Re: 1 Billion 11 Byte Words... Need to Check Uniqueness with Oracle

Is the set predetermined or will the set be trickled in to the instance? If the former, I wouldn't bother with Oracle (or SQL Server :), I'd use Unix to 'sort -u' -- which will uniquify it too, then I'd slam it into the instance using SQL*Loader.

At that point you'll show how much faster Oracle will load. Make sure to have enough disk drives configured for Oracle and not enough for SQL Server :)

--
Pablo Sanchez, High-Performance Database Engineering
www.hpdbe.com
Available for short-term and long-term contracts

"Hooty and the Blowfish" <bob_at_bob.com> wrote in message
news:v3s76uciii2742r4sgm19gruvtn5bcja9p_at_4ax.com...


>

> I have a group that is currently using Oracle but is considering


> moving to SQL Server to save some money.  One of the business cases

> they're working with is the testing of 1 billion 11 character words

> for uniqueness.  Apparently they've been sold on the idea that SQL

> Server will rock their monkey.

>

> I tend to disagree and believe that Oracle will handle this task

 much


> more elegantly.


>

> So that I can give this group some guidance:  How would you go about

> testing uniqueness on 1 Billion 11 character words using oracle?

>

> Ideas to get you started:

> 1) Write small clients to do database inserts and distribute them

> across the network on small desktops.  Run Oracle with one table,

 one


> column and one constraint (that it be a primary and unique key).


> Start the inserts and wait for an exception.  You could play with

 the


> number of inserts per commit and number of boxes submitting


> connections to the database.

>

> 2) Install Oracle on 10 boxes and split the 1 Billion words into 10

> segments.  Insert the numbers into the database and check for

> uniqueness.  This won't prove uniqueness across the entire set so

> you'd then have to bulk insert or import all data from each Oracle

> database into one master database that checks uniqueness.  Maybe

 this


> would be faster than checking uniqueness on every insert.


>

> 3) ??? Any other ideas?