|
; From two Trace collections imported to Rdb Expert from the same
; Rdb database where the data hasn't changed the node size suggested by
; Rdb Expert differs on several indexes, why?
I'm not trying to be cute when I say that the input data is
likely different between the two sets. Maybe something like
one of the workloads was all read-only reports scanning
major portions of an index and the other workload had updates,
etc. Whatever they're calling the "power user's guide" these
days reviews some of the rules and I am sure you'll get a
more informed opinion later.
...
; (Please bear with me, I'm new to Rdb Expert and Trace)
Welcome... we enjoy new faces and customers !
; is there a way to tell Rdb Expert not to remove indexes...
It is not that it is removing them as much as it could not
establish a need for them based on the input workload. You
can get the best reading by taking several workloads and merging
them together in a single Trace database to get uniform coverage
(i.e. collections during the day, the nightly report batch runs,
etc. and then on the input to the format separate the various
collected data files by commas). This was all of your bases
are covered.
; is it possible to "flag" that an hash index is not to be generated...
Not to my knowledge. Since I'm just usually pilfering what I
want a la carte out of the procedures though and I never use
the stock generated procedures directly it has never been much
of an issue for me... until I get the "perfect workload" of
course, but I've never encountered such a beast ;-)
Regards,
rcs
|
|
Hi Peter,
I think the fundamental concept that you need to understand
is that the Trace collections represent what Expert uses to
create the workload.
The collection consist of a snapshot of what activity was
occurring at the specific period of time that the Trace
collection was run.
In Expert the workload represents all of the transactions
that occurred during the collection.
If one collection was run during a time period where the
transactions against the database contained a lot of inserts
and updates and the other collection was run during a time period
where the transactions were mainly queries and read only transactions
they would include very different workload data.
We recommend that you run several collections at various times to
enable a representative sample of database activity to be included.
Otherwise, Expert won't have a complete workload and will not generate
good recommendations.
It is crucial for Expert to have a complete representative workload.
To get a good workload, you should run multiple collections throughout
the period that there is activity against the database. Then you
can format them into one Trace database or bring them into Expert
as seperate workloads and merge them before you do the analyze.
Typically, you wouldn't want to run a few long collections, but
many short ones.Otherwise the collection file gets too big and the
processing time is too long.
> From two Trace collections imported to Rdb Expert from the same
> Rdb database where the data hasn't changed the node size suggested by
> Rdb Expert differs on several indexes, why?
The node size is determined by the volatility of your data. If during
one collection as Richard mentioned you had mainly read only transactions
occurring Expert would recommend that you have larger node sizes. If you
have more volatile data with Inserts and updates occurring Expert would
recommend shrinking the node sizes.
The node sizes store the storage of the indexes. Expert would recommend
that the node size shrink with volitile data to help prevent locking at
the node level.
> What parameters could influence this?
There are parameters that influence this, if you feel that you need
further information on them let me know. If you have a good workload you
wouldn't necessarily want or need to modify the parameters.
> (Please bear with me, I'm new to Rdb Expert and Trace)
Welcome! This stuff is confusing until you understand what's going on.
> And more:
> - is there a way to tell Rdb Expert not to remove indexes that have
> not been used during the workload collection? It seems like Rdb Expert
> will drop all indexes that haven't been "touched" during the workload
> collection when the new database is generated?
If you have a good collection this is this still an issue?
An important tip is that once you do the analyze and generate
we don't recommend that you take the generate script and run it as
is. The idea is that you can look at what bits and pieces make sense.
> - is it possible to "flag" that an hash index is not to be generated
> because the data in the table will change over time and an hash index
> would not be the correct choice for this table.
You can set parameter 85 (Enable easy physical) to 2. It will
cause sorted indexes to be created and will prevent hashed indexes
if there are only direct accesses in the workload.
This parameter is located in a file located in RDBX_LIBRARY:
The file is named: RDBX_DB_RDB_n_n.INI
Were n_n represents your Rdb version.
Hope this information is helpful. I'd also recommend that you look
on your kit for the Power User Guide. It has some valuable information.
Let us know if you have any other questions.
Have fun,
Sheri
|