve
  data.

54
Ambients

• a bounded place
• can be nested
• can have a collection of local running processes
• can moves as a whole with all its subcomponents
• has a name

name
55
Ideals for Wide Area Languages

• Final Goal
• to program the Internet in some high-level
  language
• need to find programming constructs

56
Compatibility Requirements

• WAN-soundness
• No primitives which entail
• action-at-a-distance
• continued connectivity
• global consensus,
• security bypasses.
• WAN-completeness
• can mimic the behaviour of
• web surfers,
• mobile agents,
• users,
• any other entities which roam the network.

57
Necessary ingredients of WAL (1)

• Naming
• a solid reference to entities across barriers.
• Names should be pure in the sense that they are
  not assumed to contain any information about
  their creation.
• Migration
• migration of of active software components is not
  possible at the moment
• it must be supported at the OS level.
• Hot plug in/out of devices.

58
Necessary ingredients of WAL (2)

• Dynamic connectivity
• of software component.
• Static linking (at compile time)
• Dynamic linking (e.g., library)
• Ambient is a mixture of both
• Communication
• asynchronous at the wide-area level
• synchronous at the local-area level
• synchronisation between communication and
  movement operations remains open.

59
How Can/Should Things Interact?

• Remote interaction Asynchronous only.
• No instantaneous remote communication.
• No instantaneous remote mobility.
• But we cannot make everything asynchronous...
• Local interaction Synchronous is better.
• Synchronous local communication good for
  awareness of communication for both parties
  (sender and receiver) and particularly good for
  reactive/real-time systems.
• Synchronous local mobility good for awareness of
  movement for both parties (agent and host).
• They fit together (globalreactive)
• Mobility can be used to turn remote/asynch into
  local/synch.
• (e.g., send agent to do real-time control of Mars
  probe)
• Taken from Cardellis slides.

60
Necessary ingredients of WAL (3)

• Security
• Security abstractions should be provided at the
  programming-language level

key ... door ... try open_sesame (
key, door ) catch ( ex ) ...
open_sesame (key, door)
lang. construct
Some authentication mechanism based on
cryptographic protocol is assumed
61
Question 9

• The author claims that the followings must be
  supported at the language-design level. Do you
  think this claim is plausible?
• Ingredients of WAL
• Naming
• Migration
• Dynamic connectivity
• Communication
• Security

62
Some Examples- Relations between foundational
calculus and prog languages -

• Lambda calculus
• Functional programming languages (Scheme, ML,
  Haskell)
• First-order Predicate logic
• Logic programming languages (Prolog)
• Ambients
• Wide Area Programming languages?

63
Wide Area Challenge- A Conference Reviewing
System -

• Management of a virtual program committee meeting
  for a conference
• Roles in this system
• the conference chair(s)
• the programme chair(s)
• authors
• committee members
• external reviewers

64
Wide Area Challenge- A Conference Reviewing
System -
Prog. chair
Conf. chair
Review Form
Sub. Form
Comm. members
authors
Notification
Review Form
Final version
Review Form
Proceedings
Extern. Reviewers
Publisher
65
Next Seminar

• Mobile Agents White Paper by J. White.
• Some Remarks
• concentrate on the overview of underlying
  concepts
• dont need to poke your nose into programming
  examples (we will experience many examples
  later), but read the paper carefully .