U. Glaeser

In the case of Fig. 8.15, the model with 1 HSC50 disk controller yields e min = 1.53%, and for δ = 0.2
we have:
Finally, the predictive power in the case with two HSC50 disk controllers is much better: we have e min =
0.64% yielding
Conclusions
Even in cases of fully controlled simple synthetic workloads, the performance modeling of disk subsystems
is not a simple task. Basic popular load independent queuing models (load independent convolution
algorithm or MVA) cannot be used for modeling modern disk subsystems, which use caching and disk
access optimization. The use of load dependent MVA models is rather efficient and we proposed models
for optimized disk access, cached disk access, and combined optimized and cached disk access. Presented
models need a careful calibration procedure. In the majority of cases based on symmetric random disk
accesses the modeling errors for optimized disk access were less than 2%. In more complex cases with
optimized disk accesses and caching, our models regularly achieve errors below 10%. Experimental
verification of our models has been successfully performed in both VAX/VMS and PC/NT environments.
References
( 1 + δ )emin 1.84%, p ∗ = ( 0.2)
= 0.7, q( 0.2)
= 35.3%
( 1 + δ )emin 0.77%, p ∗ = ( 0.2)
= 0.35, q( 0.2)
= 76.5%
1. Basket, F., K. Chandy, R. Munz, and F. Palacios, Open, Closed, and Mixed Networks of Queues with
Different Classes of Customers. Journal of the ACM, Vol. 22, No. 2, pp. 248–260, April 1975.
2. Buzen, J.P., Computational Algorithms for Closed Queueing Networks with Exponential Servers.
Communications of the ACM, Vol. 16, No. 9, pp. 527–531, September 1973.
3. Dujmovic, J.J., Multiprogramming Efficiency Analysis for Computer Evaluation and Selection Studies.
Proceedings of the 11th International Symposium Computer at the University, 1989.
4. IBM, Hard disk drive specifications—Deskstar 60GPX. S07N-4780-02, Publication #2818, February 1,
2001.
5. Jain, R., The Art of Computer System Performance Analysis. John Wiley, New York 1991.
6. Lee, E.K. and R.H. Katz, An Analytic Performance Model of Disk Arrays. Sigmetrics ’93 Proceedings,
pp. 98–109, 1993.
7. Menasce, D., V. Almeida, and L. Dowdy, Capacity Planning and Performance Modeling. Prentice-Hall,
1994.
8. Nelder, J.A. and R. Mead, A Simplex Method for Function Minimization. The Computer Journal, Vol. 7,
No. 4, pp. 308–313, 1965.
9. Ng, S.W., Advances in Disk Technology: Performance Issues. Computer, pp. 75–81, May 1998.
10. Reiser, M. and S.S. Lavenberg, Mean-Value Analysis of Closed Multichain Queueing Networks. Journal
of the ACM, Vol. 27, No. 2, pp. 313–322, 1980.
11. Ruemmler, C. and J. Wilkes, An Introduction to Disk Drive Modeling. Computer, pp. 17–28, March
1994.
12. Shriver, E., Performance Modeling for Realistic Storage Devices. PhD Thesis, Department of Computer
Science, New York University, May 1997.
13. Silbershatz, A. and P.B. Galvin, Operating System Concepts. Addison–Wesley, Reading, MA 1994.
14. Worthington, B.L., G.R. Ganger, and Y.N. Patt, Scheduling Algorithms for Modern Disk Drives. Proceedings
of SIGMETRICS’94, pp. 241–251, 1994.
15. Worthington, B.L., G.R. Ganger, Y.N. Patt, and J. Wilkes, On-Line Extraction of SCSI Disk Drive
Parameters. Proceedings of SIGMETRICS’95, pp. 146–156, 1995.
© 2002 by CRC Press LLC