I am running LDMS version 9 SP2.
A Source representative should NOT be used if the Replicator is local (fast network) to the Source.
That's the fine print I missed! All's well in replication land now.
I have around 150 Preferred Servers and I was getting around 75% failure rate, with some jobs taking over 9 hours to replicate around 300MB !!!!
Since I removed the Source Rep things seem to have improved.
I can't even set a Source Representative any more. Everything was working fine a week ago. Then I add a patch to my share and all of the replicator jobs flip out. Grrrr.
A few items i've discovered:
Source rep builds the file lists of the source. This is probably just creating the file hashes and serves the purpose of reducing CPU load on the source server. Trying to hash across a slow or unreliable link is going to take forever.
Replicators will download the files from the source then put the file on the preferred target. The Replicator can reliably download files across the slow or flakey link, but can't reliably put files on the preferred target if the link is slow/flakey. Your replicator should be close (or the same machine) to the preferred target.
After making adjustments to the Replicator locations in the network my content replication was much more reliable.