home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.dcom.modems:10899 comp.sys.ibm.pc.hardware:20048
- Path: sparky!uunet!uunet.ca!xenitec!zswamp!geoff
- From: geoff@zswamp.UUCP (Geoffrey Welsh)
- Newsgroups: comp.dcom.modems,comp.sys.ibm.pc.hardware
- Subject: Re: Help on upgrade to 16550A UART
- Message-ID: <XgmaoB5w164w@zswamp.UUCP>
- Date: Tue, 21 Jul 92 23:06:08 EDT
- References: <1992Jul19.112553@mycro.UUCP>
- Organization: Izot's Swamp
- Lines: 41
-
- scott@mycro.UUCP (Scott C. Sadow) writes:
-
- > There is only 1 reason to replace a 8250/16450 with a 16550A -
- > interrupt latency.
-
- I'm usually the first to say something like that, but there is one more
- advantage (however slight): you can set the receive interrupt to trigger at
- certain FIFO buffer fill levels, allowing a single interrupt to read up to 16
- bytes and in stead of one interrupt per byte, lowering the CPU overhead of
- going into and out of interrupt service code. Similarly, you can write 16
- bytes on a transmit hold buffer empty interrupt in stead of just one, with
- similar gains. So, the per byte average CPU overhead is lowered... on
- multitasking systems you can sometimes feel the difference while high-speed
- transfers are underway and, on single-task systems, you can sometimes measure
- small increases in performance, not to mention that the CPU has more time to
- write incoming data to the screen while it's still pouring in the port, giving
- another small but perhaps measurable increase in the speed of applications.
-
- > My experience has been that a 12 Mhz 80286 is just barely fast enough
- > to be able to receive characters at 115,200 baud. At this baud rate, the
- > PC is handling roughly 11,520 interrupts/second, so the interrupt handler
- > has to be very small and tight.
-
- Utilities like LapLink, FastLynx, etc. use polled, not interrupt-driven,
- mode to do 115 kbps. Of course, you can only get away with this on single-
- task systems.
-
- > Also, any program that switches the 80386/80486 into protected mode and/or
- > virtual 8086 mode will increase the interrupt latency (EMM386 does this)
-
- That's not half as bad as an AT-compatible 286 switching from protected to
- real mode (i.e., after an extended memory access under DOS); the CPU is
- actually reset, and that doesn't happen in a single clock cycle. <grin>
-
- > Should you replace you current UART with a 16550A? I would only do so
- > if you are seeing problems like reduced throughput and/or high error rates.
-
- I'm with you 100% on that!
-
- Geoffrey Welsh, 7 Strath Humber Court, Islington, Ontario, M9A 4C8 Canada
- geoff@zswamp.uucp, [xenitec.on.ca|m2xenix.psg.com]!zswamp!geoff (416)258-8467
-