Re: Significant performance difference between v8.0.6 and v8.1.7?

NOARCHIVELOG Mode on both. Additionally, all the user tables have NOLOGGING.

Redo Log files = 64MB each (8 times) on the v8.0.6, presumably default on the v8.1.7 (which I believe is 3 Redo Log files of 1MB each).

Import on v8.0.6 with COMMIT=Y using the command line interface. OEM presumably used on v8.1.7 and I don't know what the defaults are there, but I think the customer used COMMIT=Y there as well. Additionally, the Import on v8.0.6 was done in two passes - ROWS=Y, INDEXES=N on the first pass, ROWS=N, INDEXES=Y on the second pass - whereas the Import on v8.1.7 was done in a single pass. BUFFERS=50000000 on v8.0.6 and I think the customer used the "MAX" setting in OEM on the v8.1.7 (which should be right around 50,000,000).

Like I said ... according to all the pictures in the book, the v8.0.6 installation should be the better performer. Less checkpointing, indexes imported separately, I/O spread out across 4 SCSI drives, more memory, more processor. I just don't see resource contention as the cause for the difference in Import times. It's got to be (in my mynd, at least) either some really poor code from Oracle or an O/S driver issue.

"Niall Litchfield" <n-litchfield_at_audit-commission.gov.uk> wrote in message news:3cdfa9c2$0$8506$ed9e5944_at_reading.news.pipex.net...
> "MyndPhlyp" <nobody_at_home.com> wrote in message
> news:abo8m5$ett$1_at_slb2.atl.mindspring.net...
> > I suspect the performance problem is one of two possibilities, but would
> > appreciate other ideas (or confirmation I haven't yet lost my last
> marble):
> >
> > * The SCSI host adapter drivers may be out-of-date or improperly
> installed.
> >
> > * There were very significant performance issues with v8.0.6 that were
> > resolved by v8.1.7.
> >
>
> Difference in archive log mode/redo size? also doing import with commit=y
or
> commit=n?
>
>
>
>
Received on Mon May 13 2002 - 07:13:45 CDT