This document discusses parallel virtual machine (PVM) and how it handles parallel programming. It explains that PVM allows tasks to join named groups and perform collective operations on group members. It provides functions for tasks to join and leave groups dynamically. Synchronization between tasks can be achieved using blocking receives and barriers where tasks must wait at barriers until a specified number of group members have arrived. Message passing is also used to enforce precedence constraints between tasks by forcing the receiver to block until a message is received.

1. Parallel programming in the
parallel virtual machine
Submitted by
N.Jayanthi
Msc(cs)

2.  Change at any time during computation.
 Groups are useful in cases when a collective operation is
performed on only a subset of the task.

3.  A broadcast operation,which sends a message to all tasks
in a system,can use a named group to send a message to
only the members of this group.
 A task may join or leave a group at any time without
informing other tasks in the group.
 A task may also belong to multiple groups.
 PVM provides several functions for tasks to join and leave
a group,and retrieve information about other groups.

4. i = pvm_joingroup(group_name)
 Group named group_name.
 Created pvm_joingroup is called for the first time.
 The first caller gets 0 as instance number.
 Starts at 0 and is incremented by 1 every time a new task
join the group.
 Set of instance numbers may have gaps sas a result of
having one or more tasks leave the group.
 New member will get the lowest available instance
number.
 Maintaining a set of instance numbers without gaps is the
programmers responsibility.

5. info = pvm_lvgroup(group_name)
 Leave the group group_name.
 info will have a negative value.
 Task decides to rejoin this group at a later time, it
may get a different instance number because the
old number may have been assigned to another
task that may have joined.
 There are a number of other functions that can be
called by any task to retrieve information without
having to be a member of the specified group.

6.  Function pvm_gsize() can be used to retrieve the
size of a group.
 It takes as input the group name and returns the
number of members in the group.
 The function pvm_gettid() is provided to retrieve
the TIP of a task given its instance number and its
groupname.
 Function pvm_getinst() retrieves the instance
number of a tasjk its TIP and the name of a group

7. Synchronization constructs can be used to force
a certain order of execution among the
activities in a parallel program.
Synchronization in PVM can be achieved using
several constructs,most notably blocking
receive and barrier operations.
Example:
 Member of a group that finish their work
early may need to wait at a synchornization
point until those tasks that take a longer time
reach the same point.

8.  Message passing can be used effectively to force
precedence constraints among tasks.
 The blocking receive operation (pvm_recv())
forces the receiving task to wait until a matching
message is received.
 Sender of this matching message may hold its
message as long as it wants the receiver to wait

9. Two task: T0 and T1
 function g() in T1 is not executed until T0 has completed
the execution of the function f().
 order of execution using a send operation after calling f()
in T0 and a matching blocking receive operation before
calling g() in T1.
f()
Pvm_s end(200,t ag)
T0
(TIP=100)
Pvm_r ecv(100,t ag)
g()
T1
(TID=200)

10.  Parallel tasks can be sychronized through the use
of synchronization points called barriers.
 Members of a group can choose to wait at a barrier
until a specified number of group members check
in at that barrier.
 Function:
info=pvm_barrier(group_name,ntasks)
 Two input: group name ,number of group members

11. Group slave(T0,T1,T2)
Info=pvm_barrier T1 T2
PVM_barrier(“slave”,3)
PVM_barrier(“slave”,3)
PVM_barrier(“slave”,3)
wait
wait
proceed proceed proceed
Group: slave
T0