Re: Which one is better, Oracle or SQL server - clear answer !

Hi,Bill !

ASYNC I/O in Kernel = KAIO is a absolute minimum requirement for OS. Otherwise only one thread can write to the filesystem or raw device. All other threads will be blocked.

Linux hasn't KAIO yet. But - i have RedHat 8.0 with libaio, a recompiled glibc running. I tested several highly parallel applications on it, 30% performance boost. I really wouldn't recommend running any database on Linux at the moment. HP-UX, Solaris, AIX are quite ok. I had lots of problems implementing this feature in redhat 8.0, 9.0 will be running soon.

Oracle has transaction level read consitency, statement level read consistency, no page locks, no lock excalation to table locks, yes and a very differentiated table lock mechanism:

row share (RS), row exclusive (RX), share (S), share row exclusive (SRX), and exclusive (X)

It has transaction locks, parse locks, MCRC (multi version read consistency)....

State of the art is MVCC, where all transactions can write into/read from table without blocking each other and without showing the symptom of dirty reads or phantom entries. This is well possible in PostgreSQL.

So, IMHO oracle is not perfect.

SQL Server even hasn't MVRC. So, this missing feature alone disqualifies   SQL Server from being a candidate for enterprise systems with many clients, complex transactions and high workload.

This benchmark here show, that there is something wrong with SQL Server:

http://www.pcmag.com/article2/0,4149,1172043,00.asp

Have you ever tried to reindex online, without causing serious performence/locking problems, doing a full online backup with SQL Server ? No ? Try it.

Technical features decide, weather a database server beaks at high loads or still continues to work, but a little bit slower. chaotic behaviour (nonlinear, nondeterministic) is typical for SQL Server, not oracle, not   PostgreSQL, not Informix. I think, i have pointed out the reasons for it.

Look into the usenet groups and see, how many developers have serious problems with locking.

Why do you argue ? MS SQL Server programmers have admitted, that MVRC/MRCC ist still missing. This feature will be implemented in autumn 2004.

reagards, Guido Stepken

Billy Verreynne wrote:

> As for one transaction locking the actual rows (no page locking ever
> in Oracle) to be updated, thus preventing another transaction from
>
Received on Mon Aug 18 2003 - 05:51:26 CDT