Question

Suppose a typical program has an instruction mix of 40%, 15%, 20%, and 25% for instruction types

A, B, C, and D respectively. On processor P1, the CPIs for these instruction types are 1, 2, 3, and 4

respectively, and the clock rate is 2GHz. One representative program, S, executes 1 million instructions.

(a) What is P1’s throughput? Don’t forget that the unit of throughput (in this context) is MIPS.

(b) How long does S take to run on P1?

(c) A compiler has options that allow it to generate 2 instructions of type A instead of 1 instruction

of type C. How much faster can you make a typical program run?

(d) The VLSI designers tell you that they can improve the clock rate to 3GHz. However, due to the

need to handle the extra power dissipation, a cooling system that adds 30% to the cost of the

package is needed. Assuming you want to maximize the performance/cost ratio, what do you tell

them to do?

(e) Processor P2 has the same ISA but needs 2 CPI for all instruction types. What clock rate would

P2 need to have the same performance as P1?

(f) An I/O operation of b bytes takes 3b+5 μs. Program S’ is like S but must also do k bytes of

I/O, though the I/O may be split into any number of transfers (e.g.,k transfers of 1 byte each,

k/2 transfers of 2 bytes each, . . ., 1 k-byte transfer). The system designers would prefer to do

multiple small transfers to avoid hogging the I/O bus. There is also a requirement that S’ takes

at most 3 times as long to run as S. How best do you split the I/O while keeping system designers

happy? You may assume that k is large enough to avoid using floor and ceiling functions (as an

approximation), and ranges over values for which your result makes sense.

Submission

Submit this on paper at the beginning of class, one problem per sheet.

c This

Answer 1