Elardus

> 'Maximizing the SNA RU size reduces end-to-end acknowledgments at the 3270 
> application level.'

This is just common sense.

In order to illustrate the point I'm assuming you have to send 10K of data and 
the flow control parameters are such that each unit of data sent must be 
acknowledged - just to keep it simple.

- If you send 10 units of 1K you are obliged to wait while 10 acknowledgements 
are returned.

- If you send 1 unit of 10K you are obliged to wait while 1 acknowledgement is 
returned.

QED!

It's interesting that Bill Gates's men[1] have obscured the fact that, in the 
document to which Mike Schwab directed our attention, they do not discuss the 
other topic which it is a dereliction of duty not to discuss when talking about 
networking performance.

The two topics which always go together are (1) the size of data units which 
can be employed and (2) the flow control regime which applies.

Incidentally, the flow control in SNA can be implemented as follows:

- the application is such that a limited amount of data is sent before a 
limited amount of data is received

- the application requires definite responses after limited amounts of data are 
sent

- where relatively large volumes are to be sent in one direction such as for a 
printer application, SNA-defined pacing

It is the latter where the PACING and VPACING operands apply - and their use is 
*very* dependent on the configuration of the path over the SNA network. Except 
in the case of printers supported by LU type 1, it is usual to set both of 
these operands to 0 wherever that appear. Also the PSNDPAC, SSNDPAC and SRCVPAC 
operands of the  MODEENT macro corresponding to any mode table entries employed 
should specify 0. Note that these operands are shown as not specified in the 
listing of ISTINCLM. Fortunately the default values are 0 - which does *not* 
apply to the PACING and VPACING operands!

In order that you are not further misled, an application imposing its own 
pacing internally is a contradiction in terms. While most parameters specified 
in mode table entries, so-called "session parameters", serve as advice to the 
application[2] (strictly the application using the "record mode" API and not 
the APPC API), the pacing parameters (reset according to the values specified 
by PACING, VPACING and AUTH=(N)VPACE operands and the configuration of the 
path) are "policed" by VTAM and - I expect without ever having actually checked 
this - other SNA implementations.

-

Another somewhat misleading comment concerned VTAM's buffer pools.

I am a great advocate of regular monitoring of buffer pool patterns of use. 
First of all I advise that dynamic buffering should be used. Then I advise to 
guard against "too much" or "too little".

- "Too much" means that the initial allocation is (far) too large, will never 
or hardly ever be reached and is just wasting storage. This obviously has no 
effect on anything other than the wasted storage.

- "Too little" means that there are always/sometimes periods of rapid expansion 
and contraction. This wastes processor capacity and/but just may be perceptible 
in terms of network activity - but, with today's machines, unlikely!

Note that this is *not* a trivial topic and needs fully to be understood before 
tinkering begins.

-

[1] Anyone who has followed my posts in the so-called "BizTalk Server Host 
Integration Server Forum"[1.1] vainly attempting to correct the impression 
Gates's men seem to have that they are qualified to advise on anything to do 
with VTAM will know that I would advise the utmost scepticism over any advice 
emanating from that quarter.

[1.1] http://social.technet.microsoft.com/Forums/en-US/biztalkhis/threads

[2] It is relevant to this discussion to point out that, indeed, most of the 
"session parameters" are there for the application to use if it so desires. 
Obviously a well-written application might be expected to follow its 
instructions, as it were, and even refuse to start a session where any 
parameter was potentially dangerously out of the range supported by the 
application. Paying attention to these parameters definitely applies to the 
request unit (RU) sizes. IND$FILE may or may not follow the relevant RU size 
however large it is. Perhaps someone who really knows IND$FILE can check 
whether or not there is an upper limit. Alternatively, the point can be checked 
by someone with a "sandbox" ideally with NetView Session Monitor set up.

-

Chris Mason

On Wed, 19 Sep 2012 12:48:32 -0500, Elardus Engelbrecht 
<[email protected]> wrote:

>Mike Schwab wrote:
>
>>Here is Microsoft's suggestions for increasing the speed, from 2004.
>http://support.microsoft.com/kb/125881
>
>Interesting! Thanks Mike!
>
>I see this eye-catcher from micro$oft on that page:
>
>'Maximizing the SNA RU size reduces end-to-end acknowledgments at the 3270 
>application level.'
>
>My searches on IBM Internet Library Collection failed to give me more info 
>about this. Perhaps I used wrong search keywords.  Is there any info from IBM 
>about this tidbit? 
>
>TIA!
>
>Groete / Greetings
>Elardus Engelbrecht

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to