Infolob Solutions is the premier Oracle Exadata implementation partner in the US today. Oracle Corp constantly features us as their go-to partner and we have more than 50 engagements where we have helped companies implement, manage, and optimize their platforms using the Oracle Exadata features available to them.

Today, Infolob’s resident Exadata guru and our CTO Tim Fox writes about a few key features that are worth knowing:

Oracle Exadata features hidden in plain sight

Exadata has a few key features that are sometimes overlooked by those who either don’t know they exist or have trouble finding them. What sets Exadata apart as a platform from others is how it executes queries, using something called a smart scan.

A smart scan is a full segment scan the platform does use the direct path read operation. A smart scan on an Exadata can be anywhere from tens to thousands of times faster than the same query running on a traditional platform.

There are three types of smart scans:

1. Column projection:

   The smart scan will only return data from the columns in the select list. The machine still scans the whole segment, but it only transfers the data in the columns referenced in the select list over the storage network to the computer nodes. This method reduces network traffic between database servers and storage.

2. Predicate filter:

   This only returns the data from the row(s) that match your where clause or predicate. You can use column projection and predicate filtering in conjunction to further enhance query performance while also dramatically reducing storage area network traffic.

3. Storage index:

   When scanning storage with a smart scan, the Exadata improves speed by identifying which storage areas do not contain what is being queried. To do this, the Exadata keeps a storage index, or an in-memory structure that keeps track of enough information to determine which areas of the disk contain data relevant to the query.

Most Exadata users miss these features right off the bat:

Small table threshold: The machine will use direct path read to scan any table that is bigger than the STT during a full segment scan. If it’s smaller, it won’t. If your buffer cache is too big, your STT is large as well, so you won’t get smart scans except for large tables. Reducing the buffer cache size makes your small table threshold smaller, and therefore allows you to utilize smart scans for more tables.

Full segment scans are an absolute requirement for using Exadata smart scans. Not as much data in the buffer cache forces you to go to disk, which is where all the Exadata features are. In this case, memory isn’t necessarily faster than disk, which causes disbelief and skepticism among experienced DBAs.

IORM feature: IO resource manager. Exadata is the only place in Oracle where you can run a database, specify the importance of certain databases, and prioritize their IO resources. Many customers largely ignore this feature.

These Oracle Exadata features are often counterintuitive to an experienced DBA, which is why they get swept under the rug so often. Since a lot of people don’t even know what they’re missing with their Exadata machine, their opportunity cost can be enormous.

Exadata implementation tips

• Size the Exadata appropriately
  • For consolidation, take the number of homes into account, or how many binary sets you’re going to load on the Exadata
  • Reducing SGA/buffer cache size promotes consolidation—faster with fewer resources used
  • Get the disk group sizing right—DATA/RECO/SPARSE
• Build databases using the Exadata template from DBCA
  • Forklift migrations will not have Exadata-specific parameter settings (which is not ideal)
  • Exachk reports will complain about invalid database parameters
  • Disable adaptive optimizer plans (smooths your performance profile)
  • Gather system statistics with Exadata parameter (does an IO calibration without actually running the IO calibration process)
• If you require network separation, use virtualized Exadata
  • Exadata compute nodes are configured with Oracle virtual machine (OVM)
  • Each OVM cluster has its own disk group (almost total separation)
  • All Exadata cloud machines are configured with OVM
• Consider a backup/recovery strategy
  • Exadata storage is expensive for backup (consider ZFS/ZDLRA)
  • ZFS is best for backup/recover of a few very large databases
  • ZDLRA is best for backing up hundreds to thousands of Oracle databases
  • We highly recommend external backup via Infiniband (ZFS/ZDLRA)

The Exadata platform is most effective in the following areas:

• ERP: Oracle E-business Suite/PeopleSoft/SAP
  • Exadata handles OLTP very well and excels at batch/reporting
  • Most of Infolob’s customers run ERP on Exadata
• Data warehousing
  • Hybrid columnar compression (HCC) is very effective—this function collapses columns (rather than rows) into compressed files and stores them in an Oracle block
  • Partitioning together with HCC is the basic information lifecycle management
• Business intelligence
  • Large data sets with heavy computational requirements—Exadata vastly improves business intelligence reporting

Exadata in the cloud

There are multiple cloud implementations for Exadata if you choose to go that route.

1. Oracle Cloud at customer/Exadata cloud machine: We install the Exadata in your data center but Oracle manages it. This looks like bare metal, shows up virtualized, and comes with a cloud plane, which is a controller that allows you to add and subtract CPUs. We consider this a private cloud implementation.
2. Oracle Exadata Cloud Service (ExaCS): This machine exists in the public Oracle Cloud but is used by you.
3. True private cloud (right to use): You lease a platform from Infolob and you can run it like bare metal in your data center. At the end of your lease, you get another one rather than buying it.

Each option has pros and cons, and thankfully Oracle offers a full range of cloud and bare-metal solutions.

Exadata machines are easy to implement and are known for their extreme performance. Infolob often sees machines running tens to thousands of times faster than their incumbent systems. There is ample support available for Exadata versus custom systems, as well as on-prem, private, and public cloud versions.

Are you thinking of implementing an Exadata? Do you have one already and just want to take advantage of its features? Either way, Infolob is here for you. Write to us at engage@infolob.com.

This post was compiled by Carson Collins from CTO Tim Fox’s recorded webinar. You can reach Carson at carson.collins@infolob.com and Tim at tim.fox@infolob.com.