OPERATING SYSTEMS ECSE 427/COMP SECTION 01 WEDNESDAY, DECEMBER 7, 2016, 9:00 AM

Transcription

1 December 2016 Final Examination OPERATING SYSTEMS ECSE 427/COMP SECTION 01 WEDNESDAY DECEMBER :00 AM Examiner: Assoc Examiner: Student Name: Answers McGill ID: INSTRUCTIONS: This is a CLOSED BOOK examination SPACE IS PROVIDED on the examination to answer Qs # 1-20 (Part I) & #1-4 (Part II) You are permitted TRANSLATION dictionaries ONLY NO CRIB SHEET permitted STANDARD CALCULATOR permitted ONLY This examination is PRINTED ON BOTH SIDES of the paper This examination MUST BE RETURNED This exam includes 11 PAGES (including the cover page) and has 24 QUESTIONS Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 1 of 11

2 node - Part I Short Answers Provide the answers in the allotted space Answer all questions Each question is worth 3 points VERY IMPORTANT: Provide precise and up to the point answers Long and rambling answers will receive 0 marks 1 Consider a directory /users/robert/os/assign in a UNIX file system Suppose the root directory is already loaded into the memory Determine the number of disk accesses that are necessary to translate the given directory to i-node Assume all directories in the path are small and can fit into a single disk block For example the /users directory can fit into a single data block - i lasers ; data lasers ; inode1users1robertjdak1useos1robertji-nodetusers1r1osjdnk1user1r1os@iisfotaclzx 2 Consider a UNIX file system that uses an i-node structure discussed in the class Suppose the disk block size is 1024 bytes Assume that a disk block pointer is 8 bytes long What is the size of the largest file that can be created in the system? 1024 t (102-84) 1024 t [02-84] A file is removed using the rm command in a UNIX file system What is the minimum number of disk accesses that are needed to carry out this command? Carefully state all the assumptions you make If you write down a number without the assumptions it won t be accepted as the correct answer Dir containing the file Is loaded l update I - at least 3 writes node 4 Consider the following operations open() close() write() read() and seek() associated with a file in a UNIX file system Group these operations based on the minimum number of disk accesses needed to execute them That is when you apply one of those operations on a file it is going to create certain number of disk accesses Find the minimum number of disk accesses and use those numbers to group them It is very important that you state all assumptions you make regarding the state or structure of the file system (eg what is already loaded in memory) I - update directory ; ; update free bitmap ; days }se Intec gknc ) a ready writer ) sizes dir Final Exam - ECSE 427/COMP 310 (Fall 2016) already Page 2 of 11 ( within the block ) loaded multiblock

3 ' 5 Consider the SCAN disk scheduling algorithm Suppose the disk scheduler receives as data requests and the disk read/write head is at track 60 and moving upwards (towards higher numbered tracks) and the disk has tracks numbered 0 to 199 What is the total number tracks crossed by the SCAN algorithm? Suppose the requests arrived in two batches: { } and { } What is the total number of tracks crossed by the SCAN algorithm? For the second part assume that the read/write head is at 60 and moving upwards at the beginning of processing the two batches Go -> tracks in total /5 tracks ; treks 32g ; 188 tracks Use the SSTF (Shortest Seek Time First) algorithm with the previous disk scheduling problem What is the total number of tracks crossed? start at ( ; = Consider a system with 4 processes: P1 P2 P3 and P4 There are 3 types of resources: X Y and Z in the system The following are the matrices denoting maximum resource requirement and resources held by the different processes respectively Also the currently available resources are given by the availability vector Determine whether the system is safe or not If it is safe is it very safe? P P Max = 2 [ ] P P X Y Z Need = Hold = ( 4g ) P P 2 [ ] P P X Y Z Avail = [ 3 3 3] system is safe first Py Any and can run then Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 3 of 11

4 - the 8 If the above state is safe what is the largest request that could be launched by P1 that would be accepted by the Banker s algorithm? The size of the request is 3*x+2*y+z where x is the number of X type of resource y is the number of Y type of resources z is the number of Z type of resource respectively If the state is not safe minimal addition (computed the same way as above) that will make the system very safe Largest request request is [o / 1] from P system is still safe 9 State the four necessary conditions for deadlock to happen in a system with multiple processes and resources 1 Mutually exclusive 2 Hold and wait access 3 No rollback 4 Circular wait 10 Under what situations are the above conditions not sufficient? With multiple units of a resource the above is not sufficient 11 Consider four jobs that arrive at the scheduler at time 0 with the following burst (run time) requirements: J1(10) J2(15) J3(20) J4(8) If round-robin (RR) is used as the CPU scheduling policy what would be the completion times of the different jobs? Assume RR quantum of 5 J : 25 Jz : 43 : 53 Jg : 38 Jy Different at#asai:rat quantum sizes attempts Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 4 of 11 Cmp times

5 12 If shortest job first (SJF) is used as the scheduling policy what would be the completion times of the different jobs? Titles ) J ( 18 ) Jz( 33 ) Jg (5 3) 13 If first-come first-served (FCFS) is used as the scheduling policy what would be the completion times of the different jobs? Assume that the jobs arrived at time 0 in the order J1 J2 J3 and J4 J Go ) Jz( 25 ) JB (54 5) J4± 14 Consider a Multi-Level Feedback (MLF) queue scheduler where the top level queue has a time quantum of 1 unit and the next level queue 2 units and the following one with 4 time units The last queue uses FIFO as the scheduling discipline The MLF algorithm services a higher level queue (queue with shorter time quantum) if it has a job and services the next queue only after completing all jobs at the top level It follows the same strategy with all the levels The MLF algorithm does not preempt That is if the algorithm is servicing the 4-unit queue and a new job arrives to the top level it does not preempt the currently servicing 4-unit queue It completes the servicing of the 4-unit queue and switches to the top level queue at the next schedule cycle Suppose you are given the following jobs show the execution trace: (0 4) (1 3) (4 5) (5 3) The first number is the arrival time and the second number is the required execution time in the time units J ( 04 ) j Tzcl 3 ) ; Jz ( 457 ; JYCIB ) durate r I (g) Tgs#$T3CaHJaCEDH EGAD Mf µ d t wwws#dnbenkshmkad )wm5c Haas at qtz qq t t d done {3 done Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 5 of 11

6 15 State the Amdhal s law in words Note that Amdhal s Law is not the speedup equation in Improvement performance of an enhancement is limited by the fraction of time the ebnhauawet can 16 Consider an application that has a serial part and a parallel part The runtime of the serial part cannot be improved by mapping into multi-core machine (ie parallel machine) The parallel part can run on a parallel machine with up to eight processors The performance of parallel part improves linearly with the number of processors That is doubling the number of processors reduces the runtime by half The total runtime of the application in a serial machine is 600 seconds The serial fraction of the application is about 20% The parallelization overhead is half of the serial part and is independent of the number of processors The parallelization overhead itself cannot be parallelized What will be runtimes of the application in a four core and eight core machines respectively? St be applied p = 600 S = = 60 On 4 processors On 8 = 120 t processors : = t 60 = A binary Buddy algorithm is used to management a block of memory that is 64 KB in size Suppose the memory system is receiving the following requests: allocate A (8 KB) allocate B (4 KB) allocate C (8 KB) allocate D (2 KB) Show the state of the memory a lb EE#hfEFIo unallocated B allocated 18 Now take the state of the memory given in the above problem and run the following commands: deallocate A deallocate B Show the state of the memory 8 TIED D teach Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 6 of 11

7 19 Consider a logical memory system that uses segmentation The logical address is 32 bits long Suppose the segment selector is 6 bits long What is the largest segment size you could have in the memory system? segment offset =3 2-6=26 Largest seg size = 226 bytes 20 You have an application that needs to work in an embedded device with constrained memory You ask several developers to create implementations with different algorithmic designs You test two of the best implementations and get the following traces for the logical addresses accessed by them Implementation A: Implementation B: Which implementation would you select and why? waking set (A) = 4 51%126 waking set (B) = 12 34> >1012 settee A because the Watne set is Smaller Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 7 of 11

8 Part II Long Answers Your answers should still be up to the point Provide the answers in the allotted space Answer all questions 1 [10 points] Consider a virtual memory system with logical (virtual) addresses that are 36 bits long The system uses 2048 byte pages The size of a page table entry is 8 bytes How many levels are needed in the paging tables? 4 levels Show the allocation of address bits for the different components of the paged logical (virtual) address k -3 Sketch the organization of the page tables and show how the different portions of the logical (virtual) address are used to index the different tables and eventually find the physical address Mark all important components including how the CPU points to the page tables Ethniki t I## #i top level page table Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 8 of 11 (page directory )

9 2 [10 points] Provide the pseudo code for the following operations in a simple UNIX like file system: open() and write() You can assume that the open() and write() have similar input and output parameters as you encountered in the programming assignment Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 9 of 11

10 3 [10 points] Develop a semaphore-based solution for the Dining Philosophers problem You need to recall the Dining Philosophers problem as discussed in the course material Your solution should have no deadlock and have maximum concurrency That is as many philosophers as possible should eat at the same time State all assumptions Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 10 of 11

11 4 [10 points] Develop a monitor-based solution for the bounded buffer producer-consumer problem You monitor should provide a method for the producer call when it produces an item and the consumer to call when it wants to consume an item Your solution should work with multiple producers and consumers If no consumers are there producers should block after filling up the buffers Similarly the consumers should block after emptying the buffers State all assumptions Final Exam - ECSE 427/COMP 310 (Fall 2016) Page 11 of 11