Skip to Main Content

Exadata Office Hours

Free tips and training every month! Subscribe for reminders and more from Office Hours. FAQ

Header container

March 12, 2020

17:30 UTC   Start Times Around the World

Subscribe to be notified of changes to sessions and give us feedback!

Having trouble watching the video on this page? Open the video in your browser.


Cecilia Grant's Database Performance Tuning Tips
Join Cris and Gavin for a chat about database performance tuning with Cecilia Grant (Architect, Oracle Development). Please review Cecilia's performance tips post on the Exadata blog (link below) ... we bring you the expert, and you bring the questions!

00:11 Welcome
00:43 Blog post
00:58 Discussion is about Database Performance everywhere, not just on Exadata
01:39 Gavin introduces Cecilia Grant
02:34 Safe Harbor slide
02:49 The 2-3 most important performance tips
02:57 -- Let the data guide you
03:50 -- Database tuning doesn't change much for Exadata
04:20 -- Keep good baselines
05:09 -- Recap & discussion
06:08 Should you tune performance on Exadata?
06:25 -- "Smart features" let you get away with a lot, but ...
07:01 -- You still need to write good SQL and design schemas properly
07:32 What is a Storage Index - the best I/O is the I/O you don't do
08:19 -- Storage Indexes/Indices are ephemeral - how are they rebuilt?
08:41 -- Conventional Indices/Indexes are still relevant - persistent and less CPU cost
09:19 How do I keep a baseline?
09:52 What is IORM? Should I use it?
09:55 -- IORM is on by default on Exadata
10:09 -- IORM useful to control I/O when hosting multiple databases
10:34 -- Backtrack: IORM defined
11:05 -- Plans and tiers with IORM
11:24 -- Examples of tiers
11:51 I/O outliers on Exadata?
12:13 -- Make sure it's an I/O problem
12:26 -- We expect workload to be distributed evenly
12:59 -- We saw recently a case like that, but...
13:20 -- Normally you reach max capacity on all disks
13:35 -- Apply top tips: use AWR, look at the data
14:15 Where do I find data showing Exadata performance benefits?
14:28 -- Reducing I/O, AWR shows SI savings and CC savings
14:45 -- The SQL monitor report - cell offload efficiency
15:08 -- Comparing to non-Exadata machines
15:33 What about performance on Autonomous DB?
15:50 -- You get an AWR report, but restricted to the PDB
16:12 -- The data you get is ...
16:25 -- Performance tuning?
16:52 -- Concentrate on the upper level / application
17:07 How to architect your data?
17:29 -- Normal forms are good but denormalization may be useful
17:48 -- Beyond the model, look at access patterns
18:01 -- Access patterns, explained
18:34 -- Example: model too complicated
19:24 Persistent Memory (PMEM) in write-through instead of write-back mode?
19:47 -- How it works with flash - write-through
19:58 -- How it works with flash - write-back
20:03 -- How it works with PMEM
20:20 -- Why use write-through with PMEM
20:32 -- -- DB writes are asynchronous, i.e., user doesn't wait for them
20:53 -- How is it different from flash?
21:54 Is the PMEM cache write-through by default?
21:58 New features planned for PMEM?
22:22 -- See also last month's AskTOM on Exadata X8M, with Jia Shi
22:49 -- We used PMEM in the architecturally best manner
23:20 How Cecilia got into performance tuning?
24:13 What it takes to work in performance
25:09 How do the customer experiences connect with our technology design and tuning?
25:30 -- A number of teams bring different facets of experience with the product to dev
26:08 -- Similarly with the Support escalations
26:51 If PMEM is not available, can I still benefit from, e.g., RAM cache features?
27:15 -- Flash speeds v. PMEM speeds
27:34 Websites, whitepapers and other resources?
28:45 Your ideas and questions for topics and experts are welcome
29:00 Do old DB performance methods / tools still apply?
29:27 -- Example recent escalation
31:12 Can Storage Index be sto

Your Experts

    Cris Pedregal

    Cris Pedregal

    Cris Pedregal joined Oracle as a senior developer in 2005, working on the design and development of high availability functionality, compression, and the initial cloud deployments of the Oracle Database. In product management he launched product features and evangelized Oracle technologies. He also managed company-wide technical relationships with key external partners, including a major cloud computing company and CERN. Before Oracle Cris was on the Computer Science faculty at the University of New Mexico, Albuquerque. He speaks one language with 2.5 accents.
    Gavin Parish

    Gavin Parish   

    Gavin is a product manager in Oracle’s Exadata team, with a focus on Exadata’s software and hardware roadmap. Prior to the Exadata team, Gavin was a founding member of the team responsible for launching the industry’s first on-premises public cloud technology - Oracle Cloud at Customer. In his 15 years in Oracle Product Management, Presales and Consulting roles, Gavin has developed a robust understanding of all things Oracle, helping customers architect and implement a variety of infrastructure and application technologies.
    Cecilia Grant

    Cecilia Grant

    Cecilia is a Performance Architect in the Exadata Development Group, working on customer performance escalations and performance diagnostic tools. She has been with Oracle for over 25 years, in various capacities focusing on performance; having worked on customer benchmarks, database performance and server manageability.


We are running multiple databases on Exadata X6-2 and everytime we had applied bundle patch on Exadata, we seems to lose all the learnings and it takes upto two weeks before performance of DB is back to pre-patching. How to avoid this learning (cache) post patching on Exadata?

Not quite enough details to go into it here, (are you applying to storage, db node, GI, DB, rolli Show more.

Answered by Gavin Parish -