I finally got back to this. I updated to the new versions of Multimac, now available in the eCS betazone.
First, I used NETIO126 to test between two windows machines. The numbers for TCP were about 60% of what I see in eCS, but UDP has no problems.
Then, while testing between windows, and eCS, I started getting messages that I was out of buffers, on the NVIDIA machine (using NETIO126.EXE -t -u -s). After some investigation, I found a suggestion that I could add "/mbuf:512 /gdt:256 /mem:3600" to the DEVICE=C:\MPTN\PROTOCOL\SOCKETSK.SYS line, in CONFIG.SYS. NETSTAT -M did not really indicate a problem on any of my systems, except the NVIDIA machine, where it seems that the spare buffers were not plentiful. Anyway, I added the suggested line to all of my systems, except the NVIDIA, and I added "/mbuf:1024 /gdt:512 /mem:7000" to it. Now, there are lots of buffers, and memory usage really hasn't gone up very much. Side effects are still to be determined. NETIO126 performance has not changed however, although it appears that I am no longer running out of buffers. I think the numbers that I am using are too big, but I don't know of any reason to reduce them.
I should probably explain that the NVIDIA machine is an Asus A8N-SLI motherboard, with 3 GB of memory. I use it as a NAS device, and have RSYNC, SAMBA, WEB/2, WEASEL, and Peter M's FTP sever running. I rarely use WEB/2 or the FTP server. I am using eCS 2.1 for testing.
At the moment, I am not sure if all of the MULTIMAC drivers are broken, or if NETIO126 in eCS is broken (the NVIDIA machine does fail, just not as badly as the others). TCP seems to work fine, but UDP doesn't work very well, if at all. I see the same problem between eCS and windows, that I see between two eCS machines. I do not see the UDP problem between two windows machines.
