On Sunday, 10/01/2006 at 06:53 AST, David Boyes <[EMAIL PROTECTED]> wrote: > Thus my suggestion for using PVM. If we really need to solve this > problem, there has to be some packetization going on somewhere to > circumvent the byte-banging-kills-us problem. The PVM line protocol is > relatively efficient, and while it might be a bit jerky, we can pick the > individual characters out of the incoming stream and respond fairly > elegantly w/o constant polling, eg. it would allow presenting a > byte-oriented function w/o making the machine do byte-oriented I/O.
QDIO handles the real live data coming in from the wire, and it doesn't suffer from the traditional I/O interrupt problems. I also think our atavistic fear of interrupts is rooted in memories of a distant past when (a) we had only one CPU, and (2) that CPU was REALLY slow. Today we tout the mainframe as being best-of-breed for context switching. But what triggers those context switches? Interrupts, of course. Timers. But for these quasi-emergency uses of character-at-a-time, I don't think it matters. You can see it in action today with linemode telnet. Every character you type at the client is immediately shipped to the host. There the telnet server lies in wait, watching for the unsuspecting CR. When it finally arrives, the server sends the accumulated data to CP. The design challenge is to connect a linemode telnet session to the guest virtual ASCII console, bypassing the buffering and the ASCII->EBCDIC->ASCII translation. Alan Altmark z/VM Development IBM Endicott ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
