Path: utzoo!attcan!uunet!know!zaphod.mps.ohio-state.edu!ub!uhura.cc.rochester.edu!rochester!pt.cs.cmu.edu!o.gp.cs.cmu.edu!cs.cmu.edu!Eric.Thayer
From: Eric.Thayer@cs.cmu.edu (Eric H. Thayer)
Newsgroups: comp.sys.next
Subject: Re: Summary: Oscilloscopes and Real Time
Message-ID: <1990Oct26.185541.9354@cs.cmu.edu>
Date: 26 Oct 90 18:55:41 GMT
References: <1990Oct24.210232.24492@wam.umd.edu>
Sender: netnews@cs.cmu.edu (USENET News Group Software)
Organization: Carnegie Mellon School of Comp. Sci.
Lines: 36


In article <1990Oct24.210232.24492@wam.umd.edu> rgc@wam.umd.edu (Ross 
Garrett Cutler) writes:
> The rate of data I need to sample is not much,
> but if I lose a millisecond anywhere, it's no good.  I've talked to the
> makers of Digital Ears and they admit that you could lose data and 
there's
> *no* indication if you do.

I'm sorry I don't have anything non-annecdotal concerning data loss on the 
NeXT.   But, it should be straightforward to design a system which would 
dectect lost data.  The DSP subsystem itself on the NeXT is not subject to 
the scheduling whims of the kernel.   You can design your DSP code such 
that with every DMA transfer it would send the size of the data received 
so far on the DSP SERIAL PORT ITSELF.  This number will be accurate 
(modulo 2^23).  So on the 68k side, you could check to see that the tally 
of the amount of data you have received agrees with the DSP's tally.  If 
they don't agree, then somewhere some data was dropped.  If they do agree, 
you can be confident that you have received the same amount of data.  If 
you were still concerned about integrity of the data, you could gain more 
confidence by keeping a running checksum going on either side too.  What I 
haven't said that is still important is that on the DSP side you should 
also check for overrun errors on the DSP serial port.

The sounddriver has a reasonable setup for data aquisition.  You 
need to ensure that you always keep at least one region waiting to be 
recorded, but if your code can meet that requirement everything SHOULD be 
ok.  Our data aquisition code so far has been able maintain this requirement.

So far, we haven't experience problems that were attributed to data loss.  
But your message has gotten me thinking about putting the above tallying 
feature into our data collection system just to make sure.

----------------------------------
Replies can have NeXT attachments in them
Phone: (412)268-7679