Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!gem.mps.ohio-state.edu!ginosko!uunet!mcsun!unido!rmi!infoac!msn!m_hanft
From: m_hanft@msn.rmi.de (Matthias Hanft)
Newsgroups: comp.sys.ibm.pc
Subject: >re: ems support on a 386
Message-ID: <890912.145659.852@msn.rmi.de>
Date: 12 Sep 89 12:56:59 GMT
Reply-To: m_hanft@msn.rmi.de (Matthias Hanft)
Organization: Mailbox System Nuernberg
Lines: 33
X-Length: 1500

>
> palowoda@fiver.UUCP (Bob Palowoda) writes
>
>  How long does it take to bankswitch? Are interupts enabled
>  during bankswitch?
>  I was wondering if it would affect communications programs to
>  the point that you would loose characters at 9600bps.
>

  I believe it *does* affect communications programs: I have
  a Compaq 386/20 running CEMM and CACHE which use Extended Memory
  for simulation of LIM/EMS hardware and as a hard disk cache.
  If I mae extensive use of bank switching, some incoming
  characters are lost - but only at 9600 bps and higher
  (communication up to 4800 bps works well.)

  So, I guess bankswitching is rather a memory *transfer* than
  switching. Compaq's VDISK driver lets you install the maximum
  number of sectors which you want to trasfer at one time, and
  they clearly say that if you use too big blocks, some interrupts
  can be lost because all interrupts are definitely disabled during
  protected mode memory accss (the memory above 1 MB).

  I wonder why there is no such "number of sectors" switch
  with the CEMM and CACHE programs. Maybe they use really
  bank switching? But, in that case, I do not understand why
  interrupts are lost (I guess bank switching is so much faster
  than memory transfer so that it would not affect any interrupts).

  In any case, incoming characters at 9600 bps *are* lost when
  CEMM is often called. Something *must* be wrong. But what???

  Matthias Hanft (m_hanft@msn.rmi.de)