Subject: | |
From: | |
Reply To: | |
Date: | Mon, 9 Nov 1998 16:58:16 -0800 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Mike writes:
>
> While browsing docs.hp.com, I came across the online documentation (Thanks
> HP, I'm sure that the trees of the world thank you as well!) for long
> awaited and much kept under wraps, Master Dynamic eXpansion (MDX) feature.
>
> http://docs.hp.com/dynaweb/smpe/b1019/b840/@Generic__BookTextView/30148;hf=o
>
> What is not clear is why the master dataset is only expanded when it is
> "almost full"! And, what the definition of "almost" is. This seriously
> negates the value of this feature because many systems start to encounter
> performance problems long before their master data sets get anywhere near
> full! I have seen clustering erupt into a performance issue even in the %65
> full levels.(yes, they did have prime capacities)
...
In the design of MDX, several methods of deciding when to trigger an
initial expansion were discussed. Also discussed was when when to use
an existing expansion area.
The current released method is the first phase of allowing
per-database (or possibly per-dataset) configuration.
We currently say:
Search the primary area for up to N blocks and up to K percent of
the primary area. If we don't find a hole before either of those
limits, we'll use the expansion area.
At present, N is 40, and K is 10%.
These values seemed like a good compromise. We're not searching
the entire primary area, but we're giving secondaries some chance
to be clustered near their primary entry.
Now...if we make them configurable, setting either to 0 would mean
"always use expansion area".
SS
|
|
|