When we made the conversion to OS/390 2.6, I actually got to talk to one of
the TSM server developers. It wasn't related to throughput, but I talked to
him about it. According to him, you won't really see a throughput
improvement so much as a reduction in TCP CPU utilization. By coding
directly to the BPX callable services, they are saving the 'translation' of
the TCP requests into the BPX sockets services. He told me that the big
throughput improvements come with the giga-bit Ethernet OSA cards. The
overhead of the Fast Ethernet OSA is what is preventing the thoughput since
the system still sees it as a channel-control unit-device, and still goes
through the whole IOS path. The gig-bit adapters are direct memory
accessible.

My thoughtput is about 10-12 GB/hour. That is testing a NTserver on the same
switch as the OSA card.

Bill Boyer
DSS, Inc.