Re: Teradata article about exadata
Date: Wed, 27 Nov 2013 18:21:06 -0500
Message-ID: <CAPXEL0+9x182BhVbVgtY=V2xdpr97fJ-=5_txuM7=hkohTBRqA_at_mail.gmail.com>
I haven't worked on Exadata or Teradata and being from Oracle camp always thought Shared disk architecture is much better than shared nothing for a long time. However that is not the case especially for Data warehouse kind of applications when the schema/data is properly partitioned across nodes to match the query patterns.:
- Aggregates calculations usually don't cross the node boundaries and if they do, most can be computed cumulatively from data on individual nodes
- Sharding and replicas avoids data concurrency/availability issues
- Consistency conflicts is hardly a concern on data warehouse (C from CAP) where usually single writer (batch jobs) does DML
- Lots of so called NoSQL databases scales using shared nothing architecture (albeit not to TB ranges) but they prove the concept
- Netezza, Teradata, SQL Server, Scale base, Scalearc etc. have proven customer stories of scaling using shared nothing architecture
- And the big elephant in the room Hadoop is a shared nothing architecture too (yes their support for standard SQL is evolving at rapid pace)
Thanks
Paresh
416-688-1003
On Wed, Nov 27, 2013 at 2:07 PM, Ric Van Dyke <ric.van.dyke_at_hotsos.com>wrote:
> About a 1000 years ago I worked on a major Teradata implementation. There
> are some interesting similarities between the two. However as Tim mentions
> they don’t have anything like PL/SQL (at least not that I’m aware of) and
> even the SQL implementation I found lacking. At that time they did have
> the upper hand on handling massive amounts of data, however today I think
> they have lost that battle as well. Oracle can easily handle the same
> volume of data rather easily.
>
>
>
> One of the basic differences between Teradata and Oracle (Exadata or not)
> is that Teradata works form a “shared nothing” architecture and Oracle is a
> “shared everything” architecture. Each AMP in Teradata can only access
> it’s slice of the data. Where as in Oracle any instance connected to a
> database can access any and all of the data. They work from an idea that
> data needs to be sliced and diced and only one “engine” so to say should be
> able to access any part of the data. So this idea of “shared disks” is
> somewhat of a wild concept to Teradata folks. (BTW- for fault tolerance
> each AMP can access some other AMP’s data in case an AMP fails, but in
> normal processing it can only access it’s part.)
>
>
>
> *From:* oracle-l-bounce_at_freelists.org [mailto:
> oracle-l-bounce_at_freelists.org] *On Behalf Of *Tim Gorman
> *Sent:* Wednesday, November 27, 2013 9:42 AM
> *To:* oracle-l_at_freelists.org
> *Subject:* Re: Teradata article about exadata
>
>
>
> Even prior to 12c ADO, ILM isn't unmanageable; you just have to know what
> you want and code for it.
>
> In Teradata, which (I believe) still lacks an embedded procedural language
> equivalent to PL/SQL (is this true?), this is more problematic than for
> Oracle.
>
>
>
> On 11/27/2013 7:31 AM, Justin Mungal wrote:
>
> *Oracle advocates the SAME allocation policy for data warehousing because
> it believes that in its shared disk environment this policy optimizes
> access performance across diverse access patterns to different tables.
> While it’s possible to control data allocation manually, as the number of
> tables grows, the complexity of specifying data placement manually becomes
> quickly unmanageable.*
>
>
>
> Hmm. The paper predates 12c ADO, So I'm guessing they consider 11g ILM to
> be "manual" data allocation that quickly becomes unmanageable?
>
>
>
> Marketing...
>
>
>
> On Wed, Nov 27, 2013 at 8:19 AM, Stephens, Chris <Chris.Stephens_at_adm.com>
> wrote:
>
> SGA/buffer cache and all the locks/latches/mutexes necessary to coordinate
> access to those buffers.
>
> Teradata doesn't implement acid as far as I know.
>
>
>
>
>
> -----Original Message-----
> *From: *Dba DBA [oracledbaquestions_at_gmail.com]
> *Sent: *Tuesday, November 26, 2013 11:02 PM Central Standard Time
> *To: *ORACLE-L
> *Subject: *Teradata article about exadata
>
> This is a marketing article. I have not used teradata or exadata. I think
> Teradata is basically Oracle running on custom hardware sold by Oracle that
> is specialized for DB performance.�
>
>
>
> anyone know what they mean by "shared disk"? Its on page 2.�
>
> I'd like to avoid an oracle fan argument. I know people who have used
> teradata and find it a very a good product.�
>
>
>
> www.*teradata*.com/white-papers/*Exadata*-is-Still-Oracle/<http://www.teradata.com/white-papers/Exadata-is-Still-Oracle/>
>
>
>
> While Exadata improves Oracle�s I/O performance, Exadata
>
> does not tackle Oracle�s underlying performance and scalability
>
> problems with large-scale data warehousing that stem from
>
> its shared disk architectural foundation.
>
>
>
> CONFIDENTIALITY NOTICE:
> This message is intended for the use of the individual or entity to which
> it is addressed and may contain information that is privileged,
> confidential and exempt from disclosure under applicable law. If the reader
> of this message is not the intended recipient or the employee or agent
> responsible for delivering this message to the intended recipient, you are
> hereby notified that any dissemination, distribution or copying of this
> communication is strictly prohibited. If you have received this
> communication in error, please notify us immediately by email reply.
>
>
>
>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Thu Nov 28 2013 - 00:21:06 CET