Test 1 for Computer Architecture | CSCI 4717, Exams of Computer Architecture and Organization

Download Test 1 for Computer Architecture | CSCI 4717 and more Exams Computer Architecture and Organization in PDF only on Docsity! Points missed: _____ Student's Name: __________________________________ Total score: _____ /100 points East Tennessee State University Department of Computer and Information Sciences CSCI 4717 – Computer Architecture TEST 1 for Fall Semester, 2006 Section 001 Read this before starting! • The total possible score for this test is 100 points. • This test is closed book and closed notes • Please turn off all cell phones & pagers during the test. • You may use one sheet of scrap paper that you will turn in with your test. • When possible, indicate final answers by drawing a box around them. This is to aid the grader. Failure to do so might result in no credit for answer. Example: • If you perform work on the back of a page in this test, indicate that you have done so in case the need arises for partial credit to be determined. Binary Hex Binary Hex Power of 2 Equals 0000 0 1000 8 24 16 0001 1 1001 9 25 32 0010 2 1010 A 26 64 0011 3 1011 B 27 128 0100 4 1100 C 28 256 0101 5 1101 D 29 512 0110 6 1110 E 210 1K 0111 7 1111 F 220 1M 230 1G “Fine print” Academic Misconduct: Section 5.7 "Academic Misconduct" of the East Tennessee State University Faculty Handbook, June 1, 2001: "Academic misconduct will be subject to disciplinary action. Any act of dishonesty in academic work constitutes academic misconduct. This includes plagiarism, the changing of falsifying of any academic documents or materials, cheating, and the giving or receiving of unauthorized aid in tests, examinations, or other assigned school work. Penalties for academic misconduct will vary with the seriousness of the offense and may include, but are not limited to: a grade of 'F' on the work in question, a grade of 'F' of the course, reprimand, probation, suspension, and expulsion. For a second academic offense the penalty is permanent expulsion." 1. For each of the following traits of system component design, indicate which implementation method is the best, hardware (HW), software (SW), or firmware (FW), and which is the worst. Best Worst Characteristic _HW_ _SW_ Least expensive during development (2 points) _SW_ _HW_ Least expensive during manufacturing (2 points) _SW_ _HW_ Ease of upgrade (2 points) _HW_ _SW_ Reliability (2 points) Since the performance characteristics of firmware fall between hardware and software, you usually wouldn't see it in one of these "best"/"worst" ratings. 2. For the following items, indicate whether the characteristic more closely describes a top-down (TD) or a bottom-up (BU) design method. (1 point each) _BU_ Best for small production quantities, i.e., we're not going to sell a lot of these _TD_ Doesn't have any excess components, e.g., one goal is to reduce chip count for size _BU_ New design is closely related to long-standing, existing product line _TD_ Cost is not a concern _TD_ There are critical performance goals that need to be met 3. Name an important performance measurement for a video conferencing system? Justify your answer. (3 points) The primary concern with video conferencing is getting huge amounts of data from one point to the other and vice versa. Therefore, throughput is of utmost concern. The CODEC (compression/decompression algorithm) requires a great deal of processing power, therefore MIPS may also be a good measurement. 4. Name an important performance measurement for a PDA? Justify your answer. (3 points) The biggest concerns with PDA's are portability and "rugidization." Portability affects power consumption (directly related to heat dissipation) because of its effects on battery life. Rugidization is important because we need to worry about things like dropping the PDA on the floor and hoping it still works. Some people wrote MIPS as a measurement. This is not nearly as important since MIPS usually relates directly to frequency, and higher frequency means higher power consumption. If, however, you said processing power as it relates to things like handwriting recognition, that would be okay. You don't necessarily need MIPS for handwriting recognition if your instruction set is designed to support it specifically. Speed is also found through compact applications. 18. Describe how the Least Frequently Used cache replacement algorithm works. (3 points) A counter is associated with each line in the cache. Every time a block is loaded, the counter is cleared. The counter is then incremented each time the block is accessed. If the cache needs to decide between multiple lines which one to delete, the one with the lowest "hit" count gets replaced. 19. Assume the processor needs to check the cache for data from the 20-bit memory address 3C45D16 = 001111000100010111012. The cache is a fully associative cache with 210 = 1024 lines and a block size of 23 = 8. Derive the tag from the address that will be used to compare to the tags in the cache. (3 points) In a fully associative cache, no line numbers are used. Therefore, the address is broken into only two parts: the tag and the word id. Since an 8-word block requires 3 word id bits, the lower three bits of the address are the word id. Everything else is the tag. tag 00111100010001011 word id 101 20. Repeat the previous problem for a direct cache instead of fully associative. (3 points) In a direct cache, the middle section of the address identifies the line number in which to store the block. Since there are 210 = 1024 lines, then 10 bits are needed to identify the line number. These bits are immediately to the left of the 3 word id bits. Everything else is the tag. tag 0011110 line number 0010001011 word id 101 The table below represents a small section of a cache that uses direct mapping. Refer to it to answer questions 21 and 22. Assume the processor's memory bus uses 24 bits for an address. Word within block Tag (binary) Line number (binary) 00 01 10 11 100111011010 0101001011 0816 1916 2A16 3B16 110110011000 0101001100 4C16 5D16 6E16 7F16 101101010101 0101001101 5016 6116 7216 8316 010010111101 0101001110 9416 A516 B616 C716 111001110011 0101001111 D716 E916 FA16 0B16 011001111001 0101010000 1C16 2D16 3E16 4F16 Col 0 Col 1 Col 2 Col 3 21. What is the size of the cache in words (bytes)? (2 points) The size of the cache in words is equal to the number of lines times the number of words per line (i.e., block). Since the number of bits in a line number is 10, the number of lines in the cache is 210. Since the number of words per block is 22 = 4, then the size of the cache is: 210 × 4 = 4K Row a Row b Row c Row d Row e Row f 22. A block containing the address 8B453F16 = 1000101101000101001111112 is not contained in the cache. When loaded, which row (a-f) and column (0-3) will its values be stored in? (3 points) Begin by dividing the address into its tag, line id, and word id. Ten bits are needed for the line id and 2 bits are needed for the word id. This gives us: tag 100010110100 line id 0101001111 word id 11 Matching the line id with the lines from the above cache segment reveals its destination to be row e. The word id of 112 = 310 shows us that it will be stored in column 3. 23. Assume a 2-way set-associative cache with 211 = 2048 lines is being used in a system with a 24- bit address using a 16-word block size. If a block containing the address 7BE45316 = 0111101111100100010100112 is stored in the cache, what would the tag be? (2 points) The set associative cache uses a set id instead of a line id to figure out where to store a block. In other words, a block maps to a specific line within a set instead of mapping to a unique line. And just line the line id, the set id comes from the middle section of the address. Since there are 211 = 2048 lines, and since there are 2 lines per set, then there are 210 = 1024 sets. This means that 10 bits are needed to identify the set number. These bits are immediately to the left of the word id bits. Since a block is 24 = 16 words, the right most 4 bits serve as the word id. Everything else is the tag. tag 0111101111 line number 1001000101 word id 0011 24. Assume a memory access to main memory on a cache "miss" takes 40 ns and a memory access to the cache on a cache "hit" takes 10 ns. If 75% of the processor's memory requests result in a cache "hit", what is the average memory access time? (2 points) a.) 17.5 nS b.) 32.5 nS c.) 10 nS d.) 40.0 nS e.) 7.5 nS f.) 2.5 nS First, since we know that not all of the memory accesses are going to the cache, the average access time must be greater than the cache access time of 10 ns. This eliminates answers 'c' and 'f'. Similarly, not all accesses are going to the main memory, so the average must be less than 40 ns eliminating 'd' as an answer. For the exact answer, you need to see that 75% of the time, the access will be 10 ns while the rest of the time (25%) the access time will be 40 ns. This gives us the following equation. (0.75 × 10 ns) + (0.25 × 40 ns) = 7.5 ns + 10 ns = 17.5 ns 25. For each of the following statements, identify which represent write-back (WB) and which represent write-through (WT) cache writing policies? (2 points each) _WB_ When using this write policy with multiple cache-equipped processors sharing a single RAM, it is possible for RAM to contain an incorrect value for a significant period. _WT_ This write policy generates higher write traffic to DRAM. 26. Why does a DRAM use column and row addressing? Circle all that apply. (2 points) a.) It reduces the number of pins needed to interface with the DRAM b.) It's necessary in order to store data to capacitors. c.) It can be used to speed up transfers of sequential data within the same row by using page mode. d.) It allows for simultaneous writing data to and reading data from the DRAM. e.) The cache uses the columns and rows to organize its own data. 27. True or false: The processor may still read from the DRAM while the DRAM is being refreshed. (2 points) 28. True or false: A time critical interrupt should allow nested interrupts. (2 points) Any process that is critical, especially one that is time critical, should not be interrupted until it is finished. Yes, there might be interrupts that are of a higher priority, and if those exist, the designer should take precautions to allow those interrupts to nest, but in general, no nesting should occur here. 29. Name and describe one of the four types of interrupts, i.e., what types of interrupt triggers are there, that we discussed in class. (2 points) The following is copied directly from the notes: • Program – Something that occurs as a result of program execution such as illegal instructions, arithmetic overflow, divide by zero, or memory handling error • Timer – Generated by one of the processor's internal timers so that the processor can perform some time-scheduled task • I/O – Generated by an I/O controller to request service from the processor such as keyboard, mouse, NIC, disk drive • Hardware failure – signifies some error condition with the hardware 30. Match the DRAM type with its primary characteristic. (5 points) Fast Page Mode overlaps data read with column address write for next read Extended Data Out uses fixed row address for multiple column reads SDRAM requires adherence to exact bus specification DDR-SDRAM incorporates a cache on the memory stick RAMBUS returns data to processor synchronized to a full clock pulse CDRAM returns data to processor synchronized with both the rising and falling edges of a clock pulse 31. For single error correction, what is the minimum number of check bits required for 12 data bits? (2 points) a.) 4 b.) 5 c.) 6 d.) 7 e.) 8 f.) 9 g.) 10 h.) 11 i.) 12 For single error correction, the number of data bits, M=12, and the number of check bits, K must satisfy the expression M + K < 2k – 1. The following works this expression out for a few of the values of K to show where the minimum one is.