|
|
|
|
|
|
|
distributed over possible locations on the track. What is the total time to access and transfer the requested data? |
|
|
|
|
|
|
|
|
2. In a certain system, requests to a disk system (from multiple buffered sources) arrive at the rate of one each 33 ms (l). The disk has service time of 20ms. Use the open-queue M/G/1 model (c2 = 0.5) to find: |
|
|
|
 |
|
|
|
|
(a) Total response time (service time plus the waiting time). |
|
|
|
|
|
|
|
|
(b) Average number of requests queued at the disk awaiting service. |
|
|
|
|
|
|
|
|
3. Plot the effect of c2 on waiting time for various values of r (r = 0.1, 0.2, 0.5, 0.7, 0.9). At what point(s) does c2 contribute more than a 10% effect on total response time (Tw + Ts)? |
|
|
|
|
|
|
|
|
4. Repeat study 9.2(a) with the following changes: c2 = 0.5, Tuser = 20ms (i.e., 200K user-state instructions before an I/O request), and n = 3. |
|
|
|
|
|
|
|
|
5. Now find the effect of a four-times-faster processor; that is, repeat study 9.2(b) using the parameters given for the previous problem, except that c2 = 1. |
|
|
|
|
|
|
|
|
6. Study 9.4 assumes that using n = 2 (two processes) residing in the same memory will double the relative paging memory miss rate and hence double the I/O request rate. At that point, multiprogramming (n = 2) decreases performance. At what relative paging memory miss rate (for n = 1) will the user performance for both n = 1 and n = 2 be the same? That is, find Tuser (n = 1) so that the user time per second n = 1 and n = 2 are the same. |
|
|
|
|
|
|
|
|
7. In study 9.3, suppose we increase the server processor performance by a factor of 4, but all the other system parameters remain the same. Find the disk utilization and user response time for n = 20. (Assume c2 = 0.5 for the disk.) |
|
|
|
|
|
|
|
|
8. A RAID technique is to be applied to a disk array of 18 drives (including correction drives). Suggest suitable redundancy configuration for drives with data distributed as: |
|
|
|
|
|
|
|
|
(a) (4,1). |
|
|
|
|
|
|
|
|
(b) (1,8). |
|
|
|
|
|
|
|
|
(c) (1,1). |
|
|
|
|
|
|
|
|
9. Find the effect of workload on TR using the open-queue model. Repeat study 9.5 using the alternative workload (workload #2) described in study 9.6. |
|
|
|
|
|
|
|
|
10. Suppose we now apply the "square root" server (from study 9.6) to study 9.5. Again using the data of study 9.5 and the open-queue model, find TR for each configuration if  |
|
|
|
|
|
|
|
|
11. Reconsider study 9.6. Now assume |
|
|
|
|
|
|
|
|
(a) m = 1, |
|
|
|
|
|
|
|
|
(b) requests are uniformly distributed across servers. |
|
|
|
|
|
|
|
|
Evaluate la for workload #1 for (1,1), (16,1), (1,16), and (4,4). |
|
|
|
|
|