Chuck, if you have records that are relatively large compared to the block
size on a Unidata dynamic file that is KEYDATA, even though the record size
is not the split-level percentage of the block size, it will trigger a
split.  I found that a single record written to an empty group, where the
record  size was about 65% of the block size, would trigger a split, even
with the split load set to 95%.  Please note: this is from observation of
file behavior only - I have not read IBM's code and can not be giving away
any trade secrets!  I just needed to be able to explain file-sizing to our
client companies, and the only way to ensure that I was telling the truth
was to experiment a lot and record the results.  So, for records with a
relatively large average record size, I would recommend KEYONLY, or, if you
insist on KEYDATA, at the very least, max out your blocksize in order to
minimize splitting.

We had a client whose file was mushrooming every week, shrinking every
weekend when it was resized, and mushrooming again each week.  It was
Dynamic Keydata with a large average record size, and it was extremely slow.
Making it Dynamic Keyonly improved their performance markedly.  Your mileage
may vary, of course, but I would advise experimentation!

For optimal performance, I tell our clients to stick with static files and
to use FAST monthly to resize their files, rather than rely on dynamic files
to size themselves properly.  The only time I advise dynamic files is when
the file approaches 2  gig - at that point, there is no choice.  (And, no, I
don't work for FAST, so this is not an Ad, just my recommendation based on
experience as their customer.)

Again, from experimentation, not an official pronouncement from anyone who
has read the code, Dynamic Keydata seems to work very well with an average
record size of about 100 bytes, in a 2k block size - unfortunately, none of
the files in our application meet that criteria.

Susan Lynch
F.W. Davison & Company, Inc.

----- Original Message ----- 
From: "Chuck Mongiovi" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Thursday, March 17, 2005 2:20 PM
Subject: [U2] UDT dynamic files


> Does anyone know what kind of a perfomance hit you take for using dynamic
> files? .. I had thought that it was pretty minimal, since splitting
> shouldn't occur too much, and merging almost never happens ..
>
> Anyway, I was archiving data off of a dynamic file today and noticed that
a
> COPY command was taking a really long time .. I re-wrote the copy in BASIC
> so I could put in display counters and got the same results .. I did some
> testing and found that doing the same process (READ/COPY/DELETE) using a
> STATIC file is faster by a factor of about 10 ..
>
> Any ideas?
> -Chuck
> -------
> u2-users mailing list
> [email protected]
> To unsubscribe please visit http://listserver.u2ug.org/
-------
u2-users mailing list
[email protected]
To unsubscribe please visit http://listserver.u2ug.org/

Reply via email to