Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Mailing Lists -> Oracle-L -> RE: 175 Terabyte Objectivity Database

RE: 175 Terabyte Objectivity Database

From: MacGregor, Ian A. <ian_at_SLAC.Stanford.EDU>
Date: Tue, 06 Feb 2001 10:54:53 -0800
Message-ID: <>

way Objectivity sets up a federated database is  that you have a  master database which records information  about the federation.   An individual database can be attached or detached from the federation.   An individual database is comprised of a database file ,which holds logical structrures termed containers, which in turn hold the persistent data, termed basic objects.  The data is stored in a hierarchical file system, HPSS,  with Redwood tape drives providing the near-line storage. 
are numerous  load balanced data servers which handle parts of the federation. 
<FONT face=Arial color=#0000ff
size=2>Stanford Linear Acclerator Center <A
href=""> <FONT
face=Arial color=#0000ff size=2>  

  <FONT face=Tahoma
  size=2>-----Original Message-----From: Mohan, Ross   []Sent: Tuesday, February 06, 2001 5:46   AMTo: Multiple recipients of list ORACLE-LSubject: OT:   175 Terabyte Objectivity Database
  OOooohhhhh, how COOL!......Objectivity is neither Oracle nor   SS.... is it ( gasp ) "federated" in any sense?   

  Can you tell us more?  This is interesting......   

  -----Original Message----- From:
  MacGregor, Ian A. [<A
  href="mailto:ian_at_SLAC.Stanford.EDU">mailto:ian_at_SLAC.Stanford.EDU]   Sent: Monday, February 05, 2001 8:20 PM <FONT   size=2>To: Multiple recipients of list ORACLE-L <FONT   size=2>Subject: RE: OT - WHAT is a FEDERATED DATABASE ???   We have a 175 terabyte database in Objectivity.  It   houses event data  from  a physics experiments looking at the decay   of B-mesons and their antimatter counterparts, trying to find out  what's   going on with CP violation.
  Ian MacGregor Stanford Linear
  Accelerator Center   -----Original Message----- Sent:
  Monday, February 05, 2001 4:56 PM To: Multiple   recipients of list ORACLE-L
  Ross, glad to see you're starting to come up to speed here.   :>)
> But for the clustering to work, businesses would have to
  change software > and segment the data   The CNet authors obviously got tangled up in their notes and   didn't understand what they were writing about. (Not a   first.) You don't have to "segment the data" in OPS-   that's the "federated database" scene where you place   different tables for the same database app on different servers. If   you segment an enterprise package like SAP or Oracle ERP then   you have 1000's of tables to deal with. Chances are,   no matter how "intelligently" you segment your data,   just losing any random machine, and its attendant <FONT   size=2>subset of tables, will bring the application to a halt and no   more transactions will be possible even though the   database is still "up." That's a single point a   failure and that's the real problem. And to add a machine <FONT   size=2>to the federated cluster you still have to re-segment the data. I   don't believe the good folks at Dell have architected   a federated database like Microsoft did for the   TPC.
  Here's a challenge... Does anyone know of ANY enterprise ERP   type package or any other application where the   software vendor supports a "federated" architecture?   If not then it's likely no one will ever experience the <FONT   size=2>performance seen in the TPC-C benchmarks by Microsoft. If no real world   apps support a federated architecture then we may as   well just ignore all those benchmarks. And after we   throw all those benchmarks out which database engines   consistently score the best on the remaining benchmarks?   Here's another challenge... Has anyone ever worked with or   even know of anyone who's worked with a federated   database? While I wouldn't configure my database   exactly like Oracle configures those used for TPC benchmarking,   (turning off redo, etc.), in terms of physical design I do   believe my databases are at least somewhat similar or   recognizably in the same ballpark. I do not believe   anyone comes close to configuring SQLServer's physical   layout like that used in the Microsoft benchmarks. That's the <FONT   size=2>challenge and until someone can address this challenge we should   practically ignore all TPC benchmarks produced from   Microsoft's federated database architecture.   IMHO.
> the TPC is *independent*. Yes,

  and it's flawed and vendors take advantage of this to dupe the   unwitting.
  BTW, Oracle OPS / EMC doesn't have to be a single point of   failure if you implement the SRDF option but I've   never done it so what do I know? Well I'll answer that   by saying I don't know much but I do try to keep an open <FONT   size=2>minded pursuit of the truth. Sometimes I actually succeed... I   think.   ;-)
  Steve Orr
  -----Original Message----- Sent:
  Monday, February 05, 2001 3:09 PM To: Multiple   recipients of list ORACLE-L
  Very Interesting!  It appears Oracle 9i, is, in fact, a   Hybrid Federated Database! <A
  href="">   A snippet: "An Oracle spokeswoman
  said the new Oracle 9i database, due in the first half   of next year, will feature new "clustering" technology that will make   the company's databases perform faster and more reliably than   before. Clustering allows businesses to harness   multiple servers to run a very large database,   allowing servers to share work or take over from each other if one   fails. The company's previous
  clustering technology, called Oracle Parallel Server, <FONT   size=2>allowed businesses to add as many servers, or high-end computers, as   they needed. But for the clustering to work,   businesses would have to change software and segment   the data, a time-consuming effort for database <FONT   size=2>administrators, said Jeremy Burton, Oracle's senior vice president   of products and services marketing..."   -----Original Message----- Sent:
  Monday, February 05, 2001 5:55 PM To:   ''
  I have some answers, for the curious: <FONT   size=2><A target=_blank

  It appears that SS can partition data storage among   multiple machines, giving it "blow your doors off"   performance. If a machine goes ( gets dynamited at an   Oracle demo, for instance) the data goes with   it. This would be much in the same way that your data   (ALL of it) would go if you blew up the   EMC/Hitachi/StorageWorks array. Oracle Parallel   Server, in contrast, has a single location for it's   data ( read: single point of failure! ) Granted, there   are more failure points in a federated architecture, <FONT   size=2>but there is no such thing as a TOTAL failure ( like "site down"   ) since only part of the data needs to be recovered   from backup. But, with Oracle Parallel Server, if your   disk farm goes down, you lose EVERYTHING.   I suppose if i ever need to store a Petabyte or so, I'll do   it on more than one box, for disaster recovery. So,   this is the "way around" the weakness in hardware loss   for both SqlServer2K and Oracle. <FONT   size=2>And, if I run my PByte database on SS2K, I'll get my answers   alot faster. <nudge nudge>
  -----Original Message----- Sent:
  Monday, February 05, 2001 3:53 PM To: Multiple   recipients of list ORACLE-L
  What's a federated database???????? We   really need to understand this otherwise we'll be duped by Microsoft's   deceptive benchmark claims!! <FONT
  size=2>Comparing the performance of SQLServer in a federated database   configuration to Oracle in a parallel server   configuration is useless and misleading but that's   what Microsoft is doing when they tout their TPC-C benchmarks. In a   non-federated database configuration Oracle8 outperforms   SQLServer handily. Do we really want performance   without fault tolerance? How well does SQLServer   perform when it's down because of its fragility? ;-/ <FONT   size=2>Microsoft "shattered" the TPC-C record with the "federated   database" architecture but even a self-confessed   pro-Microsoft apologist pointed out that no one in   their right mind would actually setup a production OLTP <FONT   size=2>database that way. The point of the demo at OpenWorld was to highlight   the fragility and impracticality of the federated   database architecture as a real world fault tolerant   solution. The demo was quite amusing with smoke and   sound effects. While displaying transaction rates, a node in a running   cluster was "blown up" with predictable results. The   transaction rate for SQLServer went down to zero   because the database was down while the Oracle <FONT   size=2>Parallel Server cluster kept on running. Of course Microsoft does not   want to see its products trashed regardless of the   truth so, in an attempt to prevent Larry from   repeating this demo they sought an injunction based on <FONT   size=2>the fine print of their license agreement which says you can't run   benchmark tests without prior written approval from   Microsoft. (Does anyone ever read license   agreements?) We need a new, more fair benchmark to   measure transaction rates AND fault tolerance of a   database cluster. Something like a standard 4 node cluster <FONT   size=2>and a random blow up of a node. This new benchmark would need to run   a practical, real world application and measure   transaction rates before, during and after the blow   up. It would also be nice to measure the linear <FONT   size=2>scalability of adding new nodes (which is impossible under the   federated database approach without doing a complete   reorg). Oh but now I'm dreaming so it's back to   reading the reviews and making decisions based on gut feel. <FONT   size=2>IMHO, Steve Orr

Received on Tue Feb 06 2001 - 12:54:53 CST

Original text of this message