Hi
I think you should capture RAT for at least 3, 4 hours of workload. In the
old days when load tests were using tools such as Mercury Loadrunner there
was a stage in the beginning which we called ramp-up which is the warming
up stage you are describing to bring in live-like conditions. During
ramp-up response times are higher and are taken into account when analyzing
the data.
So to simplify things, run AWR every 30 minutes and consider the first 30
minutes as your ramp-up stage.
Thanks
On Sun, Oct 24, 2021 at 12:32 PM Lok P <loknath.73@xxxxxxxxx> wrote:
Hi, We are planning to use the real application testing feature of Oracle
for capturing and replaying production workloads to see the actual impact
of one of our coming versions and hardware upgrades. We have one question-
As before each workload replay the RAT database has to be flashed back to
the point in time where we have captured the workload from. So will that
flashback, remove everything from the RAT DB buffer cache and also
everything will be wiped out from the storage server flash cache? And in
that case the replay may not mimic the exact impact as the cache warmup
will be adding additional overhead to the resource and time and thus it may
not be a good comparison of before and after upgrade behaviour of RAT
database. Can you please confirm if this is the correct understanding and
how we should replay then to avoid this cache warmup thing? And if any
other RAT system related resources can also deviate the figures?
Regards
Lok
