Archives of the TeradataForum

Message Posted: Thu, 20 Feb 2003 @ 23:18:33 GMT


	<Prev	Next>		<<First	<Prev	Next>	Last>>

Subj:		Re: Details on Teradata

From:		Chris Farrell

Thomas,

I am sending this one to the list to let others take a shot at it also. I agree that many minds are better then few minds and my current impression is that there are some really talented people on this list.

Parallel Processing

not related at all. The TD system is based on data distribution based on hash keys. Oracle may distribute the data based on DBA defined partitions. As such the (data) parallelism is ingrained in Teradata and propped onto Oracle. Makes a huge difference!

The documents I read that best illustrated the approach of parallel processing used by both Oracle and Teradata are found at

www.otn.oracle.com...

www.teradata.com...

I guess I don't see the difference right now. I am having troubles focusing today due to a lack of sleep but I still don't see it. Both of the systems use multiple processing units in the form of AMP's or nodes.

Both break a SQL query into pieces.

How does the data reside on a Teradata system? Is the data duplicated on each system(shared everything), or is the data divided up among the systems in some form of table partitioning system, do different tables just reside on different systems, or is this all determined by the DBA?

Both methods really remind me of RAID where you have striping vs parallelism.

****Question 2 & Response

Query Optimization

Not sure what to answer here...my gut statement would have been that TD is really optimized for Data Warehousing whereas Oracle comes from the transactional side. As such, I personally do not think TD is extremely great to handle transactional type of loads.. but enormously fast for large or complex queries. Oracle is exactly the other way around. (This still leaves the question why open...?)

I understood that Oracle was far superior on OLTP and Teradata was superior at Data Warehousing handling complex table joins. I've read that the mere re-arranging of the tables in oracle SQL can change performance from a few minutes to days.

Wouldn't both systems use similar methods for estimating Query cost? The foundations of any optimizer would probably rely heavily on the statistics of the table and also concepts of relational algebra. Ideally you would want to minimize cartesian products. Surely they could make one query estimator for each type of query/transaction that would factor this in.

I am guessing the difference comes in the form of indexing tactics used. However as a programmer I don't know of many ways faster then an old fashioned binary search or some variant of a binary search. Do the DBMS's update the indices differently or does one use more indices then the other?

Cost

I assume you will not be able to come across good numbers for this. The reason is that a lot of the TD deals are on a case by case basis. (At least in my experience.) What I suggest is to take the TPC values and calculate backwards....this, in theory, should allow you to get similar or at least comparable values. Look at www.tpc.org.

I realize this would be very difficult to state which is probably why it isn't provided on their site, at least that I could find. However I would think there is some way to give a baseline estimate of price. I'm guessing the DBMS software can be bought as a enterprise site liscense for both? In both cases, I would think the price difference in hardware would be very close. Teradata uses less hardware resources(from what I have read) and Oracle uses more but can use cheaper, non-proprietary systems. I don't need NASA precision, I just need rough estimates.

I'm not trying to be difficult here. I am just a fascinated student trying to learn and understand the differences. I am very impressed at the level of assistance I have recieved so far and I'd like to offer my thanks to all who responded.

Chris Farrell


	<Prev	Next>		<<First	<Prev	Next>	Last>>

Attachments

Library

Quick Reference