[TriLUG] The thin line...

[Len Boyle]

Ah, but tcpip was a little different, when it came along. At least with the IBM supported release. 

There was special industrial rack mounted AT machine with a channel card (big heavy twin multi circuit cables) and I believe up to two lan cards. The details are a little bit fuzzy as it was a long time ago. 
But the box booted up msdos then started a multi tasking program that would could emulate an ibm i/o controller using the channel card. Later on IBM would redo the box using os/2 and newer hardware. 
It would receive blocks of data from the host. It would have to deblock the data and extract out the data to be send on the lan (ethernet, fddi,...) and send it out. Or receive data on the lan card. Pack several lan packets together into one block to be send up the channel to the tcp/ip program.  
They had set up special hardware to transfer data between the cards without the need to use the pc bus to speed things up. 

By the way, this box came out of the labs in Yorktown New York and not from the standard product support folks.  

Later using the os/2 version of the box, they moved some of the tcp/ip kernel functions down to this special pc. Now all these things are done.

IBM now supplies this function using special cards that are installed in the mainframe box itself. 
For a while cicso offered this function in cards that installed in a router. I am not sure if they stopped doing that when IBM started selling the new cards that are installed in the mailframe. 

len 

-----Original Message-----
From: trilug-bounces at trilug.org [mailto:trilug-bounces at trilug.org] On Behalf Of Rick DeNatale
Sent: Monday, October 09, 2006 3:01 PM
To: Triangle Linux Users Group discussion list
Subject: Re: [TriLUG] The thin line...

On 10/6/06, Ryan Leathers <rleathers at americanri.com> wrote:
> Got me thinking... fairly off topic
>
> I didn't read that Andrew was talking expressly about web servers.
> Maybe he was instead talking about mainframes as packet switches /
> routers.  Some of that went on in the early days, but it certainly
> doesn't work that way any longer.  I don't think it was ever even the
> norm.  In the mid 80's DARPANET was handing over networking
> responsibilities to the NSF.  To that point, they had been using front
> end processors to handle the network traffic chores rather than the
> mainframe proper.  It wasn't until the NSF got involved that more
> mainframes were added, and considerable growth occurred resulting in the
> need to hold sizable routing information and process packets at such a
> rate that the mainframe became interesting as a network infrastructure
> component.
>
> By the late 80's, appliance routers were already showing up.  The famous
> Cisco quote goes something like "People had a long history of buying
> things that plugged into the wall, made noises and got warm."
> So, the idea of an expensive mainframe as a network infrastructure
> component was already on the way out once the software to handle routing
> had been coupled with a cheap hardware platform.

If we're talking about traditional IBM mainframes, most network
traffic never went THROUGH them.

Even back in the days of SNA, it was the communications controllers
which handled the network traffic, only things destined for the
mainframe ever went there.

And not just network data, but IO in general. That's really what
distinguished mainframes. The CPUs weren't all that powerful, it was
the IO infrastructure. The I/O channels on IBM mainframes were really
specialized computers and could do pretty sophisticated data transfer
without involving the CPU once they got started.

What we are seeing these days is pretty much parallel, with LAN/SAN
distributed processing replacing the old IO channels.


-- 
Rick
-- 
TriLUG mailing list        : http://www.trilug.org/mailman/listinfo/trilug
TriLUG Organizational FAQ  : http://trilug.org/faq/
TriLUG Member Services FAQ : http://members.trilug.org/services_faq/