Traffic Shaping / SSH & MySQL Slowness

Very odd situation here.

Two datacenters with different ISPs. Each data center has at least a 100 mbps coection or higher.

During hours of say 1 am through 2 pm, everything is fine.

Then randomly between 2pm through midnight, MySQL replication and rsync replication between the servers (keeps them in sync) significantly slows down, discoects, fails to coect. But not all the time.

From Server A to Server B, I can ping and do a traceroute with an average of only 23 ms latency. From Server B to Server A, it is the same story - about 23 ms latency.

During the same time of the major slowness, coection from Server A to Server C (different ISP) and coection from Server B to Server C are fine with no latency. So each test fine individually through another entity. Sometimes it is bad enough where it takes an hour to replicate 10 - 15 MB of data.

Coection seems fast and I don't believe that it is due to network saturation. Both servers A and B also have the mini speedtest ruing on the web service - and when doing a speed test between the servers, the speed is very fast and normal even during the slow times. This tells me that the network isn't saturated otherwise these speedtests would suffer too.

Almost seems like some kind of traffic shaping/filtering that is being implemented somewhere... but obviously it isn't all hours of the day and only during more peak times.

Any suggestions on how that theory could be tested or other things to look for? MySQL runs on the standard port and rsync is done via SSH on the standard port. The significant slowness causes major headaches and failures for the synchronization of the servers.