13.9 Exercises


  1. Consider the system specified by Table 13.1. Solve the system performance model using Bard's approximation, as shown in Eq. (13.5.8). Compare the results with the exact ones obtained by Schweitzer's approximation. What are your conclusions?

  2. Consider the transaction system of Section 13.3. Calculate the performance measures of that model using the approximate MVA algorithm. Assume that the stopping criterion is to have a maximum difference of 0.001 for successive values of ni,r. Suppose now that the number of update transactions in the system is tripled. Recalculate the model's results using the exact and approximate techniques. Compare the computational effort required by the two algorithms.

  3. In the example of the transaction system of Section 13.3 it is noted that query transactions only make use of disk 1, which increases its utilization and turns it into the bottleneck. Having observed this problem, the support analyst wants to know what would be the effect on performance if the I/O load due to query transactions were balanced among the two disks. Compare the results obtained with the current situation.

  4. A database server has one processor and two disks, D1 and D2. The workload is divided into three classes: query (Q) transactions, update (U) transactions, and interactive (I) users. Table 13.12 gives the input parameters for these classes.

    Table 13.12. Parameters for Exercise 13.4

    Class

    Dcpu

    DD1

    DD2

    graphics/373fig01.gif

    lr

    Mr

    Zr

    Q

    0.06

    0.030

    0.06

    5

    3.0

    -

    -

    U

    0.10

    0.030

    0.09

    3

    1.5

    -

    -

    I

    0.09

    0.045

    0.00

    5

    -

    50

    15

    Use the QN solvers provided with the book to answer the following "what if" questions:

    1. What is the average response time for each class?

    2. What is the impact on response time if the arrival rate of query transactions is increased by 95%?

    3. In the scenario with an increased arrival rate of query transactions, consider the following hardware upgrades and compare the performance improvements obtained with each one of them.

      • replace disk D1 by one twice as fast.

      • replace the CPU by one twice as fast.

    4. With the increased arrival rate for query transactions and with a twice-as-fast processor, draw a graph of response time versus the number of simultaneous clients when this number varies from 50 to 250. What is the maximum number of simultaneous clients that can be supported to keep the response time below 1.5 sec?

  5. A database server has one CPU and one disk. The server's workload is composed of trivial queries that arrive at a rate of 10 tps, complex queries that arrive at a rate of 0.1 tps, and of a batch workload that generates a report. When the report generation completes, a new report generation is started in 15 minutes. Table 13.13 provides workload related information. Each physical I/O demands 0.015 msec of CPU time and 9 msec of disk service time. The last row of Table 13.13 indicates how much CPU time is required by transactions of each workload in addition to the CPU time related to I/Os.

    • Find the average response time and the average throughput for each of the three workloads.

    • Find the utilization of the CPU and of the disk.

    • Finds the residence times at the CPU and at the disk for each of the three workloads.

    Table 13.13. Data for Exercise 13.5
     

    Trivial

    Complex

    Report

    Avg. Number of SQL Calls

    3.5

    20.0

    120.0

    Avg. Number of I/Os per SQL Call

    5.0

    15.0

    40.0

    DB Buffer Hit Ratio (in %)

    70.0

    80.0

    30.0

    Non I/O Related CPU Time (msec)

    30.0

    180.0

    1250.0



Performance by Design. Computer Capacity Planning by Example
Performance by Design: Computer Capacity Planning By Example
ISBN: 0130906735
EAN: 2147483647
Year: 2003
Pages: 166

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net