Re: ACCPTHSIZ perfomance - *MAX1TB vs *MAX4GB

[Bob Larkin <blarkin@xxxxxx>]

I do have a success story. Had a 3rd party application with all 4GB access
paths. This was a client server environment, but the problem could occur in any
environment. Had an intermittent problem where the client workstations would
start locking up. Sometimes the lock would last as long as 7 minutes. Using a
network sniffer, we determined (finally!) that a server side program was common
to all of the "freezing" clients. Switching to program traces, we found that the
addition of a record to a very active file was the point where the client would
freeze. Any other client attempting to access the file would freeze.  If a
client was not attempting to use this file, it woul not freeze.

Changed accpthsiz to *MAX1TB, and the problem was solved.

As I understand, here is where the problem lies. The addition of a record to a
branch of the index that is full causes a "mini  rebuild" to split the branch.
With *MAX4GB, the number of entries in the branch is quite large. All of the
records in the branch are locked during the "rebuild".  With the *MAX1TB, the
branch contains fewer entries, so fewer records are locked. Also, the split
occurs much faster because there are fewer entries to deal with.

Pretty interesting/confusing/trivia....
Bob


>
> I found this in the Archives, and I am looking for futher detail.  The IBM
> Redbook "AS/400 Availability and Recovery..." recommends all logicals
> related to a physical should be 1TB or 4GB, but do not mix them.  It refers to
>
> performance  issues if you mix them, but it is not clear if they are referring
> to
> overall  performance, or performance in a specific program.  Anyone have any
> success stories  with this?  I need to resolve a serious (2 minute) response
> time
> issue.
> >
> > Phil