0
$\begingroup$

I am currently enrolled in a computer organization and design class, which I am struggling mightily with, and I have a final homework in my class that I need to get a perfect score on. One of the questions covers cache hit-misses, and the lecture we had on the subject was relatively brief. The question I have having issues with is:

A processor has a CPI of 1.4 with level 1 caches with a 100% hit rate. When running benchmarks, the level 1 instruction cache has a 1% miss rate and the level 1 data cache has a 3% miss rate. 30% of the instructions access memory through load and store operations. Main memory has a 50 nano second access time and the clock is running at 2.5 GHz.

What is the actual CPI of the processor with just the level 1 caches?

A level 2 cache is added to the processor. The hit time of this cache is 1.5 nanoseconds, and it has a global miss rate of 0.5%. Compute the actual CPI of the processor after the addition of the level 2 cache.

What is the speedup achieved by adding the level 2 cache?

So my first attempt is this:

I-cache miss rate = 1%
D-cache miss rate = 3%
Miss penalty to main memory = main memory access time / clock time
50 ns / .4 ns per clock cycle = 125 clock cycles
Base CPI = 1.4
Load and store = 30% of instructions
Main memory access time = 50 ns
Processor clock = 2.5 GHz

What is the actual CPI of the processor with just the level 1 caches?

Instruction miss cycles = I x 1% x 125 = 1.25 x I
Data miss cycles = I x 30% x 3% x 125 = 1.125 x I
Total number of memory-stall cycles is 2.375 I
Total CPI including memory stalls is 1.4 + 2.375 = 3.775

A level 2 cache is added to the processor. The hit time of this cache is 1.5 ns and it has a global miss rate of 0.5%. Compute the actual CPI of the processor after the addition of the level 2 cache.

Secondary cache access time = 1.5ns
Processor clock = 2.5 GHz
Miss penalty for an access to second-level cache = 1.5ns / .4 ns per clock cycle = 3.75 clock cycles
For a two level cache, total CPI is the sum of the stall cycles from both levels of cache and the base CPI
1.4 + (.005 x 125) + (.005 x .3 x 125) + (.01 x 3.75) + (.01 x .3 x 3.75) = 2.26125
*****This is where I'm most unsure. The "global miss rate of 0.5%" is throwing me off, and I'm not sure I performed this calculation correctly.**

What is the speedup achieved by adding the level 2 cache?
3.775 / 2.26125 = 1.67
***This answer is dependent on the result from the calculation above. If I did that wrong, this answer will be different.

$\endgroup$
3
  • 1
    $\begingroup$ It'd be helpful if you wrote what holds you back form answering this question. What have you tried? Where did you get stuck? We do not want to just do your (home-)work for you; we want you to gain understanding. However, as it is we do not know what your underlying problem is, so we can not begin to help. See here for a relevant discussion. If you are uncertain how to improve your question, why not ask around in Computer Science Chat? $\endgroup$
    – Ran G.
    May 6, 2015 at 22:37
  • $\begingroup$ I completely understand, let me out line what I've tried. $\endgroup$ May 7, 2015 at 2:29
  • $\begingroup$ Sorry, I've added what my first attempt was. My apologies for not outlining it earlier. $\endgroup$ May 7, 2015 at 2:38

1 Answer 1

1
$\begingroup$

You may wish to ask your TA/Prof. what they mean by "0.5% global miss rate". The way I read it is the following:

Out of $x$ accesses to the L2 cache, $0.005x$ accesses are a miss.

This interpretation makes more sense than assuming the miss rate is out of all the store/load instruction.

So each miss of the L1 cache becomes an access to the L2 cache, and out of these, 0.5% are a miss at the L2 cache. From here, it seems you can fill out the details by yourself, building on your own attempts.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.