eric@systech.uucp (Eric Lighthart) (01/17/89)
I've followed the lively debate regarding the merits of Ethernet vs. Token-ring with interest. Several articles have pointed out that actual data rates (between a given pair of nodes) depend heavily (primarily?) on the architecture of the CPU/controller interface, and on the performance of the controller itself. E.g., Phil Karn in <13137@bellcore.bellcore.com>: >Anyone with experience in writing drivers can tell you that performance >depends much more strongly on the hardware design of the controller than >anything else. What I'm interested in is this: What, in your _experience_, makes a high-performance interface architecture? What techniques let a the CPU most efficiently issue commands and retrieve status from a controller? (command registers? in-memory control blocks?) What techniques most efficiently move data between CPU and controller? (DMA? contoller-resident dual-port RAM? data FIFOs?) Can intelligence on an I/O controller effectively help unload a host CPU, or should controllers just be "stupid and fast?" Please reply directly or to the newsgroup; [ Mr./Ms. Moderator: Excuse me for barging in like this; if you let me know [ who you are, I'll come in by the front door next time. If there's a more [ appropriate group for this let me know and I'll move it there (with my [ apologies to this group). Thanks. Eric Lighthart, SW Dev. Mgr. Systech Corp., 6465 Nancy Ridge Dr., San Diego, CA, 92131 (619) 453-7400 x 255 systech!eric@{ucsd|uunet} ...!{uunet|ucsd}!systech!eric