slides01 Flashcards
1
Q
CPU dealing with I/O and memory
A
- CPU to memory at high speeds
- CPU to I/O at low speeds
- I/O needs to be asynchronous and interrupt-driven so memory can deal with other tasks and not wait for it
2
Q
Steps in interrupt processing (6)
A
- device pulls interrupt request line; sensed by CPU after executing current instruction
- CPU detects interrupt, reads interrupt number and jumps to ISR (interrupt service routine) after saving current state
- CPU processes interrupt by running ISR
- CPU clears interrupt
- completes execution of ISR
- restores previous state and resumes computing
3
Q
DMA (direct memory access)
A
- once setup, CPU not involved in data transfers
- used for bulk transfers
- interrupt per block
4
Q
storage hierarchy
A
- registers
- cache
- main memory
- nonvolatile memory
- hard-disk drives
- optical disk
- magnetic tapes
5
Q
dual mode operation
A
- kernel is supervisor mode and the rest is in user mode
- the kernel is part of the OS; applications need to defer to OS/kernel for certain high-privilege operations
6
Q
advantages of dual mode operation
A
- OS protected from errant user/program
- errant users protected from each other
- allows a method to manage privileged execution
7
Q
normal vs privileged instructions
A
- normal: in both user and kernel modes, execution proceeds
- privileged: execution proceeds in kernel mode and user mode traps into OS
8
Q
why can’t users manage interrupts?
A
- program can be written to turn off all interrupts; no way for OS to regain control of CPU
9
Q
computing systems performance
A
- measurable through execution time of an application ie. want to minimize execution time to improve effectiveness
10
Q
Amdhal’s law
A
- maximum potential improvement to the performance of a system is limited by portion of system that cannot be improved
FORMULA - speedup = Told/Tnew
11
Q
cluster computing
A
- cheap way of obtaining high performance - using multiple machines
- latency is not optimal -> takes longer to send message from one machine to another
- high performance for applications that send few but large messages
