I have an application that is bursty, but generally 100% writes and it is 24x7. It overwrites old data every week. Normally I would take the peak, multiply by 1.10, and that would be my target//normally I wouldn't try to calculate write cache impacts to the workload. However, because this is bursty, and I will be migrating a dozen machines with workloads just like this example to my new G400s, it suddenly becomes a sizable workload. I don't want to waste too many spindles.
Any thoughts as far as how you would curve the peaks of burst based on 32GB of G400 write cache if it was yours?