Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Database or store to handle 30 Mb/sec and 40,000 inserts/sec

Re: Database or store to handle 30 Mb/sec and 40,000 inserts/sec

From: Tony Rogerson <tonyrogerson_at_sqlserverfaq.com>
Date: Sun, 19 Feb 2006 18:40:44 -0000
Message-ID: <dtae2o$5ln$1$8300dec7@news.demon.co.uk>


>> mode.) You have to be careful using it so you don't slow down the DB.
>> (poor
>> implementation) It will be rare that people will actually use it in SS
>> because of the performance problems it will generate.

More likely because we don't follow that snapshot isolation religion! One site DOES NOT fit all!

>> When would you ever want to read uncommitted records?

No, its been a problem in releases prior to 2005 where people have had to resort to it to prevent the writers blocking readers problem, having said that, in a great percentage of applications its ok becuase of their 'unlikely rollback' scenario, but, as an Oracle guy you wouldn't have experience of that so the likely comment will be based on religion rather than fact and experience.

> However, in Oracle you won't get it if you are just reading. (the bitmap
> index thing is silly, Oracle recommends not using bitmap indexes in OLTP
> systems, just in datawarehouse systems)

And I guess everyone follows recommendations as per in the MS world. You know, one of the impressions I'm taking away from this thread is that people never answer the point, I raise or disprove a point and suddenly the scope of the topic changes or we move onto something else or more comically dribble is posted from some myth on how they THINK SQL Server works.

> In SS unless you enter a
> non-default isolation level (and one MS will say you shouldn't enter for
> long due to performance problems) readers will lock and cause deadlocks.

 Please, now you are being ridiculous; where on earth does MS state you shouldn't enter a non-default isolation level for long - behave, stop making things up - its laughable. The snapshot stuff works just fine, and you can get round deadlocks using it by coding properly AS YOU ARE SUPPOSE TO DO IN ORACLE lol!

The reason there is not an massive uptake in the snapshot isolation and the reason they've not coded it into the product until now is because nobody wants it! Again, its religion and your just as bad promoting the doctorine.

If you guys had only implemented more of the ANSI standard 92 then migration to SQL Server would be far simpler :-)) !

I'm looking forward to installing oracle express to see just how 'easy' it is; mind you when you have limited use until 26 March 06, I mean - what on earth is that all about! At least in MS beta's you get 4 - 6 months to move from the beta onto release code!!!

We can move the thread onto datawarehousing if you want, I didn't realise you still had a solution for that, well, one anybody uses or wants !

-- 
Tony Rogerson
SQL Server MVP
http://sqlserverfaq.com - free video tutorials


"Jim Kennedy" <jim dot scuba dot kennedy at gee male dot com> wrote in 
message news:gMGdnSljY-wDO2XenZ2dnUVZ_sKdnZ2d_at_comcast.com...

>
> "Galen Boyer" <galen_boyer_at_yahoo.com> wrote in message
> news:uzmkncj93.fsf_at_rcn.com...
>> On Sun, 19 Feb 2006, tonyrogerson_at_sqlserverfaq.com wrote:
>>
>> > You would set the database options (once) to allow this
>> > functionality....
>> >
>> > SET TRANSACTION ISOLATION LEVEL SNAPSHOT
>>
>> [...]
>>
>> >
>> > Or, if you to wait on update...
>> >
>> > SET TRANSACTION ISOLATION LEVEL READ COMMITTED
>> >
>>
>> [...]
>>
>> Hm... Even in your database code you are extremely cognizant of the
>> database locking issues setting transaction isolation level to some
>> snapshot for if you, what?, want a consistent view coming back from a
>> query, then, what?, read committed? You have to ask for that? WTF!!!
>> You actually have to deliberately ask for the database to only show you
>> committed rows? WTF? I thought SQLServer was finally caught up to
>> Oracle! Let me ask you one fundamental question.
>>
>> When would you ever want to read uncommitted records?
>>
>> There is the question. If you answer nothing else, answer that one
>> simple question.
>>
>> > Anyway, I'm done here; you guys believe what you want
>>
>> We don't believe anything. We are 100% assured of transactional
>> integrity when using the Oracle database server.
>>
>> > ; the reality is different!
>>
>> Yes, as always, Oracle remains far ahead of SQLServer in this
>> fundamental respect.
>>
>> --
>> Galen Boyer
>
> Galen,
> You need read committed so you can have a system that works the same as a
> file based database (eg Clipper or xbase et al.) The snapshot isolation
> thing in SS is a KLUDGE. (I thank my lucky stars Oracle doesn't allow this
> mode.) You have to be careful using it so you don't slow down the DB.
> (poor
> implementation) It will be rare that people will actually use it in SS
> because of the performance problems it will generate. It is there for
> marketing purposes. SO MS can say we can check that box in requirements,
> but in marketing presentations to companies the MS guys will claim they
> have
> the same feature as Oracle but because that feature causes performance
> problems in databases (lumping Oracle in and thus implying Oracle
> performance is poor because they use this feature. Leaving out the poor
> architecture decisions by MS.) no one uses it.
>
> Also Tony is rather disingenuous (or ignorant, but I suspect disingenuous)
> about deadlocks. Yes, in any database one can get in deadly embrace
> problems caused by locking particular database objects in different
> orders.
> However, in Oracle you won't get it if you are just reading. (the bitmap
> index thing is silly, Oracle recommends not using bitmap indexes in OLTP
> systems, just in datawarehouse systems) In SS unless you enter a
> non-default isolation level (and one MS will say you shouldn't enter for
> long due to performance problems) readers will lock and cause deadlocks.
>
> Jim
>
>
Received on Sun Feb 19 2006 - 12:40:44 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US