Hi,
Any chance the slower one with the standby is set to force logging while
the faster one isn't? Are there nologging operations in the database?
Harel
On Thu, Apr 23, 2020 at 2:29 PM Paul Houghton <Paul.Houghton@xxxxxxxxxxxxx>
wrote:
Hi
We have some new hardware which seems slower from an IO perspective than
the old hardware, so I downloaded SLOB to investigate, and ran it on two
databases which are copies of each other. Both on RHEL7 OS with Oracle
12.2.0.1 and the April critical patch. Both are in archive log mode.
Looking into the AWR, it seems the slower one on the new hardware is
generating more redo per transaction than the old one. I set up SLOB
identically in both databases (./setup.sh IOPS 8) and ran it identically
(./runit.sh 8). The tablespace is a smallfile and I specified it’s
location, but otherwise I used the defaults. I set up the admin user as
“sys/sys as sysdba” and for this run changed the run time to 900 seconds.
There are some differences in configuration. The slow one is supposed to
be more like production than the faster one, so it has:
· bigger buffers the log buffer is 10M as opposed to 5M in the
faster one
· uses huge pages, where the fast one doesn’t
· LOST_WRITE_PROTECT is NONE in both environments.
· a physical standby database. The faster one doesn’t
There are applications running against these databases, but I can’t see
that there was any application SQL that would make this much difference.
I am looking in the load profile section of the AWR report that is
generated. Redo size per transaction is 54K for the faster one, and 203k
for the slower one, so about 4x as much. What could be causing this?
I’d be grateful for some pointers as to where I can look to see what is
causing the extra redo.
Thanks
PaulH
