[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference ulysse::rdb_vms_competition

Title:DEC Rdb against the World
Moderator:HERON::GODFRIND
Created:Fri Jun 12 1987
Last Modified:Thu Feb 23 1995
Last Successful Update:Fri Jun 06 1997
Number of topics:1348
Total number of notes:5438

931.0. "Rdb/DLM Performance Ref*Oracle" by ODIXIE::BURTON () Wed May 22 1991 03:35

    I'm in the throes of a competitive standoff between Rdb and Oracle --
    we're down to one performance claim:  That DLM and Row Level Locking
    are not used by Digital Rdb sites due to the performance hit.  
    
    I need a reference site using a large Rdb database on a cluster with
    large number of interactive users.  I realize that "large' is
    subjective; ANY reference would help -- my customer is an excellent
    internal salesperson and wants only to be able to call one reference
    that has good performance using DLM.  A chemical manufacturer would be
    great but any Fortune 1000 customer is good also.
    
    Thanks for your help,  Karen
T.RTitleUserPersonal
Name
DateLines
931.1UKEDU::SMITHBBazzoo�Wed May 22 1991 10:0210
>
>    we're down to one performance claim:  That DLM and Row Level Locking
>    are not used by Digital Rdb sites due to the performance hit.  
>    
	Does this statement suggest that Oracle claims that Rdb does not
	use DLM and row locking?

	Interesting, it's a bit hard to turn it off!

Barry
931.2We wouldn't want to confuse Oracle with the truth!KBEAR::STENOISHDBS WestWed May 22 1991 17:5022
    To restate .1, it is impossible to turn-off row-level locking.
    
    However, Oracle might be referring to adjustable lock granularity.  You
    CAN disable adjustable lock granularity.  (When enabled, adjustable
    lock granularity tries locking chunks of pages instead of individual
    pages in an attempt to reduce the number of locks required.  If
    multiple Rdb users contend for the same chunks of pages, then each Rdb
    code automatically locks a smaller chunk of pages (it only locks chunks
    containing the pages it needs).  If there is STILL contention for pages
    in a chunk, then the Rdb code in the process locks individual pages
    that are needed.  If multiple processes frequently contend for pages
    that fall into the same chunks, then some time is spent locking smaller
    chunks.  If adjustable lock granularity is disabled, then Rdb locks
    only pages and doesn't need to spend time switching between locks for
    smaller chunks of pages.  Whether adjustable lock granularity should be
    disabled depends entirely on the application; some perform better with
    it on, others don't.  Whether it is enabled or not, Row level locking
    is ALWAYS done.  (If you need to learn more, the guide to maintenance
    and performance discusses adjustable lock granularity).
    
    Jim
    
931.3HGOVC::DEANGELISMomuntaiThu May 23 1991 05:4714
�                  <<< Note 931.1 by UKEDU::SMITHB "Bazzoo�" >>>

�	Does this statement suggest that Oracle claims that Rdb does not
�	use DLM and row locking?
�
�	Interesting, it's a bit hard to turn it off!

Hi Barry,

I guess a VAXcluster site can choose to 'turn off' DLM by simply accessing 
the database from 1 node. In fact I know of some customers who do this...
But to generalise to all Rdb sites is another example of Oracle 'marketing'.

John.
931.4UKEDU::SMITHBBazzoo�Thu May 23 1991 12:549

Hi,

	yes, it just seems real crazy that they can say these things and
	get away with it. Oracle are excellent salesmen.....maybe DEC
	should by some of them ;-)

Barry.
931.5How low can you discount?SHALOT::DUNCANJoe - CIS/EIC/DCC Groupworks TeamMon Jun 24 1991 16:033
    No thanks.  We have enough trouble with margins as it is! ;-)

    Joe Duncan @ OPA
931.6You pay for your Oracle concurrency!NOVA::NOVA::R_ANDERSONMy timing is Digital.Tue Sep 03 1991 16:2816
From my limited experience with Oracle, you have to PAY extra to get 
row-locking capabilities (I think this option is part of the transaction-
processing package???).  Anyways, the default locking is page-based, which IS
more efficient for distributed lock managers, but provides extremely degraded
retrieval concurrency.  Again, what type of locking is best depends on the
application.

With Rdb, you get page-level locking until concurrency conflicts arise, at which
time you automatically switch into row-level-locking mode.  I think someone
explained adjustable-locking-granularity earlier in this topic.

We are looking into disabling row-locking in Rdb (it's a long ways off, so don't
start offering it to customers :-) for those types of applications that don't
need high-retrieval-concurrency.
FYI.
Rick