We have been testing the performance of merge replication. We're finding the results difficult to explain.
We have a complex merge publication with subscribers on both a high speed home office network and on lower bandwidth field office networks. We had subscribers in both locations import the same 2000 rows and synchronize their subscriptions.
On the home office network (100 Mbps, low latency), subscribers consistently achieve about 2500 changes / minute. On the field office networks, the best performance we've seen was about 700 changes / minute (on a 3 Mbps, 74 ms max latency network).
Clearly the network is an issue, however, when we look at network utilization, the maximum utilization was 29% in the field office, even with 5 subscribers simultaneously synchronizing. The test was done after hours, so all the network activity was
due to replication.
The puzzling part is that if the network is the bottleneck, why is only 29% of the bandwidth used? How can we get replication to more efficiently use the network? We need to answer these questions before we can justify purchasing more bandwidth.
Thank you for your assistance.
View Complete Post