Shared memory & table buffering

[bbalci]

Hello Friends,

I'm working on a transactional SAP screen which will be used by approximately 400 users simultaneously,

and I'm designing a good performance model.

This program is using material master, customer master and some other master data too frequently,

Than off course I'm using table buffering ( especially full buffering ) too improve the performance,

because master tables in our system is not being changed frequently.

Here is my problem :

I've created an area in shared object memory using transaction SHMA,

I'm storing an object there to store material master data and this object supplies me some methods

for fast searching like BINARY SEARCH technique etc.

Everything is working fine , different users are able to access this memory too quickly

by using attach / detach methods and search mechanism is very fast.

But my problem is about the server synchronization,

I should refresh the data in that shared object when our material master tables changed by a user.

I mean I should read data again from database when the buffers of this tables are invalidated.

SAP's buffer mechanism is already achieving this process by using profile parameter rdisp/bufretime (frequency )

and logging changes in table DDLOG.

Now I want to refresh when table buffer is invalidated.

Is there any function module that returns the buffer for a table is invalidated or not?

Or how can I refresh my data parallel to SAP's buffer

without reading all data with SELECT a statement each read access ?

Thanks for your time reading this small novel

Any help will be appreciated.

Bulent

[Former Member]

Hi,

Isn't this done by preloading? Just look for the "Session ID: ABAP251" document on google. Part of this document handles preloading and it sounds to me that's what you need.

Good luck!

[bbalci]

Thanks for the answer,

But I've already read that document but it's not mentioning about how can I understand

database table has changed.

I have to use this before preloading.

[Former Member]

[SHMA/SHMM Presentation|http://www.sdn.sap.com/irj/scn/go/portal/prtroot/docs/library/uuid/98dedf90-0201-0010-7790-b299be258b63?quicklink=events&overridelayout=true]

Here's something I used for a presentation to my colleagues. Not sure it helps, but check out the propogation settings and the lifetime settings...

[bbalci]

Hi,

The link contains the same document unfortunately. I've already read this.

[HermannGahm]

Hi Bulent,

sounds like you want to reimplement the table buffer synchronization for shared objects... i'm not sure whether this is a good idea.

You can specify a life time for your shared objects. If you refresh your shared objects every few hours, is that not good enough?

The table buffers are synchronized every 2 minutes. Do you really need the changed master data that quickly in your shared objects? If so i would question the usage of shared objects i think. They were not designed for frequent updates or fast synchronization but for shared and copy free reading of data that changes rarely.

However coming to your question:

There are the function modules:

SBUF_SEL_DDLOG_DATA

SBUF_SEL_DDLOG_RECS

the fist one tells you nr. of records an minimum and maximum time stamp

the second one reads records for a given minimum and maximum time stamp and returns them

in this structure:

SYSTEM_ID

SEQ_NR

TIMESTAMP

CLASS

TABNAME

MODE

KEY_LNG

KEY

Here you could check for TABNAME.... .

But:

- the function modules are not released

- polling DDLOG frequently adds load on your system

- i would probaly not do this in a production system but instead work with the life time of shared objects

Kind regards,

Hermann

[Former Member]

> The table buffers are synchronized every 2 minutes.

But only if necessary.

He said:

>because master tables in our system is not being changed frequently.

so it is the question how often something changes. It can be there are only one or two changes in a day, but after the change the shared objects should also be adapted in few minutes.

So you can schedule a check of the DDLOG entries and trigger an update of the shared objects if necessary.

However, you should also store the time stamp of each shared objects update and should prevent that can become to frequent.

Siegfried

[HermannGahm]

> > The table buffers are synchronized every 2 minutes.

> But only if necessary.

agreed.

> So you can schedule a check of the DDLOG entries and trigger an update of the shared objects if necessary.

> However, you should also store the time stamp of each shared objects update and should prevent that can

> become to frequent.

good idea. with this measures reading only intervals every few minutes and checks that it is not updated too

often... i think as well: why not?

Kind regards,

Hermann

[bbalci]

Hello Dear Siegfried and Hermann,

Thanks for allocating your time,

Actually SAP application servers are already checking the invalidated table buffer records from DDLOG,

in cycles of profile parameter rdisp/bufreftime. ( The default value of this parameter 120 seconds in transaction RZ11 )

And than if the buffer is invalidated for a table, data of this table will be reloaded from database

on next - first SELECT statement from that application server.

So the application server knows that buffers is invalidated. Keeping this information somewhere.

What I was actually asking is there any system function or function module exists in system to get this information...

it would be more useful if exists.

But keeping the last timestamp in a shared object attribute and checking the DDLOG at regular intervals is a good idea,

it won't create a load on system and I can make the interval time as a customizable parameter.

Anymore I can put a spoiler behind the system after doing this for all necessary master data

Best regards

Bulent

[bbalci]

By the way,

if anyone reading this thread has information about the system function or function module I mentioned above,

it still would be very useful for me. Please let us know.

[HermannGahm]

Hi,

i see what you mean... .

So what you want is the status of a buffered table from the local apllication server, right?

In case it is not valid (it had DDLOG records and got invalidated) you will trigger an update of

your shared objects. In that case:

For generic or fully buffered tables:

SBUF_GENERIC_SHOW_OBJECT

input -> tabname output -> stat_id (status)

For single record buffered tables:

SBUF_PARTIAL_SHOW_OBJECT

input -> tabname output -> stat_id (status)

Kind regards,

Hermann

[bbalci]

And Dear Hermann

The functions SBUF_GENERIC_SHOW_OBJECT and SBUF_PARTIAL_SHOW_OBJECT

are exactly what I'm looking for when I was posting this thread.

Thank you for it.

[Former Member]

> So the application server knows that buffers is invalidated. Keeping this information somewhere.

This is relevant, because there are lots of tables, you must decide something on the basis of the tables connected to your shared object. I see nothing simpler than to check whether the connected tables have created DDLOG entries.

>And than if the buffer is invalidated for a table, data of this table will be reloaded from database

>on next - first SELECT statement from that application server.

no, not on the next SELECT ... there are several SELECTs which go to the database (pending and loadable buffer status). The buffer is not loaded with the next SELECT and it is only for the for application server on which this SELECT appears. Only the local application server on which the update was executed has always an up-to-date buffer.

So the table buffer keeps the load from the invalidations small, what you should also do if the shared object is invalidated.

[bbalci]

Dear Siegfried,

I researched about it,

Yes you are right .. buffer of the application server is not being refreshed at next first select

after an invalidation.

Documents I found says next "n" read access are forwarded to database to protect buffer frequent invalidations.

I use st02 to display status of buffer for a table, after I modify a full buffered table

status is set to "pending" and I use st05 to understand buffer used or not .. Buffer is not used for a long time.

Than I wrote a small a program to send 1000 select pieces to that table.. but the status "pending" is not being changed..

After a period of 24 hours I checked the st02.. is still the same.

Can u clarify me when the buffer gets status "valid" again?

Is it belong to a profile parameter?

I use many of tables except for my shared memory object,

so this pending period can cause a countable load for our system.

Thanks anyway.

Best regards.

[HermannGahm]

Hi,

> Can u clarify me when the buffer gets status "valid" again?

> Is it belong to a profile parameter?

yes. Check parmeters documentation for

zcsa/sync_reload_c

zcsa/inval_reload_c

in RZ11. However these parameters should not be changed

in production systems.

For the table that was not loaded anymore... was there enough free space

in the table buffer available? Was your statement able to use the buffer?

Btw: you can find more information about table buffer usage here:

http://www.sap-press.com/products/ABAP-Performance-Tuning.html

Kind regards,

Hermann