On Sep 13, 2007, at 1:41 PM, Michael Watkins wrote:

> * Patrick K. O'Brien wrote on [2007-09-13 12:45:24 -0500]:
>> What I'd be curious to know is how quickly does
>> ShelfStorage startup time degrade in between packings?  I'm hoping
>> that the
>> benefits of ShelfStorage don't degrade too quickly (like, startup
>> time
>> increasing dramatically after only a few transactions without a
>> pack).
>> Does anyone have any insight into this?
>
> Ok, I'll bite -- herein are the results of growing a ShelfStorage
> that has been packed but no other new object activity as yet --
> adding new objects in batches of 5,000 to my collection of 500,000
> "NewsItem" objects:
>
>             Commit New      Startup     Diff
>                           (get root)
> Baseline    n/a             0.001      0.000
> 5,000       5.394           0.185      0.184
> 5,000       5.540           0.364      0.179
> 5,000       5.813           0.556      0.192
> 5,000       5.924           0.736      0.180
> 5,000       6.080           0.918      0.182
>
> Pretty linear.
>
> I then did a pack; comparing it to the original bulk load and pack:
>                     Seconds     RAM Consumed (peak)
> Initial 500,000     247         74MB
> Subsequent 25,000   332         10MB
>
> Adding another few sets of 5,000 records and re-running the timing
> tests produced results very much in line with those listed in the
> first table.

Thanks a ton for satisfying my curiosity.  It seems a shame that
packing is so darn slow, since it brings such obvious benefits to the
startup time.  So now I'm wondering if there is any room for
optimization of the packing process.  Anyone from Durus care to
comment on that?  (Optimize, optimize, optimize.  Nobody ever wants
their database to get smaller, or slower... ;-)

--
Patrick K. O'Brien
Orbtech    http://www.orbtech.com
Schevo     http://www.schevo.org