[osflash] Object Databases
Evert | Rooftop
evert at rooftopsolutions.nl
Tue Oct 28 18:15:03 PST 2008
An ODBMS will not solve your problems. 900M is quite a bit of records,
so I can see that you'll quickly run into problems there. I would
recommend approaching this problem with the usual optimization tricks.
Start off with your requirements for the reporting you'll be needing
and work your way down. A simple example.
Your 900.000.000 records will require time-based reports and a
granularity of 5 minutes is required. You receive 500 records every
second on average. This will effectively reduce your entire dataset to
from 900M to 6000 records. Calculate these summary tables every 5
minutes, and your reports should be delivered under 20ms.
In practice you'll end up creating separate summary tables for every
field you reference in a GROUP BY statement.
If your dataset is much more complex than simple statistics, looking
into a map-reduce type scheme like Hadoop.
I hope this helps,
P.s.: recommended reading material:
On 28-Oct-08, at 12:18 PM, João Saleiro wrote:
> I am researching on Object Databases to integrate with our RIA
> solutions. We have to build a statistics system to handle complex
> databases that can have tables with more than 900.000.000 records.
> The system needs to be VERY efficient (i.e. return the result of a
> calculation in less than 4 seconds). We are curious about Object
> Databases, especially the Versant one. We have read all their
> "marketing" material, but I would like to prefer the opinion of
> someone who has used ODBMS. Does someone here has experience with
> this kind of solutions?
> João Saleiro
> CTO @ Webfuel
> Adobe Community Expert
> Co-founder of the Portuguese RIA Community
> osflash mailing list
> osflash at osflash.org
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 2433 bytes
Desc: not available
More information about the osflash