Transcript Operating Systems

Message Passing
Introduction
•
•
•
•
•
Overview of Message Passing
Implementing Message Passing
Mailboxes
Higher-Level Protocols Using Message Passing
Case Studies in Message Passing
Overview of Message Passing
• Message passing is one way in which processes
interact with one another
• Processes may exist in the same computer or in
different computers connected to a network
• Uses of message passing:
– Client-server paradigm
– Backbone of higher-level communication protocols
– Parallel and distributed programs
Overview of Message Passing
(continued)
• Two important issues in message passing are:
– Naming of processes
• Names may be explicitly indicated or deduced by the kernel
in some manner
– Delivery of messages
• Whether sender should be blocked until delivery
• What the order is in which messages are delivered to a
destination process
• How exceptional conditions are handled
Direct and Indirect Naming
• In direct naming, sender and receiver processes
mention each other’s name
– In symmetric naming, both sender and receiver
processes specify each other’s name
– In asymmetric naming, receiver does not name process
from which it wishes to receive a message; kernel gives it
a message sent to it by some process
• In indirect naming, processes do not mention each
other’s name in send and receive statements
Blocking and Nonblocking Sends
• A blocking send blocks sender process until message is
delivered to destination process
– Synchronous message passing
– Simplifies design of concurrent processes
• A nonblocking send call permits sender to continue its
operation after send call
– Asynchronous message passing
– Enhances concurrency between sender and receiver
• In both cases, receive is typically blocking
• Kernel performs message buffering pending delivery
Exceptional Conditions in Message
Passing
• To facilitate handling exceptional conditions, send and
receive take two additional parameters:
– Flags indicate how exceptions should be handled
– status_area is for storing code concerning outcome
• Some exceptional conditions:
1.
2.
3.
4.
5.
Destination process mentioned in send doesn’t exist
Source process does not exist (symmetric naming)
send cannot be processed (out of buffer memory)
No message exists for process when it makes receive
A set of processes become deadlocked when one is
blocked on a receive
Implementing Message Passing
Delivery of Interprocess Messages
• Remember that:
– An event control block (ECB) has three fields:
• Description of the anticipated event
• Id of the process that awaits the event
• An ECB pointer for forming ECB lists
Mailboxes
• Mailbox: repository for interprocess messages
– Has a unique name
– Owner is typically the process that created it
• Indirect naming
– Only owner process can receive messages
– Any process that knows name of a mailbox can send
messages to it
• Kernel may provide fixed set of mailbox names, or it
may permit user processes to assign the names
– Varying levels of confidentiality
Mailboxes (continued)
Mailboxes (continued)
• Kernel may require a process to explicitly “connect” to a
mailbox before starting to use it, and to “disconnect”
when it finishes using it
– May permit owner to destroy it, transfer ownership, etc.
• Use of a mailbox has following advantages:
– Anonymity of receiver
– Classification of messages
• Through use of separate mailboxes for different classes
Example: Use of Mailboxes
• An airline reservation system:
– A set of booking processes
– Server wishes to process cancellations before bookings
and queries after both of them
Higher-Level Protocols Using
Message Passing
• Several protocols use message passing paradigm to
provide diverse services:
– Simple mail transfer protocol (SMTP) delivers electronic
mail
– Remote procedure call (RPC) is a programming language
facility for distributed computing
• Invokes a part of a program that is located on a different
computer
– Parallel virtual machine (PVM) and message passing
interface (MPI) are message passing standards for
parallel programming
The Simple Mail Transfer Protocol
(SMTP)
• SMTP is used to deliver electronic mail to one or more
users reliably and efficiently
– Uses asymmetric naming
– Can deliver mail across a number of interprocess
communication environments (IPCEs )
– It is an applications layer protocol: uses TCP / IP
• SMTP consists of several simple commands:
– MAIL, RCPT, DATA
• Typically used with a protocol that provides a mailbox
– Internet Message Access Protocol (IMAP)
– Post Office Protocol (POP)
Remote Procedure Calls
• Remote procedure call (RPC): used to invoke a part of
a program located in a different computer
– Semantics resemble those of a procedure call
• call <proc_id> (<message>);
– <message> is a list of parameters
• Stubs perform marshaling of parameters and convert them
to/from machine independent representations
– SunRPC and OSF/DCE standards, Java RMI
Message Passing Standards for
Parallel Programming
• Parallel program: set of tasks that can be performed in
parallel
– Executed on a heterogeneous set of computers or on a
massively parallel processor (MPP)
• Parallel virtual machine (PVM) and message passing
interface (MPI) are standards used in coding message
passing libraries; provide:
– Point-to-point communication between processes
– Barrier synchronization between processes
– Global operations for scattering (gathering) disjoint
portions of data in a message to (from) different
processes
Case Studies in Message Passing
• Message Passing in Unix
Message Passing in Unix
• Three interprocess communication facilities:
– Pipe:
• Data transfer facility
• Unnamed pipes can be used only by processes that belong
to the same process tree
– Message queue: analogous to a mailbox
• Used by processes within “Unix system domain”
• Access permissions indicate which processes can send or
receive messages
– Socket: one end of a communication path
• Can be used for setting up communication paths between
processes within the Unix system domain and within certain
Internet domains
Message Passing in Unix (continued)
• One common feature: processes can communicate
without knowing each other’s identities
Message Passing in Unix: Pipes
• FIFO mechanism for data transfer between processes
called reader and writer processes
• Usually implemented in file system
– But, data put into a pipe can be read only once
• Removed from pipe when it is read by a process
• Two kinds of pipes: named and unnamed
– Created through the system call pipe
– A named pipe has an entry in a directory
• Like a file, but size is limited and kernel treats it as a
queue
Message Passing in Unix: Message
Queues
• Analogous to a mailbox
• Created and owned by one process
– Creator specifies access permissions for send/receive
• Size specified at the time of its creation
Message Passing in Unix: Sockets
• A socket is one end of a communication path
– Can be used for interprocess communication within the
Unix system domain and in the Internet domain
• Server can set up communication paths with many
clients simultaneously
– Typically, after connect call, server forks a new process
to handle the new connection
• Leaves original socket created by server process free to
accept more connections
• Indirect naming: address (domain) used instead of
process ids
Summary
• Message passing paradigm realizes exchange of
information among processes without using shared
memory
• Useful in: microkernel-based OSs, client–server
computing, higher-level communication protocols, and
parallel or distributed programs
• Sender/receiver naming: symmetric, asymmetric,
indirect (mailbox)
• Message passing is employed in higher-level protocols
such as SMTP, RPC, PVM, and MPI