CSE 378 - Machine Organization & Assembly Language - Winter 2009

HW #4

1. Here is a series of address references given as word addresses: 7, 13, 1, 28, 25, 17,
48, 49, 3, 13, 1, 9, 4, 27, 28, and 25.
a. Assuming a Direct-mapped cache with 16 one-word blocks that is initially
empty, label each reference in the list as a hit or a miss.
b. Now assume that the cache is a 2-way set associative, one-word blocks,
total 16-word capacity.

If there is a cache-miss, indicate which of the following types of miss it is

(compulsory/conflict (with what)/capacity). As a brief refresher, here are the
definitions:
compulsory miss - miss caused by the first access to a block that has never
been in the cache.
conflict miss - miss where multiple blocks compete for the same set.
capacity miss - miss caused when the cache cannot contain all the blocks
needed during execution of the program.

For each of your answers to both parts a & b fill in the table shown below. Show
the final contents of the cache in each case.

Assume a Least Recently Used replacement policy where applicable.

(a)

Address Hit/Miss Type

7
13
1
28
25
17
48
49
3
13
1
9
4
27
28
25
(b)

Address Hit/Miss Type

7
13
1
28
25
17
48
49
3
13
1
9
4
27
28
25
2. You've been tasked to upgrade a mission critical function in a visiting
computational linguistics professor's code provided here as a tarball (or here if
you prefer ZIP files). The function (as defined in find_letter.h) searches for a
string with a given letter somewhere in it. The current version runs too slowly.
We think it may have something to do with locality, but we're not certain.

Your tasks:
1. Rewrite the function to run faster.
2. Explain why the old function was slow.
3. Explain why yours is faster.
4. Explain in what circumstances your code would be slower than the provided
code.

Notes on the provided code:

Provided is some code to load the specially formatted files that the professor likes
to use in his research. You are not expected to grok any code outside of the
function you are to rewrite. Also provided is a simple main function for testing
purposes.

The file format is as follows:

int (describes how long the longest string is)

words
words
...
lastword

The file words.txt is a simple example of this type of file.

Due to the...odd string format that the professor uses, you may not use any of
the standard c string library functions. The format he uses allows embedded nulls
for some reason.
3. You are the lead engineer on CezeCorp’s debug team, and are testing the
company’s first processor that uses data caching. After finding that the processor
fails to meet its performance targets, you discover that the wires carrying the 32-
bit memory address from the Memory Access pipeline stage to the cache have
been reversed:
pipeline_address[31] connects to cache_address[0]
pipeline_address[30] connects to cache_address[1]
pipeline_address[29] connects to cache_address[2]
…
pipeline_address[0] connects to cache_address[31]

Since this reversal is consistent for loads and stores, the processor continues to
function correctly (although with intolerably slow performance). Why does this
wiring mistake decrease the processor’s performance?
4. As we saw in class, page tables require fairly large amounts of memory, even if
most of the entries are invalid. One solution is to use a hierarchy of page tables.
The virtual page number can be broken up into two pieces, a “page table number”
and a “page table offset,” which is described in the figure below.

(From Hennessy and Patterson’s Computer Organization and Design, 3rd Edition, pg 513)

The page table number can be used to index a first-level page table that provides a
physical address for a second-level page table. The page table offset is used to
index into the second-level page table to retrieve the physical page number.
One way to arrange such a scheme is to have the second-level page tables occupy
exactly one page of memory.

Assuming a 32-bit virtual address space with 4KB pages and 4 bytes per page
table entry, how many bytes will each program need to use to store the first-level
page table (which must always be in memory)? Provide information on how to
decode the address (i.e. what bits for the page table number, pagetable offset and
physical page offset). Explain your solution.
5. Do you see any problems with the indicated decoding of addresses for caches? If
so, what is it?

Index Tag Offset

31 0
6. With respect to TLBs, indicate whether the three fields (Tag, Index and Offset)
shown here should be Physical or Virtual addresses and also state the reason. For
your answer, fill in the table provided below.
Tag Index Offset

Table:

Field Physical/Virtual Reason

Tag

Index

Offset
7. Virtual Memory and TLBs:

a. Given the following stream of 16-bit 32-bit virtual addresses, update the
TLB and Page Table shown below. Assume 4 KB pages, a four-entry fully-
associative TLB, and true LRU replacement. If pages must be brought in
from disk, increment to the next largest page number.

Addresses: 4095, 31272, 15789, 15000, 7193, 4096, 8912

TLB
Valid Tag Physical Page
1 11 12
1 7 4
1 3 6
0 4 9

Page Table
Valid Physical page or in Disk
1 5
0 Disk
0 Disk
1 6
1 9
1 11
0 Disk
1 4
0 Disk
0 Disk
1 3
1 12
0 Disk
1 3
1 7
0 Disk

b. Repeat the question, but this time assume that pages are 16 KB (instead of 4
KB).

c. What would be an advantage of having a larger page size? Are there any
disadvantages?

d. Show the final contents of the TLB if it is two-way set associative (and pages
are 4 KB).
8. I/O: Given the following two applications 1) A database for storing large image
files, 2) a video game, answer these questions about appropriate disk usage:

a. Is decreasing the sector size likely to help or hurt this application?

(explain)

b. Would increasing disk rotation speed help or hurt this application?(explain)

9. I/O: Assume a hard disk has the following specifications:
– An average seek time of 11 ms
– A 7200 RPM rotational speed
– A 30MB/s average transfer rate
– 3 ms of overheads for making a request

a. Given that sectors are 1024 bytes, what is the average time to read or write
a 1024-byte sector?

b. Now assume that sectors are 2048 bytes. What is number of random 2048-
byte sectors that can be read in one second?
10. Performance: Our two favorite programs have the following distribution of
instructions of type A, B, C, and D, and our current processor has the following
CPI for those instructions:

A B C D
Fraction of program 1 20% 15% 55% 10%
Fraction of program 2 30% 30% 25% 15%
CPI 5 4 3 2

a. We have heard there is a new processor that implements Type B instructions

blazingly fast – each one now executes in only one cycle. What is maximum
performance improvement we might expect from this new processor for
program 1 and program 2?

b. What would be our performance improvement if instead we just made our

original processor run twice as fast? For program 1? For program 2?