A local backup of a server created a 200GB file in under 40 minutes. Subsequently, a backup copy job took over a week to copy this data through the VPN to a remote site. I’d like to figure out how to reduce this time by increasing the efficiency of
the backup copy job. (I’d also like to improve my understanding of TCP to more effectively troubleshoot this type of issue.)

The slower of the two ISP links is 10 Mbps. If I could transfer 200GB at that rate, this job would complete in under 44 hours. The average RTT from three pings is 42mS so I believe that the bandwidth * delay product is 52,500B.

Packet captures show that the backup copy job utilizes six concurrent streams. With 1410B frame size, that’s 8,460B so the pipe is only 16% full. Does this suggest that I should increase the number of streams from six to 36 to fill the pipe or does
TCP automatically open up the window (with no packet loss) to accomplish this?

WireShark shows 20-24% DUP ACKs depending on which side of the link these are captured. Typically, for those segments that have DUP ACKs, there are less than a dozen but I found one #25. I presume these are caused by packets getting dropped due to
congestion; are there other likely reasons? Would this level of dropped packets, by itself account for the poor transfer performance this job achieves? Is the VPN a likely cause?

Please offer tuning suggestions and any information I’m apparently missing that would help answer these questions. THANKS!

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)