page_649

< previous page

page_649

Page 649



		For (8,2) and (2,8) configurations, m¢ = 1.



		Study 9.5 Open-Queue Study of Various Disk Array Configurations



		We are now ready to compare the various concurrent disk configurations (and compare our results with Reddy's simulation). For this part of the study, we assume an open-queue model, and we compute the total response time based on that model.



		We use the Reddy workload defined earlier as:



		70% 1-block file access.



		30% accesses uniformly distributed over 216-block file accesses.



		First, we need to compute the service and transfer times for each configuration. Assuming 16 disks, we have from the preceding discussion:

Configuration

T_s

T_transfer

(1,1)

26.3

8.8

(16,1)

22.7

6.2

(8,1)

23.1

5.8

(4,1)

23.9

5.0

(1,16)

18.1

0.6

(1,8)

18.6

1.1

(1,4)

19.7

2.2

(2,8)

20.8

1.1

(8,2)

21.6

4.4



		In order to compute r, we must find m, the effective number of independent disk drives. Reddy's simulation assumes that the disks are uniformly accessed. Therefore:



		where



		and m_q = 1 if q = 1.



		Now we can approximate Reddy's workload distribution overall with a co-efficient of variance of c² = 0.15. Using tills and the previously discussed corresponding m factor for each configuration, we have:

< previous page

page_649