Le jeudi 13 Avril 2006 21:04, Timothy Miller a écrit : > On 4/12/06, Nicolas Boulay <[EMAIL PROTECTED]> wrote: > > Super IO for x86 : gigabyte ethernet card 8 or 16 ports which are seen > > from the host like one ethernet card (with pci-e speed), the card act > > like a big switch (with a kind of routing). So you could make very fast > > interconnection between server (imaging direct connection between calcul > > server and file server). > > Back in the 90's, IIRC, some engineers came up with an FPGA-based > switch, and it worked brilliantly. Due to the parallelism, it could > route packets part way through, dramatically reducing latency. They > didn't succeed because the venture capitalists thought its design was > too non-traditional and therefore too risky. To me, it's obviously a > good idea. (Beware of patents.) >
It just look like 8/16 ethernet ports card. You could "hide" some feature (make it look like only 1 card by the host OS, add the feature to change the routing algorithme, ...) > > You could also make a SATA RAID controller. The idea is to saturate a > > pci-e link (2*300 MB/s). You could do it with 8 to 16 HD. > > Does 3ware not fill this role well? I think they limit them self to 4 SATA ports on a card. > > > IO connections are the 2 domains where PC can't be compare with > > mainframe. Standard (ethernet, SATA) put the cost down. What is lacking > > is a card to connect the PC with the network and the HD. > > What can you achieve here that separate NICs and SATA cards can't? It's hard and expensive to put more than 3 or 4 pci-e card in the same mainboard. So the idea is to saturate the pci-e link. The next level, is to create a "south bridge" that connectf to a hypertransport link. So you will have 4 GB/s link to saturate... _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
