On Tue, Jul 17, 2001 at 11:42:46PM +1200, Simon Blake wrote: the 538TX's are RTL8139C - they'll stand 11.5MByte's without difficulty (that's ftp-server -> router -> router -> ftp-client , with RTL8139C's on the routers, various odd cards at either end along the way, so who knows where the bottleneck is). 8139, god, how awful. I bet they chew wads of CPU too. By the looks of things, they use a single linear ring for Rx (meaning some CPU is required to copy to mbufs, skbufs or whatever). Perhaps *BSD allow mbufs with pointers within this frame, not sure. For Tx you have a grand total of four descriptors with nasty alignement requirements. Does there even exist a PCI chipset worse than this? There is more to networking that just 'it works fast', presumably the host also has to do something, and in many cases that something is non-trivial and requires many CPU cycles, if your spending a great many cycles copying data into and out of network buffers, you don't want to have to recopy just to suit crappy network hardware. At 100M, most people who just copy files about probably won't need anything special. Once people move to GE, the need to smarts is more apparent. Consider something like an 8139 (which doesn't AFAIK support interrupt mitigation) running at GE speed with small datagrams (stream media perhaps?). It just won't happen. In fact, most modern hardware will melt with small datagrams at 100M. For the average luse95 users, an 8139 will probably work wonderfully. Personally, I'll spend a little more and use something else. --cw --------- To unsubscribe from nznog, send email to majordomo(a)list.waikato.ac.nz where the body of your message reads: unsubscribe nznog