Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!umich!samsung!interlan.InterLan.COM!interlan.InterLan.COM!backman
From: backman@interlan.Interlan.COM (Larry Backman)
Newsgroups: comp.protocols.tcp-ip.ibmpc
Subject: Re: Two TCP/IP stacks on one PC
Message-ID: <1990Jun1.113859.15686@interlan.Interlan.COM>
Date: 1 Jun 90 11:38:56 GMT
References: <9005310245.AA17156@alw.nih.gov> <9005311412.AA02654@vax.ftp.com>
Reply-To: backman@interlan.interlan.com (Larry Backman)
Organization: Racal InterLan, Inc., Boxborough, MA (1-800-LAN-TALK)
Lines: 36

In article <9005311412.AA02654@vax.ftp.com> jbvb@VAX.FTP.COM (James Van Bokkelen) writes:
>It is puzzling to hear that 3+Open mail doen't use Netbios sessions,
>in light of the way that OS/2 and NDIS are structured.  The theory of
>how OS/2 was supposed to network (as propounded by Microsoft and 3Com)
>was that all network activity was going to take place via "named
>pipes" (preferred) or Netbios sessions (via the user NCB API), and
>thus they didn't specify any means of programming the actual transport
>layer from a user application.  LAN Manager talked to a Netbios
>embedded in the kernel, user NCBs were passed to this Netbios, and
>that was it.  A vendor could presumably specify a private API for
>a particular transport layer, but Microsoft stayed out of it.
>

Ah James, performance, performance, performance.....

NDIS --> NETBIOS is a pretty interface but as with most generic interfaces
is slower than a private (read transport layer) interface.  I've gone through
at least 4 rounds of OS/2 NDIS drivers's; the things I do to go around the OS
at this point are reminiscent of DOS to say the least.

For instance, my IPX driver uses a private transport interface up to user
space.  Rounds 1 & 2 used the DosDevIOCtl call.  How do you spell ring 
transition --- slow.... Round 3 uses a shared memory queue and RAM semaphores
to speed things up.

For another instance; use the kernel debugger to step through and count the
instructions in your favorite DevHelp call, some of them are scary!  I used
to use PhystoGDT at interrupt time when I received a packet; no more! I also
lock user buffers at Open time, the DevHelp Lock/Unlock calls are basically
useless for real-time programming.

BTW, OS/2 1.2 does clean up some of the sins and slowness of OS/2 1.0/1.1.
Give Microsoft the credit for at least trying.

					Larry Backman
					backman@interlan.com