Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!sunic!chalmers!cs.chalmers.se!lindberg
From: lindberg@cs.chalmers.se (Gunnar Lindberg)
Newsgroups: comp.dcom.lans
Subject: Re: HP discless ?
Message-ID: <3310@undis.cs.chalmers.se>
Date: 30 Nov 89 08:10:46 GMT
References: <3436@chalmers.se>
Organization: Dept of CS, Chalmers, Sweden
Lines: 68

Many thanks to all of you who replied to my question - this net
*is* an amazing source of information! A few days ago I got an
answer from a HP employee who is responsible for the product and
I'll make an attempt to summarize:

>HP's discless protocol does not use IP. It uses a request/response
>protocol at transport level.  At link level, it is based on the
>IEEE802.3 protocol. The design of this protocol was published in
>1988 Oct Issue of HP Journal.

Other people have also mentioned "Discless HP-UX Workstations" by
Scott W. Wang, also from an HP journal - possibly it's the same
piece of information.

>The discless protocol will work on a single cable up to 500 feet or
>cables connected by Lan bridge or Lan repeaters.  It does not work
>across IP gateways.
>
>HP's discless protocol is designed for kernel-to-kernel communication.
>There is no user level interface. Conceptually, you can view it as
>a bus protocol for multiple processors in which the bus is a IEEE802.3
>link.

The layout seems to be:

    HP "extended SAP format", IEEE802.3
	 _______
	!	!
	!	! 6   dst addr
	!_______!
	!	!
	!	! 6   src addr
	!_______!
	!_______! 2   length (instead of Ethernet2's "type")
	!___!___! 1+1 dstSAP, srcSAP (both = 0xfc = "ext SAP fmt")
	!   !	! 1+3 control+padding
	!_______!
	!_______! 2   ext dstSAP (0x164f)
	!_______! 2   ext srcSAP (0x164f ?)
	!	!
	.	.
	.	. 24  HP diskless protocol header
	.	.
	!_______!
	.	.
	.	.     data

>Basically, there is a world-wide committee which  controls the
>protocol ID (SAP's).  Every company is given a range of SAP's and
>extended SAP's in which the company can develop its own protocol.
>All IEEE802.3 Lan drivers are designed to recognize the protocol ID
>and deliver the packets to the upper level service routine according
>to protocol ID.  If a system is not configured to serve a particular
>type of protocol, then all packets for that packet are dumped.

>Again, I would recommend you to put discless clusters in a subnet to
>isolate lan traffic.  In discless clusters, we use the network like a
>I/O bus between the disk and processor.  The traffic will be much
>higher than the typical use of a network. The performance is pretty
>good, the I/O rate through discless Lan is within 80% of local I/O.

Once more: Many thanks for all replies,

	Gunnar Lindberg

PS  My question was explicitly about HP's implementation of "discless"
    but, as several people have pointed out, HP also has NFS, YP and
    the other TCP/IP stuff.					     DS