Re: Retreiving multiple data sets
Date: Thu, 29 Jan 2015 19:34:37 +0100
Message-ID: <54CA7D3D.5040600_at_bluewin.ch>
Hi Lou,
as my former group RWPG puts it: straight select is always the fastest. I think this is correct 99% of the time. I come across some exceptions where pipeline table functions are faster. Straight sql and all in one go sounds like a great idea. Morphing rows sounds dangerous to me. Any tricking is normally turning back on you.
These is one of the requests where more background would be helpful. Or,
lets put it like that: a lot more background. Your question seems to
indicate a difficult design task.
I don't think we can discuss it adequately on high level. Put in your
position I would get a tuning and design specialist on site. It will
certainly pay off.
Thanks and good luck!
Lothar
On 29.01.2015 19:21, avramil_at_avramil.cnc.net wrote:
> Hi folks,
>
>
> A new project has come our way that will require a decent amount of
> SQL testing and tuning. We thought we would ask the community for
> some suggestions on best approaches/solutions.
>
> Among other things, it’s looking like we will have to send result sets
> of several thousand records every 30 seconds to user browsers. Those
> few thousand records will be collected from multiple tables via UNIONs
> and JOINs.
>
> We are considering several approaches:
>
> Straight SELECT statements
>
> Using stored procedures to return result sets
>
> PIPELINED functions
>
> Bulk operations
> A "regular view / views"
>
> Materialized views
>
> Possibly compressing the result set for network transmission (if this
> isn’t already done automatically)
>
> Other ideas…
>
> Further background: It’s an 11.2.0.3 database with over 400 million
> records spread out over several schemas. It receives 3.5 million
> updates per day. Most records contain alarm data. The primary goal
> will be to retrieve uncleared alarms, plus alarms that were updated in
> the last 5 minutes. Right now, there are 41,000 uncleared alarms,
> which are part of the 240,000 alarms that were updated in the last
> five minutes. We would be returning subsets of this data.
>
> The developers (and we) are looking for a straightforward way to
> retrieve everything in one shot – though that can mean, on our side,
> morphing records from one table to look like those in another.
>
> Thanks,
> Lou Avrami
>
-- -- http://www.freelists.org/webpage/oracle-lReceived on Thu Jan 29 2015 - 19:34:37 CET