🚀 go-pugleaf

In article <1992Jul23.160241.23996@inesc.pt> prc@rabbit (Pedro Ramalho Carlos) writes:

>  1-Will I loose much performance by installing TCP/IP at all - even
>    with no sessions (due to daemons, etc.) ?

It's another set of started tasks, so it will lose you some
performance due to working set and real memory residence.
Fortunately, MVS does memory management and paging/swapping
management pretty well, so if your system is reasonably well
tuned to begin with, you shouldn't see a significant impact.

>  2-What is the relative load on the host of a tn3270 session compared
>    with a native 3270 session? 

About half again as much as a SNA-based user, as a rough
guideline. With an SNA user, the device controllers are handling
a significant amount of the I/O interrupts, which reduces the
interruptions to the access method and OS. Most of the MVS
implementations I've worked with of TCP essentially serve to
connect a TCP connection to a VTAM LU and then MVS manages it as
a SNA connected user. I have less of an estimate on non-terminal
functions such as FTP. 

The difference is very similar to driving a non-SNA terminal, but
not quite as resource hungry.

>  3-And what is the relative performance from the user standpoint?

Depends a lot on network load and CPU load. There is more host
processing involved in the TCP connection, but on a lightly
loaded network and moderatly loaded host, I wouldn't expect a
significant difference. It's comparable to a cross-domain VTAM
connection between two hosts connected by a real CTC.

>  4-Will I need to upgrade the mainframe with CPU or RAM?

Never hurts. 8-) More main storage might be useful if you're
heavily overcomitted on storage, but that's no different than any
other major application subsystem. The answer to this question
really depends on just how heavily loaded your current system is
and how much more load it can tolerate without some significant
(measured on your own scale) performance degradation. 

>  5-If I use IBM's own TCP/IP will I get the same kind of support from
>    them?

The IBM TCP/IP group is one of the most vocal and externally
visible groups of  IBM developers I know of. They listen, they
propose, and they're willing to fix things if the general
consensus of the net is that they're going the wrong direction.
IBM is still heavily committed to SNA, and they'll still be a lot
less confused if you stick to SNA, but TCP support is pretty
good. 

As a side note, most heavy MVS sites I'm aware of use IBM's TCPIP
software, but purchase the hardware interfaces from other
vendors, such as Bus-Tech, Interlink, or OpenConnect Systems.
Some use a device from McData that appears to the host as an
local SNA 3174, but accepts network connections instead of CUT
terminals. 

>--pedro ramalho carlos (prc@inesc.pt)


-- 
David Boyes
dboyes@rice.edu

dboyes@is.rice.edu (David E Boyes) writes:
>In article <1992Jul23.160241.23996@inesc.pt> prc@rabbit (Pedro Ramalho Carlos) writes:

>>  1-Will I loose much performance by installing TCP/IP at all - even
>>    with no sessions (due to daemons, etc.) ?
[ response deleted ]
>>  2-What is the relative load on the host of a tn3270 session compared
>>    with a native 3270 session? 
[ response deleted ]
>>  3-And what is the relative performance from the user standpoint?
[ response deleted ]
>>  4-Will I need to upgrade the mainframe with CPU or RAM?
[ response deleted ]
>>  5-If I use IBM's own TCP/IP will I get the same kind of support from
>>    them?
[ response deleted ]

>As a side note, most heavy MVS sites I'm aware of use IBM's TCPIP
>software, but purchase the hardware interfaces from other
>vendors, such as Bus-Tech, Interlink, or OpenConnect Systems.

I've held my tongue this far, but can't let this pass.  Our OCS
servers will not work with IBM's TCP/IP implementation.  We have
our own implementations of telnet, ftp, and smtp (to profs or snads)
which work over our OCS servers.  Needless to say we feel that our
products will outperform IBM's and they definitely utilize far fewer
mainframe resources than IBM's.  (Amazing how the IBM's lined up)  
I believe that you will find that the resource impact of IBM's TCP
is significantly more that you would believe from David's response.


>>--pedro ramalho carlos (prc@inesc.pt)


>-- 
>David Boyes
>dboyes@rice.edu
-- 
Matthew Lyle                                           matt@oc.com
Technical Support Specialist                           matt@utdallas.bitnet
West/SW/SE/Mid Atlantic Region (*Phew*)
OpenConnect System, Dallas, Texas                      (214) 490-4090

In article <1992Jul24.031725.5562@ra.oc.com> matt@ra.oc.com (Matthew Lyle) writes:
>dboyes@is.rice.edu (David E Boyes) writes:
>>As a side note, most heavy MVS sites I'm aware of use IBM's TCPIP
>>software, but purchase the hardware interfaces from other
>>vendors, such as Bus-Tech, Interlink, or OpenConnect Systems.

>I've held my tongue this far, but can't let this pass. 
>servers will not work with IBM's TCP/IP implementation.  We have
>our own implementations of telnet, ftp, and smtp (to profs or snads)
>which work over our OCS servers.

Oops; brain fault -- thanks for catching it. I was thinking of
your PC 3279G emulator that runs over various DOS TCP protocol
stacks (which I use and like a whole lot) and fingers overran brain.

The point was that few sites I know of buy true-blue IBM
controllers for this purpose -- the other vendors generally produce
hardware that outperforms the IBM offering by a significant
margin -- but many run the IBM software offering with these 3rd
party controllers for compatibility and application development. 
The 3172 is not as fast as it could be, and it's considerably
more expensive than most of the other 3rd party options.

>Needless to say we feel that our
>products will outperform IBM's and they definitely utilize far fewer
>mainframe resources than IBM's. 

Undoubtedly, they do, but you must agree that it moves the tuning
issue from OS tuning for general performance to an access method
tuning problem, which is considerably more difficult to approach
and measure for someone who may not be familiar with all the gory
details. I'm *not* saying that your approach is wrong; it simply
changes the nature of the problem, albeit to one that a site with
a extensive investment in SNA *should* be relatively familiar with.

>I believe that you will find that the resource impact of IBM's TCP
>is significantly more that you would believe from David's response.

I've had reasonable results from those numbers in several
consulting arenas. I admit that they are rough, and in fact, said
so, but they certainly aren't the result of rigorous testing
programs. It's what worked for me.

>Matthew Lyle                                           matt@oc.com
>Technical Support Specialist                           matt@utdallas.bitnet
>OpenConnect System, Dallas, Texas                      (214) 490-4090


-- 
David Boyes
dboyes@rice.edu

dboyes@is.rice.edu (David E Boyes) writes:
>In article <1992Jul24.031725.5562@ra.oc.com> matt@ra.oc.com (Matthew Lyle) writes:
>>dboyes@is.rice.edu (David E Boyes) writes:
>>>As a side note, most heavy MVS sites I'm aware of use IBM's TCPIP
>>>software, but purchase the hardware interfaces from other
>>>vendors, such as Bus-Tech, Interlink, or OpenConnect Systems.
>
>>I've held my tongue this far, but can't let this pass. 
>>servers will not work with IBM's TCP/IP implementation.  We have
>>our own implementations of telnet, ftp, and smtp (to profs or snads)
>>which work over our OCS servers.
>
>Oops; brain fault -- thanks for catching it. I was thinking of
>your PC 3279G emulator that runs over various DOS TCP protocol
>stacks (which I use and like a whole lot) and fingers overran brain.

An affliction of which I am quite familiar with... :)

>The point was that few sites I know of buy true-blue IBM
>controllers for this purpose -- the other vendors generally produce
>hardware that outperforms the IBM offering by a significant
>margin -- but many run the IBM software offering with these 3rd
>party controllers for compatibility and application development. 
>The 3172 is not as fast as it could be, and it's considerably
>more expensive than most of the other 3rd party options.

>>Needless to say we feel that our
>>products will outperform IBM's and they definitely utilize far fewer
>>mainframe resources than IBM's. 

>Undoubtedly, they do, but you must agree that it moves the tuning
>issue from OS tuning for general performance to an access method
>tuning problem, which is considerably more difficult to approach
>and measure for someone who may not be familiar with all the gory
>details. I'm *not* saying that your approach is wrong; it simply
>changes the nature of the problem, albeit to one that a site with
>a extensive investment in SNA *should* be relatively familiar with.

In general if our customers require tuning it is done within VTAM.  The
majority of our sites work run with our recommended BIND images.  Tuning
issues really come up most often when someone wants to run FTP's through
the same PU that they have 3270 traffic.  I does change the nature of
the problem, as you say, from OS tuning to VTAM tuning.

>-- 
>David Boyes
>dboyes@rice.edu
-- 
Matthew Lyle                                           matt@oc.com
Technical Support Specialist                           matt@utdallas.bitnet
West/SW/SE/Mid Atlantic Region (*Phew*)
OpenConnect System, Dallas, Texas                      (214) 490-4090