TCP-IP on STREAMS


marco graziano (psauly!mark@Sun.COM)
Thu, 15 Sep 88 16:00:19 +0100


Pisa 9/15
First of all let me introduce myself. I am a system software designer at
Olivetti. My group in Pisa, four enthusiastic guys including me,
is currently involved in the development of networking software.
Olivetti produces and sells, among others, 68020 based mini-computers
running a proprietary version of Unix called X/OS.
X/OS is derived from 4.2 BSD and offers some additional features such
as multiprocessor support, NFS, real-time and system V compatibility
(release 2).

Our goal is to have a new implementation of the Internet protocol suite
and of the TLI interface, based on STREAMS. We already have a version
of X/OS including STREAMS and we are investiganting the possibility
of acquiring from third parties STREAMS based TCP-IP components.

We already know of available products by Interactive, Wollongong Group,
Lachman, Spider and Unisoft. We would like to ask some questions about
these or other products we might not know of.

About the architecture, can it be thought of as a set of different
'virtual device drivers' (in the STREAMS terminology) linked together
in a hierarchy reflecting the TCP-IP layers ?
        - If so, what are the real advantages, if any ?
        - Are the different components truly independent from each other ?
        - Can each component be accessed directly, bypassing the components
          above it ?
        - For example, can one easily link another transport
          layer driver beside the ones already existing ?
        - How could NFS be integrated in such architecture ?

About the implementation, how heavily do the system V based implementations
use kernel's functions that are not easily mapped to BSD's ?
        - Are the 'device drivers' kept alive by any deamon, or are
          they initialized by the kernel itself at boot time ?
        - Is there any global data structure/area that is shared by the
          drivers and, possibly, the kernel ?
        - In particular how are the message buffers passed among the layers
          and using which structures ?
        - In the system V implementations, has the original kernel to be
          modified in order to support uniform access to system calls related
          with either sockets or standard files ?
        - Which methodology and tools have been used to measure the performace ?

I want to thank everybody who answered my previous question and provided me
with very useful information.
Bye, Marco.

---------------------------------
 Marco E. Graziano
 Ing. C. Olivetti & C. SpA
 System & Networks Division
 via Palestro, 30
 56100 Pisa - Italy
 Phone: +39-50-500211
 Email: oliveb!psauly!mark@sun.com
 Fax : +39-50-47378
---------------------------------



This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:43:30 GMT