This content has been marked as final. Show 49 replies
Yes, you may compare. teradata works really fine with it's proprietary hardware.
Well, if I build a data warehouse by using Teradata technology and then another data warehouse using oracle exadata technology, then I believe that I can compare both? best regards
But in the context of OP,
Is it wise thing to dedicate your career to a database technology which is designed specially for DSS?
You say that you hold an experience in oracle which works well for DSS and OLTP. The time which you will dedicate to learn Teradata, can be utilized for the enhancement of his Oracle DBA skills.
OP may also visit the views expressed on Exadata by expert:
Fahd Mirza wrote:Yes, in the same sense as you can compare a train and a boat for moving cargo.
Well, if I build a data warehouse by using Teradata technology and then another data warehouse using oracle exadata technology, then I believe that I can compare both?
However, if you want to discuss career as a boat mechanic compared to career as a train mechanic, you may find some overlap in basics (both tend to use engines) but completely different skills required.
It really depends on your point of view.
In your case, you seem to be using the view from an end user or perhaps an end application developer (and especially as a Java application developer). Which could be valid. From a DBA view, the beasts are completely different.
Luckily, I have been following this thread and when i saw this thread, I was astonished.
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. I will be using my words with caution. Recently while on business trip trying to promote Exadata a few of my clients give me this link
The link is a comparison between Teradata architecture and Exadata(oracle + Exadata storage server) architecture which is now a days called Sun Oracle Database Machine.
If TERADATA can make a comparison what makes you think that they are not compariable. because the origional post want to know compartive difference between Teradata and Exadata and you said they are totally different. while the person want to know which one is better from warehousing purpose. Exadata is intelligent storage you are right but Exadata by itself is nothing. it cannot be used without oracle, it cannot be used as SAN etc. So Exadata architecture means oracle + exadata.
I have recently implemented an Exadata system for one of our big client and i must say its a worthy apponent. Oracle with RAC on SAN based system was a totally failure for Warehousing and That is why Exadata storage servers was introduced till its appearance in the market Teradata was no doubt a leading database for warehousing but things are getting in level (although Exadata still have alot of problems but its a leap forward) but oracle with exadata storage is a worthy apponent.
The main problem with oracle is it share everything architecture which is not suitable for warehousing. while Teradata shared nothing architecture is perfect for warehousing but its not an OLTP RDBMS.
with Exadata oracle took a hybrid approach. Exadata seems to me more that share nothing system. The cells of exadata dont share data as oracle does that. Also its the responsbility of oracle to Strip data on all the grid cells.
Lets leave this mango and apples quotation and discuss the paper Teradata produced on exadata v1 release. I know you are quite an active otn discussion forum user.
I have said I guess about 4 times that the Exadata is NOT the Datawarehouse only machine anymore! It was when it first was introduced by Oracle with HP as a partner. With version 2, which came with Sun technology, its now a complete database machine, be it OLTP or DSS! This should be enough to make it uncomparable with the Teradata if DSS is the point of comparison!
The second and the most compelling reason to call the comparison as mangoes and apples or apples and grapes is that we are comparing a storage solution with a software solution! How on earth you can say that both can be compared? If we are comparing Oracle Databsae Software with Teradata Database Software , that's a valid comparison. But if I shall say that SSD is a very fast thing to make my IO faster and in front of it, Oracle is nothing, will you call this a right comparison! I don't understand why this is coming as such a confusnig thing :-( ?
Version 1 of Exadata was indeed a DSS only machine! But that's gone now and Oracle has already introduced version 2. So its not correct IMHO to bring the older version to compare it with the Teradata(even the version 1 comparison was not justified if I am asked).
I have no problem comparing Exadata with Teradata. I spent years at Boeing working with both Teradata and Oracle and I don't see any contradiction.
Reduced to the common denominator both are a combination hardware and software platforms essentially purchased and run as an appliance. Both are database machines. Both can be used for DW, DSS, and OLTP operations. Lets forget about v1 the H/P Exadata and just look at V2.
Teradata can not handle OLTP.
Teradata's almost pure ANSI SQL implementation is handicapped by the lack of extensions and extensibility.
Teradata's architecture is years, many years, behind that of Exadata. There is no technology in the Teradata even remotely equivalent to hybrid columnar compression, or RAC, or Data Guard, or Exadata's SmartScan. Nor does it integrate well into a world of E-Business Suite, PeopleSoft, SAP, Business Objects, Cognos and Hyperion.
Teradata, in many respects, reminds me of SQL*Loader. If you happen to need exactly what SQL*Loader does it is a great tool. But the vast majority of what you need to do with a database can not be done with SQL*Loader. Teradata is a 1980s technology that can not compete in 2010 and the only feature it has that I can think of as one i would envy is the ability to jump onto a previously existing cursor. One check box out of a long list is not a compelling reason to buy something.
Another serious reason to walk away from both Teradata and Neteeza is the same reason people are running, not walking, away from DB2. Where do you go to hire a team that has years of working with the technology and will be with your company in a decade? There is hardly anyone with Teradata experience that isn't over 55 years old. Exadata, on the other hand, requires an Oracle skill set. A readily available commodity in every country and every industry and there are plenty of good people in their 20s and 30s actively improving their skills to continue to run these servers for decades.
I am currently working on multiple Sun M9000s with 256 cores. Would I trade one for a Teradata? Not a chance. Would I trade two for an Exadata? In a heartbeat.
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)
oracle provide all these facilities but oracle has a major major problem, its a system not design for data warehouses while Teradata is the system designed specifically for warehousing purpose (lets not discussion exadata v2 oltp facility i know oracle is king here). Under RAC environments oracle has major problem in IO systems due to its shared everything architecture ( oracle has advertised it as its advantage). The architecture is good for OLTP system but worst for Data warehouse system. In data warehousing share nothing architecture is good and proved successful and exadata is a sort of step to resolve oracle IO problems. This what oracle has been hiding from a long time but preparing for it to resolve the problems.
How much they resolve the problem. Well after get initial performance matrix oracle + exadata looks normal but outstanding and i must say Teradata's above paper holds quite a bit of weight here.
I feel that one should follow the thread give their technical opinions insteading of hanging at your point of view un-necessary.
If teradata makes are comparison between teradata architecture and exadata architecture (oracle+ exadata storage server) you as oracle professional should present you view point on it. Since i know you work with oracle database.
also as i said exadata is nothing without oracle even you cannot store anything on it without oracle database. since oracle + exadata are strongly fused in architecture and none can live without each other under exadata architecture people usually call the whole architecture Exadata. even in above paper from Teradata this terminology is being used at some point
I am sorry if you felt that I am imposing my opinion. It was certainly not my intention to do. I am not an Exadata or Teradata expert so my opinion is nothing more important than anyone else over the thread. I apologize if you felt in this manner. About the Teradata comparing the Exadata with it, I don't know how much of right technical points they would put. But again, I am not sure about it since I myself have not done work over both the techniques.
No. It means the comparison is valid an Teradata loses.
I don't have insider knowledge but I will be very surprised if, at OpenWorld 2010, Larry does not announce Oracle has orders for more mainframe-class machines than Teradata, Netezza, and IBM combined.
SmartScan is brilliant. SmartScan with Hybrid Columnar Compression unbeatable.
But, quite simply, the real brilliance is that the database isn't some off-the-wall proprietary monster that requires a specialty skill set to use. It is the market leading Oracle database which means integration into middleware and apps stacks is a no-brainer.
You are making a distinction that does not exist.
There is no such thing as a system designed for data warehouses. Well perhaps in the mind of some marketing-type sitting in an office manufacturing slogans such as "whiter whites and bright brights" but not when you are standing in the data center.
Everything real is about CPU, I/O through-put, bits and bytes per second. So lets get real and leave the marketing hyperbole to those that can't write a SQL statement or open a Linux command prompt.
Put exactly the same data into tables on a Teradata and into an Oracle database on an Exadata V2.
There are only four operations you can perform.
INSERT, UPDATE, DELETE, and SELECT. Nothing else is real.
Can you SELECT your result set faster (the only metric that matters) on a Teradata?
Can you INSERT additional rows on a Teradata faster? How about UPDATE or DELETE existing rows faster?
And with Exadata it is not a "data warehouse only" machine. I can also use it for OLTP and DSS operations making it far more flexible.
There are perhaps two other metrics worthy of being brought into the conversation for the bean counters ... cost and availability of personnel. I have already addressed the second one. There is hardly anyone in our industry with serious Teradata experience. dice.com currently lists 600 jobs with the the word Teradata in them. There are 13,146 for Oracle. Try to find a school to learn Teradata? How about Oracle? So lets focus on the cost issue. Have you compared the cost? How about the storage cost? Take a wild guess at the difference in price for disk when compression ratios as high as 40-70:1 are available.
If Teradata is still in business 10 years from now it will be in the same way that Informix is still a commercial database.
damorgan, really nice evaluation.
I agree with you on most, a straight comparison between Oracle(Exadata) and Teradata looses. but the comparison is not that straight the biggest problem is most Dataware houses have huge storage capacity. its has huge historice data which require MPP like teradata since in teradata IO system is a sort of parallel. the disk arrays are divide among AMPs. AMPs are teradata working processes(equivalent to salve processes in oracle). due to this division of disks if two or more queries runs on teradata the disk contention is not created.
on the other hand, with oracle (even in exadata storage) data is striped to all array through ASM using SAME technique. so when a query runs all the disks are involved and when two on more queries runs in parallel the contention is being created. because all the queries start fighting for same disk on which the previous query is being runing. A big problem in oracle
Also if i write something about oracle it does not mean that i dont like oracle. I am trying to be as critical as i can to make the comparison fair.
Edited by: Nick Naughty on Apr 24, 2010 4:50 AM
You are making assumptions about Exadata storage with which I can not agree. And given that I co-presented at OpenWorld last year on Hybrid Columnar Compression it is a subject on which I have more than just a passing familiarity.
If you assume that someone is going to buy an Exadata and then try to turn off all of the optimizations ... I'm not sure it can even be done ... then ASM might be an issue. But Exadata storage is not a bunch of ASM disk groups on physical disk spinning at 15K rpm. It is Hybrid Columnar Compression in which I was able, during testing with my own data, to easily hit compression ratios of 35-40TB in 1TB of disk. Oracle has published numbers as high as 70TB in 1TB of disk. So disk throughput is not what you are thinking. Secondly the decompression is happening at the cell level - the disk shelf one might say so your CPUs, the ones you pay for with your Oracle license, are not part of the equation in terms of overhead.
Then there is SmartScan ... the disk shelf, or cell as Oracle would prefer to call it, is providing two optimizations that are critical. The first is that the predicate is applied at the cell level so only the data actually required to answer the query makes it into memory. The second is that the cell, in essence, maintains an index of what is on each disk so it is not actually accessing each disk because it has no idea what is where: It knows precisely what is where. Thus your stated assumption: "so when a query runs all the disks are involved" is far from the mark.
I hope this helps.
My presentation slides are here:
Scroll down to "OpenWorld 10/09"
There is probably no one that knows Exadata better than Kevin Closson. You might want to follow his blog.