We derive asymptotic formulae for the pdf and rail probability of a ra
ndom variable D that arises as a summation of Gamma(nu,alpha)-distribu
ted observations X-i occurring as a Poisson process with rate lambda.
This is used to determine the reliability of a system that fails occas
ionally and randomly and whose dominant failure modes give rise to Gam
ma-distributed costs and/or down-times. It therefore provides accurate
confidence intervals for the associated costs or down-times in some p
eriod T of service or of observation.