home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!wupost!usc!cs.utexas.edu!swrinde!emory!ogicse!das-news.harvard.edu!spdcc!merk!rmkhome!rmk
- From: rmk@rmkhome.UUCP (Rick Kelly)
- Newsgroups: comp.os.coherent
- Subject: Re: Coherent v4 SCSI and ST-506
- Message-ID: <9301050746.09@rmkhome.UUCP>
- Date: 5 Jan 93 12:44:48 GMT
- Article-I.D.: rmkhome.9301050746.09
- References: <930102618@umunk.GUN.de> <9301041066@hphbbs.E.open.de>
- Reply-To: rmkhome!rmk@merk.com (Rick Kelly)
- Organization: The Man With Ten Cats
- Lines: 54
-
- In article <9301041066@hphbbs.E.open.de> hph@hphbbs.E.open.de (H.P. Heidinger, Essen) writes:
- >Udo Munk (udo@umunk.GUN.de) wrote:
- >: Trev Roydhouse (trev@asstdc.oz.au) wrote:
- >: : > From: udo@umunk.GUN.de (Udo Munk)
- >
- >: : > the Adaptec AH-1542 SCSI controller is the best one
- >: : > available today for ISA bus systems. But be warned
- >: : > this controller uses busmaster DMA and has problems
- >: : > with some motherboards. The problem is that you are losing
- >: : > characters on the serial lines when there is heavy disk
- >: : > I/O. This problem does not occur only under COHERENT,
- >
- >: : ... under MS/PC-DOS too. However, the problem is easily solved by installing a
- >: : replacement UART, the NS16550AFN in place of the 8250 or 16450. Of course, VLSI
- >: : serial port chipsets preclude this option, although I hear there is a 16550
- >: : VLSI chipset now.
- >
- >: Not in every case. I know about systems where this problem is worse
- >: with a 16550, because now a whole FIFO is lost. The problem is the
- > ^^^^^^^^^^^^^^^^^^
- >: bus on/off timing, if the DMA controlls the bus for a too long time
- > ^^^^^^^^^^^^^^^^^
- >: the CPU can't read out the FIFO of the UART. This problem ist greater
- >: under UNIX then under DOS, because handling the interrupts is much
- >: more expensive.
- >
- > NO! The problem is NOT the Bus-On/Off-Timing. I wrote a tool
- > 'ahaset' which could set the Busmaster-Timing in the whole range.
- > Using this tool I set the bus-timing to 'worst case' for the disk
- > i.e. 2 ys (micro-seconds) Bus-On, 64 ys Bus-Off, but this didn't improve
- > anything.
- > I believe that the new scatter/gather-support in the ADAPTEC-Driver
- > crushes other I/O.
- > EACH disk access is to detriment of tty I/O - even at only 1200
- > baud. With COHERENTs ADAPTEC driver it is definitly IMPOSSIBLE to
- > work over tty lines independent of direction.
- >
- > Replacing UARTs (->16550AFN) cures nothing but makes things worse,
- > since disk-access from the provided ADAPTEC driver trashes entire
- > FiFo-buffers. Actually it is impossible to use FiFo-UARTs, because
- > UUCPs 'uucico' has severe sycronization problems with whole
- > FiFo-buffers trashed, which results in broke UUCP connections all
- > the time.
-
- I have a 386/33 with Adaptec 1540A, and my ISA bus running at 8 mhz.
- With Taylor UUCP I rarely get broken UUCP connections run Coh 4.0.
-
- Buson/busoff is part of the problem, but the main problem is interrupt
- latency. The Adaptec SCSI driver is locking out interrupts for too
- long. MWC is aware of this, and they are working on a fix.
-
- --
-
- Rick Kelly rmkhome!rmk@merk.com merk!rmkhome!rmk rmk@frog.UUCP
-