Re: Database writing architecture

Gerald Ward wrote:
>
> I'm very curious about comparing the writing architecture
> of Oracle versus Sybase. This question stems from prior
> discussions on the difference between databases installed
> on raw partions and databases installed on filesystems
> (UNIX).
>
> Sybase may have problems with database consistency
> if filesystem databases are used because the OS buffers
> the writes. Apparantly, Oracle doesn't have this problem
> or it isn't as big of an issue. Is this indeed true?
>
> Does anybody know or have theories on how these DBMSs accomplish
> their writes and/or what would explain the above mentioned
> behavior? I am not familiar on the internals of either
> system.
>
> By using filesystem databases you loose some performance,
> but gain a great deal in the area of administration etc.
>
> Oracle has two writer processes: log-writer and db-writer.
> Sybase may have something similar but it is done under
> the covers with threads. Could this be a possible explanation?
>
> Also, what are the various issues you have encountered
> through the use of write-back caching controllers and power
> failures, etc.
>
> Thanks,
>
> Jerry.

Jerry

Having had an INGRES background (many moons ago), I had a similar question. I posted a very similar question in the INGRES forum. The response was something like... INGRES, like ORACLE, was a fairly old/mature RDBMS, and as such it progressed/evolved to the state of where it is today, but in the good ol' days to improve performance and to have absolute control over the writes, they in essense wrote their own routines. As the operating systems progressed, there really wasn't much of an internal push to change the they design because that would have meant a radical shift in the way the RDBMS was designed. So... from a fiscal perspective and a risk perspective it didn't make much sense to perform open heart surgery on a product that was well established.

INGRES actually was a very good product - their marketing and fiscal uncertainty was their downfall.

Why Sybase has chosen not to have file systems be protected is beyond me.  I also find it curious that Sybase states "Raw partitions are recommended because of recovery capabilities and increased processing performance." (Pg 3-1 System Admin Guide Supplement). I'll buy the recovery statement, but my tests have conclusively proven that the increase processing statement is wrong. I'll add a disclainer here that your own mileage may vary, but on my system, file systems kick butt compared to raw partitions. (End Editorial.)

Later,

JB Received on Fri Apr 12 1996 - 00:00:00 CEST