About a week. (!). The job ran so long that the snapshot commit time was too long and cause service outages, and it often failed.
As I recall, the bottleneck was two things:
1. The VDDK bug avoidance code for kb2042, forcing network mode instead of hot-add when virtual disks with the same name and SCSI device # were detected. nbd is much slower than hot-add.
2. At the time our backup storage array wasn't performing well.
I am hoping that with the upcoming Nimble support with 9.5 we will be able to back this thing up without affecting service, by using array snapshots.
One strategy I tried was to only start the chain with just a couple of the large data volumes, then add a couple more over the next few days, eventually having them all. Didn't work out, though.
Of course, since my deduplicating storage array cannot do forever-forward incrementals, I must do periodic Fulls. If I cannot get a Full backup of this file server in less than 12 hours, there's no point in trying.