We have systems running in Azure that connect back to some on-premises equipment over the public Internet. These connections are all made over via TCP (HTTPS) on port 443. The firewall for the on-premises equipment is wide open for that protocol
and port. Azure VMs are all running CentOS 6.5.
Connecting to our on-premises equipment from AWS EC2 and other locations globally works every time.
Azure VMs in West US (where we've been working exclusively in Azure to this point) show a unique ability to suffer connection problems, however.
Some Azure nodes suffer connection failures as high as almost 50%. Some VMs suffer rates that high initially and then magically heal themselves through some process we don't understand and then all future connections succeed. Others will show this problem forever.
traceroutes back to our on-premises equipment show nothing but asterisks (even when running with TCP SYNs). Connections back to troublesome Azure VMs from on-premises equipment also always succeeds so it seems that the problem is with traffic originating within Azure.
Does anyone have any similar experiences? If so, how did you solve the problem? Can anyone suggest any other diagnostic tools we might be able to use?