Re: BLOCK(*YES) questions

[Barbara Morris <bmorris@xxxxxxxxxx>]

Dan wrote:
> 
> So, physically, records by key can be all over the file, but BLOCK(*YES) on a
> keyed file puts them in the buffer in key sequence.  If the system has to jump
> all over the disk to build this buffer, is there really any time savings?
> 
> Another question:  Does the RPG program get control of the buffer after the
> first record is put in the buffer by the system, or does the RPG program wait
> until the system fills the buffer?
> 
> This is an eye-opener.  A colleague told me today that the BLOCK keyword for
> the F-spec has been around since v3r7!  That one must have flew underneath my
> radar when it came out.
> 

Usually, doing blocked I/O gives a boost in performance if you are doing
a lot of sequential reading.  But it's not always the right thing to do;
say you are only chaining to new records, BLOCK(*YES) would hamper
performance since the system would fetch N records for every chain when
only 1 was needed.
 
I don't know the relative costs of fetching the data from disk and
making the call to data management.  But the fetch time might not be
that bad; the system might not have to jump all over the disk, and if
the data was already paged into memory, it might not have to go to the
disk at all.  (Getting a bit out of my depth here.)

The RPG program doesn't get control until the system fills the whole
buffer.