Consolidation onto Exadata Cloud@Customer on premise cores-->OCPUs

Franck Pachot

2018-09-02 11:18:16 UTC

Hi,

In your list of consolidation options, I agree with all except: 2. Oracle
VM.
I understand that the infrastructure team fears to manage another
hypervisor but OVM is not so difficult when used for simple things (here
you just need CPU pinning). You have RAC databases, that's more complex to
manage. If you want to reduce complexity, check if you really need RAC. In
my experience, OVM was the right choice to do capacity on demand for
customers where ODA was not a solution (e.g because of lack of flexibility
in editions, in storage capacity). However, you are right that a cloud
solution may be easier to present to financial folks, and Oracle gives good
discounts when moving on-premises licenses to Cloud credits.

You didn't mention any DR solution. RAC protects the service, but not the
data. DR solutions will be different in Standard and Enterprise editions.
Do you plan to have two ***@customer Exadata? That's big hardware to run
40 databases.

Now about OCPUs, that's not easy. This 30-40% reduction means nothing. Do
not count on reduction when going on Exadata as lot of Exadata features are
there to use more CPU: parallel query, HCC compression,... You save on BI
queries if they are offloaded to SmartScan which really depends on your
workload. But when you consolidate all your databases, there's probably
only a small part of it which benefits from SmartScan.

The major reduction comes from Cloud services and consolidation. Probably
not from multitenant as most of your databases will not share the same CDB
(you have some Standard Edition, you have different versions, maybe
different charactersets, and will still separate prod environments). You
will probably run one database service per database. However, while
on-premises you size each server to cope with the peak activity, the cloud
service allows you to stop some instances when not used (e.g a test
environment used once per month) or run with low OCPU and be ready to
increase only when needed. If you have access to autonomous services, you
can even do this scale-up without application interruption.

Those are just ideas to think about. Choosing and sizing a different
infrastructure is something more complex.

Regards,
Franck.

Post by anthony Sanchez
Hello Folks,
I'm trying to figure out what my OCPU requirements will be if I migrate
multitenant)
I've heard from one exadata customer that you can expect a 30-40%
reduction in db compute cores needed. I anticipate that multi tenancy will
help the reduction of CPU cores needed as well. I understand that the
gains realized will depend on types of workloads and other factors. I'm
hoping others can share their experiences with regards to cpu core
reduction (for different workloads and overall) and also general
appreciated!
I work for a municipality. We are an Oracle and SQL shop and have 2 DBAs,
each of which wear many hats.
a mix of 11.2.0.4 and 12.1.0.2
both Oracle Linux and Windows - ODA's and HP servers
some standard edition
enterprise edition plus partitioning and some RAC
We have roughly 25 oracle databases, and are about to kick off a big
project that will add 5 more databases. We foresee growing by another 5-10
databases over the next few years. Our workloads are diverse and the
majority of our databases are under 200G, with a handful that cross into a
few TB.
We keep buying hardware to satisfy different licensing combinations. Our
leadership is not comfortable with moving workloads to the public cloud and
it must remain on premise. We have a new system coming on board with
significant Oracle hardware and licensing requirements, and I thought
perhaps this might be an opportunity to pick a solution that would enable
us to consolidate across the organization.
I want to consolidate, not worry about licensed options anymore, make
patching easier with less downtime (2 dba's, many hats), have room for
growth over the next 5 years, and have the TCO be the same or better than
the traditional HW/SW approach we have now. For the TCO I can't assign a
dollar value to things like better up time, improved security, time saved
through easier administration, all packs are included, etc. Finance folks
will reject those numbers.
1. vmware - licensing kills us
2. oracle vm - infrastructure team doesn't want to support.
3. ODAs - can't support different licensing combinations effectively
4. purchase exadata - seems to add more administrative work to our
plates based on discussions with other entities
to calculate TCO
thanks!
Anthony