LG> But, it doesn't matter much what the PCI bus bandwidth is if your
 LG> network bandwidth is still only 10Mbit/sec.
 JD>> My Intel ISA adapters get about 900kps, whereas the PCI get
 JD>> about 1200.
 LG> 'kps' ?? I'll assume kilobytes per second, but I'll question that you're
 LG> getting any card anywhere near 900Kbytes/sec, and 1200Kbytes/sec is a
 LG> theoretical impossibility.
"About 1200kps" should be interpreted as "in the high 1100s". And I promise 
you I got sustained 1150+ going between a Compaq Prosignea server with an 
EISA Netflex NIC and a Compaq Prolinea with a PCI Intel EtherExpress Pro/10 
PCI NIC. I don't bother stating CPU speeds because I've found they're 
immaterial. I wrote the benchmark myself, and the throughput was 
independently verified with my line analyzer. The only conceivable flaw in 
the benchmark is if there is some kind of data compression going on at the 
protocol layer I'm not aware of.
 LG> Ten Mbits/sec is 1250 Kbytes/sec and that's the theoretical maximum of a
 LG> 10Base-T network -if- everything is working 100% to specifications.
 LG> The best I've ever seen is about 650Kbytes/sec and that was with a 3Com
 LG> 3C-509.
Most benchmarks have way too much internal overhead. They may make some 
attempt
at measuring real-world characteristics but they rarely give a true sustained 
maximum throughput rating. My benchmark, with smaller data transfers, is 
substantially slower, so it can be used for less-theoretical tests as well. 
But the most important thing is that we can't make any comparisons between 
your 3com on your bench and my Intel on mine, because there's no reference 
point.
And by the way, I've put 3com and Intel ISA and PCI side-by-side, and the 
Intels on both bus types are slightly faster with significantly lower CPU 
utilization.
 LG> Nonetheless, I will concede that the PCI card might grant you a 10%-15%
You can 'concede' what you want, this is not a debate. I run a line monitor 
24 hours a day on a dedicated machine, and the PCI machines consistantly peak 
out 25% higher than ISA. When the LAN is the bottleneck, I take all of that 
25% to the bank.
 LG> Once again, I'll suggest that this is caused by the inefficiencies of
 LG> the circuitry ON the ISA NIC, and not due to the increase in the PC bus.
What do I care if the reason why PCI is substantially faster is either 
advanced circuitry unique to that type of adapter or due to higher bandwidth 
through the bus? If I go out and buy PCI, I get measurably and noticeably 
faster throughput all day, every day. You stated that PCs won't see a benefit 
with PCI adapters, and I insist that it's insanity not to put the fastest 
reasonably priced adapter on each computer. PCI costs me about $165, ISA 
about $110, which makes the decision automatic. The other obvious advantage 
with PCI is that there is no lower CPU utilization possible on a 32-bit 
machine.
Jeff
--- GoldED/2 2.42.G0615
---------------