| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 181.1 | A little more info ? | MUFFIT::helen | Helen Pratt | Fri Apr 18 1997 17:22 | 47 |
|
>>Hello, We have a customer running UNIX 3.2C and CICS 2.0 and
>>Oracle 7.1.6.
We have never shipped CICS for Digital UNIX V2.0. Can you check
what version you are really running. We would recommend that
you upgrade to CICS V2.1A on Digital UNIX V3.2G, given that
Digital UNIX V3.2C is unsupported from later this year.
>>They have a porting from a host application
>>and now they are making some benchmarks before production.
>>The system is a 8200 5/300, 2cpus and 1GB.
>>When they execute 100 transactions at the same time they
>>get 14 transaction per second, 200 trans. get 10 transactions
>>per second, 1200 trans. get 4 trans/sec., and 2600 get
>>1,5tran/sec.
Firstly how are you obtaining these figures? Without any
context it's somewhat difficult to explain. Please can you
post the contents of your RD stanza, and your environment variable
settings when starting the region. What sort of clients are you
using?
Please can you also post the output from either syd or vmstat.
>>
>>First, Does Any one know if this numbers has sense?
Potentially they make sense if the system is not fully turned - the
above information should help determine this.
>>
>>In this system I can see a strange memory consuming. When
>>the system startup there is a lot of free memory but during
>>the day the memory come down and it doesn't recover if the
>>system has not any activity. Is it normal with CICS ?
>>or is it a annormal behavior ? because I don't see any swap
>>activity in any case.
It sounds as if something is leaking memory. Are you using
an epi client?
Helen.
|
| 181.2 | | NNTPD::"[email protected]" | [email protected] | Mon Apr 21 1997 10:09 | 20 |
| Thank you very much for your quickly answer.
I'm sorry for my mistake, the CICS version is 2.1.
They have not any client, they are doing a very heavy
local benchmark with an own transaction emulator.
They are going to repeat the test and they are going
to send me the output of vmstat and iostat.
Also, I recomended the upgrade to CICS 2.1A and UNIX 3.2G
but this is not a good momment because they are moving
to production the application.
I'll put here more info when customer do the test again.
Thanks for all
Regards,
Bego�a
[Posted by WWW Notes gateway]
|
| 181.3 | Complete set of info please.... | CICS03::helen | Helen Pratt | Mon Apr 21 1997 10:28 | 24 |
|
>>Thank you very much for your quickly answer.
>>I'm sorry for my mistake, the CICS version is 2.1.
>>They have not any client, they are doing a very heavy
>>local benchmark with an own transaction emulator.
They must be using a client of some sort. You need a client
to run transactions. The client may be a variety of things,
from one supplied, to a home produced client which uses a
client API provided by CICS. We need to know which in order
to comment the results.
>>They are going to repeat the test and they are going
>>to send me the output of vmstat and iostat.
Please check that you obtain the information requested in
the previous note - that information is necessary to
comment on the results - the output from vmstat and
iostat is not sufficient. Please also make sure that
the statistics gathered all relate to the same run.
Regards,
Helen.
|
| 181.4 | | NNTPD::"[email protected]" | [email protected] | Mon Apr 21 1997 19:38 | 24 |
| Helen, Thanks again.
My problem is the benchmark is making Andersen Lisboa
for a final common (Andersen and Digital) Spanish
customer (big Bank in Spain) and I'm in Madrid.
Now, Andersen will repeat the benchmark in some days and
send us all information I request them.
So, I've requered the RD stanza and environment variables
(I think all is by default) as you mentioned,
the vmstat and iostat output for the test interval
and now more information about how the transactions are invoked.
In any case, I remember some thing like they have a oracle
table where they insert a record for each transaction they
want to run and then, a program read this table and
invoke a on-line transaction for each record at the same
time.
Regards,
Bego�a
[Posted by WWW Notes gateway]
|
| 181.5 | Please ask... :-) | CICS03::helen | Helen Pratt | Tue Apr 22 1997 11:48 | 11 |
|
Begona,
Please can you ask the people doing the bench mark
which client (or clients) they are using. This is
very, very important in relation to the memory problem
you described in the base note.
Thanks,
Helen.
|
| 181.6 | the client is a cicsterm | MIPSBX::"[email protected]" | Ricardo Lopez Cencerrado | Tue Apr 22 1997 19:28 | 30 |
| Hello,
I have also been involved in the benchmark effort made by this customer and
I
guess this information about the test program can clarify some points.
The customer developed a transaction which made several "EXEC CICS START"
calls to launch as many asynchronous transactions as required. They read the
transactions and the data required for each transaction from an Oracle table
and launch all the transacctions found in the table.
The test transaction is called from an standard cicsterm terminal and they
are
not using any API (ECI,EPI) to implement the test. They simply launch as many
transactions as required, with no pause between them, an measure how long it
takes to finish all launched transactions. They launch the transaccions as
fast
as they read the database table.
Hope this information makes it easier to foresee possible botlenecks in the
CICS system that cause those performance numbers.
We will try to post ASAP the region definition and environment files. If
other
information might be useful, please let us know it so that we can post it.
Thanks in advance,
Ricardo.
[Posted by WWW Notes gateway]
|
| 181.7 | Please rethink...... | CICS03::helen | Helen Pratt | Wed Apr 23 1997 11:54 | 33 |
| >>
>> The customer developed a transaction which made several "EXEC CICS START"
>>calls to launch as many asynchronous transactions as required. They read the
>>transactions and the data required for each transaction from an Oracle table
>>and launch all the transacctions found in the table.
This is not a valid performace test of the system. If the benchmark is
attempting to simulate multi user load this is unrealistic. This test
measures the ability to manage the ATI queue (ICE chain), when only
one session is adding transactions to it. A better test would be to
have multiple client sessions, each invoking transactions.
Running the benchmark in this way explains the results in the base note.
With an increased number of transactions the ICE chain will just get longer
and longer and require increasing amounts of time to manage.
>>
>> We will try to post ASAP the region definition and environment files. If
>>other
>>information might be useful, please let us know it so that we can post it.
These files are vital to any understanding of performance and tuning, however
you must have a valid test. Untuned CICS performance can sometimes be
several times slower than a tuned system, even for a real customer load.
We will need also need CICS stats output.
Please rethink how you're going to conduct this benchmark.
Regards,
Helen.
|