The Exadata system is specifically optimized for database workloads. But
the key [Exadata specific] features that we leverage for our workloads is
Exadata Smart Scan, Smart Flash Cache, Hybrid Columnar Compression, and of
course all the 12c Oracle Database which are integrated with Exadata stack;
e.g, Multitenant/PDB.
We have consolidated a large portion of our disparate database deployments
from across the enterprise onto the Exadata. This has saved us and our
customers Capex and datacenter floor space. But more importantly, it's
forced us to rationalize and standardize on a specific software stack. For
example, earlier we had eight or nine different versions of Oracle Grid
Infrastructure and Oracle Database. By going to Exadata platform, we forced
the best practice to a single stack version.
The key areas of improvements are better documentation, Exadata hardware
stack monitoring, and improved awareness (self-tuning) application/SQL for
Exadata. For monitoring, most of the admins had difficultly with the OEM
Exadata Plugin, and thus we wrote our own monitoring. In addition, most
customers have custom network configurations, these don’t always fit into
the Exadata “out-of-box” supportability. Oracle ACS had to work
particularly closely with Network team to resolve firewall, port-forwarding,
DNS name resolution issues, etc.
We have had issues with storage cells failing and rebooting, but with proper
configuration and redundancy decisions (upfront) we barely noticed the
event. Nevertheless, when we had component failures it was difficult to
diagnose and triage.
No, this is one issue we didn’t encounter. We decided on the appropriate
Exadata configuration and we deployed and consolidated. Moreover, Exadata
has Elastic configurations as well. Nevertheless, there are some internal
areas where the admins need to set and rationalize perception. Some SQL
is poorly written and even Exadata can’t fix that. The perception is that
Exadata is a single “red” pill to cure bad code, that’s just not the case.
Oracle Exadata Tech Support has been great. We had many issues with Oracle
ACS and Platinum Support; specifically around the areas of scheduling
patching and upgrades. For these reasons, we ended building that skillset
in house.
We had disparate configurations, ranging from IBM AIX P-series, HP systems,
and older Solaris systems. We began our consolidation approach and
standardized on the Exadata for our Oracle Database 11.2 and above systems.
The initial setup required a lot of planning from all stack parties. All
features and configuration need to be discussed upfront. However, with
detail planning, the execution and deployment, via “OneCommand” process, was fairly straightforward.
This is where the Exadata hangup occurs. Not only are users paying for
Oracle EE and options such as RAC, and most of the must-have features such
as Advanced Compression, and Partitioning; but also the Exadata Storage
software. Add in the Support costs for all that storage. However, Oracle doesoffer “capacity on-demand” options, which do mollify the compute
licenses (not storage licenses) to an extent.
Yes, we looked at Hyper-Converged Systems and UCS compute with
All-Flash-Arrays. And we did buy those systems for specific applications
that didn’t fit the Exadata solution.
Understand pricing and cost of skillset acquisition. In addition, define the
organizational layout for supporting the Exadata, since Exadata is a
“converged and engineered system” per se.
