On Tue, 10 Apr 2001, Jeff Garzik wrote: > Jones Olatunji wrote: > > Intel is good so also is 3c9x series. > > I think either of them is good enough for you, I have use both of them before and they are quite okay. > > > > But if you still insist on a preference any, maybe I will pick Intel. > > Intel is nice hardware, but their driver needs cleanup and review, and > no docs are available. Plus, there is a patent issue with their > software license, which makes me encourage everyone not to use Intel's > driver software. > > So, only if you have docs and are willing to write your own driver would > I suggest using Intel hardware at the moment... acenic is a good one to > look at instead. The acenic, hamachi and yellowfin are all aging at this point. I'm developing the following drivers here at Scyld: intel-gige.c Intel "e1000" board, optical only ns820.c National Semiconductor DP83820 and DP83821 chips The Intel chip has only a PCI memory space mapping, so that driver doesn't have a matching diagnostic program. The advantages of my driver over the Intel-written driver are It's released (or will be released) under the GPL It eliminates the bogus descriptor cruft of the multi-OS Intel driver It has endian-correction, etc. to support non-x86 kernels The downside is that the driver was written from a "clean room" spec, and thus doesn't use any magic Intel features. The ns820 driver works with the new sub-$100 gigabit boards. It has only been tested with the 10/100/1000 transceiver on the D-Link board, and might require configuration tweaks to work with other boards. Right now the driver uses 32 bit addresses. I plan to switch to 64 bit addresses, which means I can't use the priority queues. Donald Becker becker@scyld.com Scyld Computing Corporation http://www.scyld.com 410 Severn Ave. Suite 210 Second Generation Beowulf Clusters Annapolis MD 21403 410-990-9993 - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
