Threads
So far, we assumed that a process has a single thread of
execution: i.e. a single sequence of executed instructions.
In many applications, however, several parts of the same
process can run in parallel. These parts share memory
address space and other resources.
In a multithreading system, a process may be composed of
several threads that run in parallel.
Memory space, I/O devices, ..etc. are assigned to a process.
However, CPU time is assigned to a single thread. Each thread
has independent state and context.
ELC-403b Lecture 4
Page 1
Threads
Advantages of Multithreading
Process can remain responsive if a part of it is blocked or
performs a long operation.

Creating a new thread of an existing process requires less
time and resources than spawning a new process. Also context
switching between threads of the same process is faster than
switching to another process.

In a mutliprocessor system, threads can be assigned to
different processors (or cores), thus speeding up the execution
of a given process.

ELC-403b Lecture 4
Page 2
1
Multiprocessor Scheduling
We consider the shown bus-based architecture which is
suitable for small number of processors.
CPU 1
CPU 2
cache
cache
Shared
RAM
Bus
ELC-403b Lecture 4
Page 3
Multiprocessor Scheduling
Two possible approaches:
Master-slave: A master CPU runs the kernel processes. Other
CPUs run user processes, and direct system calls to the master.
This simplifies coordination between processors, but master
becomes a performance bottleneck.
Symmetric Multiprocessing: code and data of OS are stored in
shared memory, and any CPU can run the system calls.
This avoids overloading the master, but more complex
coordination is required. E.g. what if two CPUs select the
same shared memory area?
ELC-403b Lecture 4
Page 4
2
Multiprocessor Scheduling
For multiprocessor scheduling, a simple approach is to have a
common ready queue (or multilevel queue) for all processors.
As a processor become idle, it immediately extracts a ready
process from the above queue.
This method has the advantage of load balancing: no CPU
remains idle while others are overloaded.
ELC-403b Lecture 4
Page 5
Multiprocessor Scheduling
Example: A dual processor system uses FCFS scheduling with a
common ready queue. Both processors become idle when
queue contain processes with process times of 2.4, 6 , 6.7, 7.5,
and 9.7.
Find the average waiting time for the above processes and
compare it with the case of single processor with double
speed.
ELC-403b Lecture 4
Page 6
3
Multiprocessor Scheduling
A process will repeatedly become blocked then run again.
Migrating a process from one CPU to another usually causes
the overhead of refilling the cache.
Thus, processor affinity may be preferred: i.e. attempt to keep
each process always on the same processor.
We have then two problems:
 Assignment of processes (or threads) to CPUs.
 How each CPU will schedule its share of processes.
Load balancing is more difficult in this case.
ELC-403b Lecture 4
Page 7
4