Disk Scheduling Algorithms Simulation Lab

Transcription

1 Disk Scheduling Algorithms Simulation Lab In this lab, we will start to look at the various disk scheduling algorithms which we discussed in the course. Here is a brief summary of the algorithms: 1. FCFS (First Come, First Served) o perform operations in order requested o no reordering of work queue o no starvation: every request is serviced o poor performance 2. SSTF (Shortest Seek Time First) o after a request, go to the closest request in the work queue, regardless of direction o reduces total seek time compared to FCFS o Disadvantages starvation is possible; stay in one area of the disk, if very busy switching directions slows things down 3. SCAN o go from the outside to the inside servicing requests and then back from the outside to the inside servicing requests o repeats this over and over o reduces variance compared to SSTF 4. LOOK o like SCAN but stops moving inwards (or outwards) when no more requests in that direction exist 5. C-SCAN (circular scan) o moves inwards, servicing requests until it reaches the innermost cylinder; then, jumps to the outside cylinder of the disk without servicing any requests o o repeats this over and over variant: service requests from inside to outside, and then skips back to the innermost cylinder 6. C-LOOK o moves inwards, servicing requests until there are no more requests in that direction; then, it jumps to the outermost outstanding requests o o repeats this over and over variant: service requests from inside to outside, then skips back to the innermost request The Saylor Foundation 1

2 For this exercise, we will run a Java disk scheduling simulator. Please make sure that you have at least the Java runtime environment setup. The simulations can be run on any platform that supports Java. For the purposes of this exercise, we will focus on running this on a Windows PC. Initial Setup 1. If you do not have Java installed, please download and install* the Java runtime environment. 2. Create a folder called Rec10 on your computer. 3. You may read through the University of Texas at San Antonio s user s guide* if you would like. Download disk.zip by clicking on the link labeled disk.zip. 4. Extract the zip file to the Rec10 folder. 5. If you are running Windows, start an MS-DOS command window (choose run from your Windows start menu and type command in the window). 6. Once you have an MS-DOS command window, type cmd/e. 7. Switch to the Rec10 folder. The command to do this is: cd \Rec10 *Terms of Use: Please respect the copyright and terms of use on the webpages linked above. Getting Familiar with the Simulator 1. Execute the rundisk batchfile. 2. You will see the main simulator window which should look like Figure 1. Click on the quit button in the upper right hand corner to exit the simulator. Figure 1 The Saylor Foundation 2

3 3. Go through the simulator tutorial that you can find in the directory that you expanded disk.zip into. Edit the diskheadconfig file so that it has your name instead of New User. 4. In order to perform a new experiment, we need to edit the diskheadconfig file and change myrun to myrun1 and myexp to myexp1. Your diskheadconfig file should look similar to Figure 2 Figure 2 5. Copy myrun.run into myrun1.run and myexp.exp into myexp1.exp. 6. Edit myrun1.run, changing the first line to name myrun1. Change the second line to reference myrun1 instead of myrun. 7. Edit myexp1.exp, changing the first line to name myexp1 and change all occurrences of myrun to myrun1. 8. Run the simulator. You should get the same results are before. 9. Let s make a few changes to the experiment. o Change 30 in first nextblock line to 25. This is a probability distribution for o generating the next block. The next block can be between 25 and 40. In the first numblocks line change 20 to 30. This tells the simulator how many blocks to request to generate from a distribution. o In the second interarrival line, change 5 to 3 and 15 to 10. The line should now read: interarrival uniform 3 10 This specifies that there should now be an interarrival time somewhere between 3 and 10. o In the second numblocks line, change 4 to 20. The Saylor Foundation 3

4 10. Run the simulator again. When the run is done, click the Show Table Data button. Make sure there were 3 runs, each having 50 seeks. (Check the Count column.) You should see something similar to figure 3. Figure To help understand what is happening a bit more, let s look at the actual data. Your results might by a bit different, since this is a random distribution. To show this, click on Show Run Data Table (see Figure 4). Figure 4 All of the data can be found in Appendix A. For the purposes of our discussion, I have included a portion of each table. Let s first look at FCFS, a portion of which is shown in Table 1 below: The Saylor Foundation 4

5 Block Time From To Request Start End Seek Turnaround Table 1: FCFS Notice how the first request starts at time 2.23 and ends at The starting block is 4, and the ending block is 33. The next request comes in at 3.12 but does not start until 4.13, since it has to wait until the first request is finished. First come, first served (FCFS) is going to simply take each request as it comes in and service it. All servicing will be done sequentially. If you look at the full table in Appendix A, you will find the statistics for this part of the experiment at the bottom of the table. Let s now look at STTF, which is a bit more complicated. Block Time From To Request Start End Seek Turnaround Table 2: SSTF Once again, we need to pay attention to the request time. Obviously, if a request has not been received yet, there is no way for the algorithm to determine that the seek time to that request is shorter. Looking at the second request, notice that the algorithm starts at 33 and ends at 117. The third request starts at 117 and ends at 37. You might wonder why we just did not go from 33 to 37 and then eventually to 117. However, notice that at the start time (4.13), the request for service to block 37 had not been received, so the algorithm had no knowledge of a closer request. As you look through the table data after running experiments, please keep track of the request time as well as the start time, since these are very The Saylor Foundation 5

6 important when determining the order of servicing requests. The full SSTF table can be found here. Finally, let s take a look at the table for the C-LOOK algorithm. I ve included a bit more data in Table 3 in order to help you understand exactly how the algorithm is working. The full table is available for viewing here. Block Time From To Request Start End Seek Turnaround Table 3 To make this easier, let s place the requests in order of arrival using the from block: 4, 33, 117, 36, 32, 37, 30, 35 Table 4 breaks down the servicing of requests as well as the requests in the queue for each point in time that we need to examine. Req No From To Request Start End In Queue In Queue Start at Finish ,4, ,5 4,5,6, ,5,6 4,5, ,6,8 4,6, ,6 4, Table 4 The Saylor Foundation 6

7 Let s discuss what is happening at each request: Req No 1 Since the service time ends at 4.13, request 2 is placed in the queue for waiting requests. 2 By the time that this service request has finished, requests 3, 4, and 5 have arrived and are in the queue. Notice that the disk head is sitting at block 117. While requests 4 and 5 are closer to 117 (37 and 35), the C- LOOK algorithm specifies that the disk head returns to the furthest request and then starts to work back towards the inside of the disk, so request 3 will be serviced next. 3 Requests 4 and 5 are in the queue when the servicing of this request starts. During the servicing, 6 and 7 also arrive in the queue. Since request 7 requires a move to block 32 and the disk head is moving inward, this will be serviced next. 7 Requests 4, 5, and 6 are in the queue at the beginning of servicing request 7 as well as at the end. Since request 5 requires a move to block 35, this will be the next request serviced. 5 Request 8 comes in just about the time that request 5 starts, so 4, 6, and 8 are in the queue. Since 8 only requires a move to block 36, this will be serviced next. 8 Request 4 and 6 are in the queue. If you look at the entire table of data in the appendix, you will see other requests that would be in the queue. However, we are just looking at a small segment of the data. Since request 4 only requires a move to 37, this request will be serviced next. 4 At this point, in our subset, only request 6 is left, so this will be serviced next. 6 Request 6 is now serviced. There are additional requests serviced beyond 6, but for the purpose of our discussion, we have only looked at processing through request 6. Table 5 The Saylor Foundation 7

8 12. Let s take a look at the summary data for each algorithm: Blocks Traveled in one Request Algorithm: FCFS Seek Time Turnaround Time Mean Minimum Maximum Standard Deviation Blocks Traveled in one Request Algorithm: SSTF Seek Time Turnaround Time Mean Minimum Maximum Standard Deviation Blocks Traveled in one Request Algorithm: C-LOOK Seek Time Turnaround Time Mean Minimum Maximum Standard Deviation Table 6 As you can see, SSTF and C-LOOK are pretty close as far as mean seek time. However, SSTF has a shorter turnaround time (the time from which a request enters the queue until it has been serviced.) As such, SSTF would be the best algorithm in this case. The Saylor Foundation 8

9 13. As a final step to familiarize yourself with the simulator, click on the Show Run Data Graph button. Make sure that you choose all runs. The graphs should be saved automatically to the log file. In the next section of the lab, you will look at the log file to analyze your results. Figure 5 The Saylor Foundation 9

10 Running an Experiment 1. Create a Rec11 directory, and do the following to create a separate set of files: o Copy the files from your Rec10 directory into this one. o Edit diskheadconfig, and change myrun1 to myrun2 and myexp1 to myexp2. o Copy myrun.run to myrun2.run and myexp.exp to myexp2.exp. o Edit the lines of these files so that they refer to myrun2 and myexp2. Be sure to change all of the run line of myexp2.exp. 2. You should now change myrun2.run so that a set of 250 requests with a first arrival time of 0 is used for the run. You also will want to modify the interarrival time so that it is constant at 10 and the NextBlock is uniform between 0 and 260. Please complete the following steps: o Delete the last four lines of myrun2.run. o Change the interarrival line form constant 2 to constant o Change the nextblock line from to o Change the firstarrival from 2.23 to 0. o Change numblocks from 20 to 250. o Change the layout line from uniform 2 to uniform 1. o Change the movement line from linear 0.5, 0.10 to linear Remove all but the last run line from myexp2.run. This line refers to the CLOOK algorithm, which we use for this experiment. 4. Check to make sure that the simulator will run with 250 requests. 5. Let s now take a look at how bad blocks affect disk performance. First, a little background. In order to account for bad disks, most disk drives today transparently map any disk blocks marked as bad to some special blocks on the disk that are reserved for that purpose. This makes the disk appear to have no bad blocks. For example, if the reserved blocks are on cylinder 245, then a request for a bad block on cylinder 30 will automatically go to cylinder 245. Because the operating system thinks that the disk in on cylinder 30, it will use this to make some decisions about where to move next. Because of this, there might just be some extra head movement. Let s now run some experiments to determine how bad blocks can impact disk performance. 6. Add the following two lines to myrun2.run: badfraction 0.0 baddestination constant 245 The first line tells us that there are no bad blocks, and the second line tells all that cylinder 245 has been reserved as a space to redirect bad blocks to. 7. Let s run the simulator again. What results do you obtain? Why? The Saylor Foundation 10

11 8. Now add a second run line to myexp2.exp. Copy the run line and add to the end of it: badfraction 0.1 This tells the simulator that every 10 th seek will be to cylinder 245, since 10 percent of the blocks are bad. 9. Let s run the simulator again. What do you find? Why? 10. Let s experiment with the fraction of bad blocks. Experiment with several values of the fraction between 0 and 0.1. As the fraction increases, does the turnaround time increase slowly, or is there a value above which it increases rapidly? 11. Now modify myexp2.exp so that it contains about 5 runs with the values of badfraction that you think are most instructive. Create a log file. Write a short paragraph summarizing your results, and put it at the end of the log file. 12. Compare the results of your log file with the answer key. The Saylor Foundation 11

12 Appendix A: Show Run Data Table FCFS Block Time From To Request Start End Seek Turnaround The Saylor Foundation 12

13 Mean Minimum Maximum Standard Deviation The Saylor Foundation 13

14 Block myrun1 SSTF Time From To Request Start End Seek Turnaround The Saylor Foundation 14

15 Mean Minimum Maximum Standard Deviation The Saylor Foundation 15

16 Block myrun1 C-LOOK Time From To Request Start End Seek Turnaround The Saylor Foundation 16

17 Mean Minimum Maximum Standard Deviation The Saylor Foundation 17