Page 122 - Computer_Science_F5

(e) Multithreaded programs can (c) It is important that in SMT, the OS
potentially scale well with increasing does not make the mistake of treating
numbers of processors. By distributing all threads equal, even if some come
tasks across multiple cores, workloads from different cores. This can lead
can be completed faster. to inefficient workload distribution.
FOR ONLINE READING ONLY
Imagine a restaurant manager Chapter Two: Performance and optimisation of computer processor
Challenges of multithreading assigning all tasks to the same waiter,
(a) Register conflicts: Instructions from even when others are free.
different threads might use the same
register names, requiring techniques Examples of SMT architectures
like register renaming to avoid data
corruption. (a) Intel Hyperthreading (HT)
(b) Program counter management: Each Launched in 2002, Hyperthreading was a
thread has its own program counter revolutionary technology that transformed
(PC) keeping track of its execution single-core Pentium 4 processors into
position. logical dual-core processors from the
(c) Virtual memory management: The operating system’s perspective. Each
operating system needs to manage physical core houses two architectural
memory access for multiple threads sets, essentially creating two logical cores
within a process and potentially for the OS.
across multiple processes.
These sets have their own:
(i) General-purpose registers: Used to
Considerations when designing SMT-
enabled processors: store temporary data during program
execution.
(a) SMT works well if threads are
well-behaved: There are not too (ii) Control registers: Manage the state
many demanding threads, and they of the processor and execution flow.
use different processing resources (iii) Interrupt control registers: Handle
(like doing both math and graphics hardware interrupts and exceptions.
calculations). Think of it like having (iv) Machine state registers: Track
a team of specialists who do not all various processor settings and
need the same tools at once. configurations.

(b) SMT works well if there is harmony (v) Benefits: Hyperthreading
in the workload: Threads do not fight significantly improved performance
over the same resources in the core, for workloads that could benefit from
allowing for smooth execution. TLP (Thread-Level Parallelism).
Imagine chefs in a kitchen with Applications like web browsing,
separate chopping boards and stoves, multitasking, and media processing
minimising bottlenecks. saw a noticeable boost.

113
Student’s Book Form Five

Computer Science Form 5.indd 113 23/07/2024 12:33