Xref: utzoo comp.dcom.modems:1188 comp.misc:1615 comp.protocols.misc:194
Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!lll-tis!ames!hao!gatech!dcatla!dncar
From: dncar@dcatla.UUCP (Chris A. Roussel)
Newsgroups: comp.dcom.modems,comp.misc,comp.protocols.misc
Subject: Re: Inquiry on the program 'blast'
Message-ID: <2542@dcatla.UUCP>
Date: 31 Dec 87 16:47:21 GMT
References: <335@caus-dp.UUCP> <2395@macom1.UUCP> <59@amelia.nas.nasa.gov>
Reply-To: dncar@dcatla.UUCP (Chris A. Roussel)
Organization: DCA Inc., Alpharetta, GA
Lines: 55

In article <59@amelia.nas.nasa.gov> rfox@amelia.nas.nasa.gov (Richard Fox) writes:
>
>John, 
>
>	do you have any performance measurements that you could share
>with the net? Also can you tell us basically what kind of protocol is
>used?
>
>thanks
>	richard fox
>	rfox@ames-amelia.ARPA
>	(415)694-4358


Glenn Smith and myself wrote the current version of the BLAST protocol when
we were employed at Communication Research Group a few years ago. 

The earlier verssions of BLAST used a protocol which required 8-bit transparencyon circuits.  We sold a lot of that product and we also learned a lot about
how it should be done right.  

The current version of BLAST (I don't think they've changed it) uses a 
protocol that can tolerate 7-bit circuits, but will make use of all 8-bits 
for higher throughput if available.  

Only printable ASCII characters are used in packets ([A-Z][a-z][0-9]['"]) 
so that the packets themselves would be ignored by most system input facilities.
Files are treated as a stream of binary bytes which are encoded into 
packets (similar to uuencode/decode).

The protocol is also flow controllable by the usual XON/XOFF technique.  

It's a full-duplex protocol with a 16-packet maximum window.  
File transfer is supported in both directions simultaneously.

Packet length defaults to around 100 bytes but can be made as large as a 
given implementation will allow.  

The protocol senses when the receiver can't keep up, and slow down packet
transmission (important for going in to mini's).

Selective-retransmission is used, that is, only those packets known to be
received in error are retransmitted.  This is much more efficient on
noisy circuits than "go back N" but requires much more information in the
ACK.

Most of these features are negotiated to the lowest common denominator of
both ends when the protocol starts up.

Note that I am no longer an employee or a stockholder of CRG, but I still
believe that it's a pretty neat product.

-- Chris Roussel
   Digital Communications Associates
   1000 Alderman Drive
   Alpharetta, Georgia 30201
   (404)442-4722