I have Solaris 10 1/13 x86 running on a Cisco M-240 UCS. 28G RAM, 1000M network line. I have a Linux-based server doing a data restore pulling data from the Cisco via an SCP restore. Analysis shows large gaps where data is not being sent as the full pipe is not being filled and utilized. The send and receive TCP window sizes are set to 512K and max buffer set to 9M, but the average throughput is still only about 6.5Mbit/sec when according to calculations should be in the 27-30Mbit/sec range at least given the average RTT. Any ideas of what else to look at to increase throughput? The SCP host has plenty of memory and CPU available, the system is hardly working at all, so it is not due to resource starvation. The NIC utilization is near zero so there is plenty of bandwidth available. Because of the amount of data to restore, it runs for days and eventually times out when there is a network glitch somewhere and the connection stalls. PCAPS show some retransmits.
Any ideas appreciated.