Re: pros/cons of dedicated vs. MTS connections?

Joe Weinstein wrote:
>
> Mark Bole wrote:
>

>> Joe Weinstein wrote:
>>
[...]
>
> Hi Mark! Yes I continue to forge my own posts! ;)
> I have recently discovered I don't know everything. ;)
[...]
> I appreciate your confirmation about a dedicated server.
> Is there any official Oracle documentation stating such
> as clearly as that? Or can you point me to a post that
> delineates why?
> thanks again!
> Joe Weinstein at BEA

Joe:

FWIW, following is an excerpt of an in-house e-mail I wrote in January 2002. This was Oracle version 8.1.6, as I recall (I don't recall the Java version, probably 1.2 or something). This test did not include a JDBC connection pool, but rather plain vanilla JDBC (OCI) connection initiated via a batch job. Since I cannot post anything more specific, this should be categorized as "anecdotal". However it was sufficient to convince the client (customer of both Oracle and BEA) to abandon MTS forever (and, incidentally, MTS isn't even a good or correct name, which is why it's known as "shared server" these days).

"Here [spreadsheet attached] is some experimentally controlled data that I collected on the performance of MTS and Java. For the first time, I can say that I have seen some hard evidence of performance problems due solely to MTS vs dedicated server for Oracle. What's VERY interesting is that the difference is MUCH more pronounced when it's a matter of Java and Oracle processes running concurrently [on the same server]. The difference was almost negligible for pure Oracle-to-Oracle transactions (meaning, on our production database server, this experiment does not predict any great difference due to MTS, which I think we've seen to be the case).

[...]

"Results: for the portion of [...] which involves creating new tables from old tables, there was a very slight improvement under the dedicated server config. Also, the tests without MTS were more consistent (lower standard deviation). For the portion of the [...] run that involves Java code (the XML extract), the MTS results were 2-4 times slower, and the deviations (variance) were signifcantly higher."

-Mark Bole Received on Tue Sep 13 2005 - 20:30:36 CDT