b. If the legacy application is only failing due to stress then double and load-balance it. This will
drop the number of failures per 500,000 to 9,194.
c. Alternatively treble and load-balance it (with reference to section 18.6.6). This will drop the
number of failures per 500,000 to 6032.
From this, three questions remain:
1. How do we calculate the reliability of each element of the transaction?
a. Create a test harness for each element of the transaction and subject it to an increasingly-heavy
transactional load. Find out the point at which it fails (and make sure the system isn??™t simply
shedding load), and then subject it to a load just below that point for a long period of (say) a
month. Ensure that the harness is capable of restarting immediately after the system fails. Count
FIGURE 5.6 Probabilities of failure
0.9982*0.99997 2*0.9999962*0.9952*09962*0.92
=POWER(B13, 2)*POWER(B14, 2)*POWER(B15, 2)*POWER(B16, 2)*
POWER(B17, 2)*B19=0.8998792047
0.999975 = [1 ??“ ((1 ??“ 0.995) * (1 ??“ 0.995))]
Item
Customer??™s browser R(Br)
Internet R(In)
Firewall R(Fi)
Web server R(We)
Banking application R(Ba)
Legacy application R(Le)
Database server R(Da)
Reliability
0.998
0.99997
0.999996
0.995
0.996
0.9999998
0.92
0.99991
Application objects R(Ap)
Testing and the Web 81
how many times it fails during that period (say it is 6). Then over a period of 30 days * 24 hours
(=720 hours) it fails 6/720 times per hour or 0.
Pages:
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228