Useful Process Structures for Distributed Applications

1
Useful Process Structures for Distributed
Applications

• proprietor, administrator, worker, courier, and
  notifier processes
• W. M. Gentleman, "Message Passing between
  Sequential Processes the Reply Primitive and the
  Administrator Concept", Software Practice and
  Experience, Vol. 11, pp. 435-466, 1981.

2
Thoths Message Passing Primitives

• Send
• Sends a message to another process
• Blocked until a reply is received (also called
  Request)
• Receive
• Receives a message from another process
• Blocked until a message is received
• Reply
• Replies to a request
• Does not block

3
Courier Process

• convey information between two processes, neither
  of which wishes to block
• uses a pair of Sends
• no work performed
• code structure
• message null
• loop
• / message is from server /
• Send(message to client)
• / message if from client /
• Send(message to server)
• endloop

4
Notifier Process

• used to notify an administrator of the occurrence
  of some event
• blocks to event and then Sends message to the
  administrator to notify
• e.g., timer notifier
• code structure
• loop
• await_interrupt(timer tick)
• Send(message to server)
• endloop

5
Worker Process

• designed to do work on behalf of servers
• always Sends to server process
• reply message from server contains work
• initial Send would contain null result but
  subsequent Sends would contain results of work
• increases parallelism in servers
• code structure
• message null
• loop
• Send(message)
• do work
• message result
• endloop

6
Proprietor Process

• the simplest form of server
• processes messages in order received (which may
  be different from order sent)
• completes one request at a time ? limits
  parallelism
• e.g., disk server
• code structure
• loop
• Receive(request-message)
• process the request
• Reply(result)
• endloop

7
Administrator Process

• more complex form of server
• receives service requests from clients
• receives work requests from workers
• forwards clients requests to workers
• administrator itself does little or no actual
  work
• always Receives and Replies but never Sends
• only blocked when waiting for more service
  requests by clients, or service completion from
  workers
• can reorder requests
• can be used to serialize a critical section
  without restricting parallelism

8

• Administrator Code structure
• loop
• Receive(message)
• if (message from client)
• put request in work-queue
• else
• if (message from worker)
• Reply result to client
• get new request from work-queue
• Reply work to worker
• endif
• endif
• endloop

9
Example File System Process Structure
Operator_dialog
Device_ proprietor
Timer
Pathname_ administrator
. .
Couriers
From users
. . . . . .
. .
Freespace_ allocator
..
Drive_handlers
Backing_store _administrator
. .
(drive O)
From users
(drive I)
..
. .
(drive N)
10
Exercise

• How would the administrators algorithm be
  changed if multiple workers were employed?
• 1. If all workers performed the same work?
• 2. If workers performed different work?