This content has been marked as final. Show 49 replies
Nothing of my own on bitmap indexes but both Jonathan Lewis and Richard Foote have written extensively on them so I would point you to their posts both here in the OTN forums, their blogs and what you can find that they have posted via google.
Personally I have found that very rarely is a bitmap index the solution for a problem I am facing. And to bring this back to the Exadata server ... when you get to Exadata, full table scans (with SmartScan) are so efficient one can make a good argument for index scans becoming obsolete.
yes, I agree here all the optimizations are there. But what will optimizations will do if there is a disk contention. As i describe above the concept of Teradata has succeed heavily because there are parallel cpus which can utilize the disk in parallel with about sharing other query's disk and that is why if you run two parallel queries on teradata they never create disk contention. this was not the case with oracle RAC since all the disk system was either on SAN with shared everything approach. when all the queries share the same set of disks it create serious contention.
in exadata did they solve this problem?. because if they dont solve this problem every optimization is useless. because how high you many make the IO channel at one point the shared disk system will become the cause of contention. That is why oracle RAC failed badly in Warehousing technology.
I am not making assumptions. most oracle customer says that shared everything is better than shared nothing they are lying here. if there shared everything strategy is so successful why did not they take the warehousing market from Teradata and why they have to create exadata. which is one again a failure due to above mentioned problems and now they are trying to sell it as OLTP+ warehousing machine.
It is not my intention to be insulting here Nick but it appears you really do not understand what an Exadata storage server is or how one works. You are making assumptions that it is just a big RAC cluster with ASM that are not valid. It would be easier to have a discussion with you if you invested some time in learning the differences between the Oracle database and an Exadata server with the Oracle database.
Your question "in exadata did they solve this problem?." is rhetorical.
It is not just Oracle customers that say shared everything is superior. It is also every customer of IBM's DB2 on the mainframe which is also a shared everything environment. If one compares the decisions made by the corporate world, internationally, between shared everything and shared nothing it is not difficult to see that almost all of the planet's data is on shared everything. Not because Oracle is investing billions of dollars promoting it ... because it is technically superior. The only commercial databases running shared nothing are DB2 on Windows/Linux (an insignificant footprint), Informix (great 15+ years ago when databases were MB and GB), Sybase (no longer a contender) and SQL Server (sufficient for some small to medium sized businesses). Shared nothing has a microscopic footprint in the corporate data center. You could make the argument that E-Business Suite, PeopleSoft, and Siebel are on shared everything because Oracle owns them. But try making that argument with respect to SAP which is the largest reseller of Oracle licenses in the world. SAP is not installing on Oracle because of their love of Larry Ellison.
You are in fact making assumptions ... unfounded assumptions. You may have experience on Teradata but you do not have experience on both platforms which I do and which is what is required to be making the pronouncements you are making. Why not take the market from Teradata? It has been less than two years. In 7 years Teradata sales will be as common as sales of Informix.
PS: The use of inflammatory language in these forums "they are lying" is a violation of the terms of service agreement. Please be civil.
For those interested in learning more about this marketplace ...
Another source of independent information is:
which is a required filing.
According to the above Teradata sales in 2009 were 772M$.
Will anyone be surprised if Oracle announces sales that large by the end of its next fiscal year?
damorgan, no need to lose temper this is just a discussion forum and one needs to be fair here.
I have been working with databases from more than 12 years and more than half of it is with oracle and I have worked with very high profile oracle customers and i ensure you that i can take any topic of oracle as easily as anyone can and if you have any doubts you are welcome.
Also i would prefer that if you give technical answers to my questions. If i was making some assumptions it should not be difficult for you to refute them. The problem is i am trying to make some technical analysis and you are constantly giving emotional answers and got heart with it. No need for that as i said i will be critical and you can counter me will nice technical answers after all you have more experienced in oracle than i am and off course an ACE member of this forum.
I never discussed oracle applications you have described. I am only discussing Exadata with oracle Vs Teradata and that is all. The sales figures you are showing is of oracle may be true but Oracle is mostly sold because of its main strength and that is OLTP and the many great features it provides in that field. How much sales oracle have in very high profile data warehouses, very few because mainly here Teradata rules. you have to accept this fact.
Also you have to accept that Teradata is shared nothing technology as i have been saying in my above post and now exadata is also shared nothing. However the cells of Exadata knows about oracle. My question is:
why oracle turn to shared nothing technology in exadata storage ?. I need just a technical answer no aggressive comments.
Nick Naughty wrote:You misunderstood the point I think.
Aman, billy all,
you guys said you are comparing mangoes with apples and then you go on saying that teradata is shared nothing architecture and oracle is shared everything. But the point here was Exadata Storage has been introduce to support oracle Datawarehousing capabilities and if you both architectures are for Datawarehousing so why not compare it.
Exadata itself is a technology used by Oracle RAC for database storage. It alone does not equate to a data warehouse.
Compare that to building a house. Exadata in this analogy means using different type of piping for water and electricity. But that alone does not make the house.
In a data warehouse, there are numerous parts that makes the whole. What about using RDS as Interconnect communication protocol (a lot less latency than using IP)? What about using PQ to process large data volumes? What about using analytical SQL functions? All these technologies (and many more) is what makes an Oracle data warehouse.
Why pull a single technology out of context and compare that technology on its own with another data warehouse product like Teradata?
You either compare the sum of the whole with one another - e.g. Oracle as a data warehouse product/solution with Teradata as a product/solution.
Or you compare technology with technology - e.g. Exadata with whatever storage technology used by Teradata.
In that case, for starters - Exadata uses Infiniband at 40Gb. It can scale (with port bonding using the latest IB switches) to 120Gb. GigE only recently broke through the 10Gb barrier. Fibre channels, at 2Gb rates, fail to compare.
The RDMA (Remote Direct Memory Access) protocol used for storage access can provide port to port latency of below 70ns. This is so efficient, that this protocol specifications have now also been made available by The Infiniband Trade Association to the Ethernet vendors for implementation. RDMA over Converged Ethernet or RoCE (pronounced "Rocky").
So if you want to compare storage layers, then let's do so by all means. But do not confuse a technology with the end solution - especially a complex one like data warehousing that consists of many technologies.
this was my first post in which i requested the members of forum this things
Thanks all the Guys to participate in discussion and respond positively. really nice comparison between teradata and oracle but it was not in context with Exadata. yes i agreed that oracle provided all the facilities which teradata dont. but the real point of discussion is oracle exadata architecutre with oracle ( let name oracle + exadata storage = Exadata for convience only)+
As i said in my above posts lets assume that (exadata + oracle RAC)=Exadata architecture. The same thing you said to me, I have repeated in this thread. Exadata is storage server no conflict here. we are trying to evaluate how (Oracle RAC + Exadata ) vs Teradata will perform. I think you missed that post.
I hope, I make this clear.
I think the confusion is when i say Exadata you are taking Exadata Storage Server. Lets first decided what we should use (Exadata + oracle Rac) collectively. Otherwise the confusion will continue.
Can i call (Exadata + oracle RAC) with all its hardware =Exadata architecture OR oracle database machine. lets first decide it.
Edited by: Nick Naughty on Apr 26, 2010 12:47 AM
Edited by: Nick Naughty on Apr 26, 2010 12:50 AM
There is also one point that the offered characteristics of up to 1.000.000 IOPS and 50GB/second of uncompressed and 500GB/second of compressed data bandwidth is
for Full Rack Version, it is very and very good characteristics but with the price around 1 000 000 USD, I think it is normal :) Half Rack is 500 000 IOPS with price around 500 000USD. As I understand 1 IOPS = 1 USD ?:| ! Yes Oracle Database Machine (ODM) is incredible but it is not for all, I think only very huge companies can benefit from its best characteristics.
But as I know they sell Sun Flash Accelerator for 5000USD it has 96GB flash memory and can provide up to 20000IOPS. On each cell of ODM 4 Flash PCI cards is installed. I think
it will be possible to buy one sun storage and 4 Flash Accelerator cards and install them on this storage and use with Oracle Database 11g! It will be very cost saving solution.
It is also possible to get ODM Basic System version, but there is no price for it in the pricelist, but if 1IOPS is 1USD :) then it can be 75000USD and it will be good price!
I mean when we compare two things not only the characteristics must be considered. It is like "Yes Mercedes-Benz is *very good car* but Hyundai is just 20000USD" ;)
P.S.: If I will choose, then my choice will be ODM!
Information about prices I was get from http://www.oracle.com/corporate/pricing/exadata-pricelist.pdf
I am aware of a number of small to medium sized organizations that have either purchased an Exadata solution or are seriously evaluating it. Apple, for example, runs iTunes on an Exadata. Apple may be large but I would hardly call iTunes itself equivalent to a huge company. Neither would I call a grocery store chain I know of in San Francisco a huge company or a university in Norway.
No single technology is the magic bullet that solves all problems big and small. There are places where Exadata is appropriate and places where it is not. But to pigeon hole it as for huge companies only is off the mark.
Here's an example, using a non-Oracle server technology, that might best illustrate the point. IBM z10 series mainframes can be used to run the Oracle database. If one looks at the comparison of in terms of cost for CPU licenses to Oracle, cost for power and air conditioning, floor space in the data center, etc. A z10 may be expensive ... but it costs less than the hundreds of power-hungry 1U and 2U servers racked and stacked in most organizations. Not to mention infinitely more secure and it runs RedHat Linux as an operating system exactly the same way those Dell and H/P boxes do.
Take a seriously look at Exadata ... it is worth the time.
Hi damorgan and other gurus,
I am new to exadata and in fact in process of clarifying my understanding on this giant. One point i am still unclear about exadata being a data warehouse solution is that in a typical warehouse database, the queries usually will look for tons of data residing in huge unnormalized tables (many a times causing FTS). I am not able to understand how exadata helps in returning loads of data to the end user reports in such straight forward warehouse scenarios. Of course, the Infiniband is capable of transferring this data from storage layer to the database layer. But still it is a fact that this data will be fetched from disks (because flash cache won't be used for large scans/FTS). Though smart scan is there, but will that actually help when there is not much of predicate filtering and column projection??? Say for example, select * from a huge_fact_table with 2 where clauses. All the groupings, sortings still happen at the database layer i guess.
Please help me understand this part of the warehouse optimization claimed by exadata or rather oracle. Is is just relying on Infiniband, no matter how huge the data set being returned from storage layer to the database layer? Will that actually speed up such reporting queries?
I hop i am clear in my question :)
I believe that Oracle RDBMS with TimesTen , Flash disk and now Exadata will gonna smash Teradata in few years.
I work with Oracle for more than 10 years, and if right now, someone offer me twice of my salary to jump in Teradata, I woudn´t do it.
Oracle is the owner of MySql, and I believe that in the future they may create a transparent migration from MySql to Oracle.
IMHO Exadata didn´t born yet, right now all we can see is the head coming out , e.g. (The optimizer not even have the ability to change from index access to Full Scan with Smart Scan. When they do it in Oracle 12 perhaps, it will suprise all of us.
Even though HCC have serious issues regarding update, it´s already a great tool, imagine when they improve it !!
But Exadata is not an appliance - it is appliance-like . In fact the Exadata storage server is an appliance. That is why the storage server is imaged and locked down with almost no changes allowed to it where as the compute node is much more open and the database can be tweak with as much freedom as an other 11gr2 oracle instance.
Exadata is similar to an appliance in that its fast to set up and comes in a single box, but it is not a closed locked ecosystem with a rigid interface. And that is the major advantage of this platform.
Some how I tend to agree with you fundamentally, Nick.
Your point is very logical and to add to it, I have seen the query performance on Teradata compared to oracle as a data analyst in DW environment.
and TD fascinated me with excellent performance on tables will billions of rows and more than 10 years of data.
Oracle has more followers and they are attached to it. TD is much simple compared to Oracle to study. there is no harm in trying it out and then make decision about what is better as a DW solution.
The only concern I have is cost. Can you throw some light on cost involved in teradata implementation v/s oracle implimentation v/s oracle with exadata implimentation for a 10TB, 100 TB and 500 TB data warehouse. Will appreciate your answer very much.