Oracle OpenWorld 2012: Monday

Database 12c Features

Following yesterday's big announcements and new rumors I made a few schedule changes ensuring I allotted more time in the demo grounds to talk to various Oracle specialists concerning Database 12c. The demo grounds are truly amazing with a wealth of contacts to be made and things to be learnt from talking to the various "informed" vendors (including Oracle). Even more amazing are the large number of people that come solely for the purpose of winning free stuff - not that I'd complain if I won one of the numerous iPads on offer or even better the $10,000 offered by EMC (good to know where my companies money is going). I'm curious if any of these vendors do any analysis on the "real" contacts made vs. just those looking for stuff and follow-up sales made.

There are many new and usable features with 12c, and I would argue this will be the biggest release Oracle will have done to date (when it ships) in terms of changes and features. The upgrade process via DBUA has been given some attention with parallelism during the upgrade itself, fix-it scripts, resumption from some failures (instead of starting from scratch for everything) and a post-upgrade health check. Transportable Tablespaces (TTS) via Data Pump will be more efficient by automatically running all pre-requisite checks, and doing the full export and import. Meaning it figures out all the metadata dependencies, creates all the users, objects, grants, etc. from the source unto the target then copies across all the data and viola!

Some questions (but not all) surrounding Pluggable Databases (PDB), which I've mentioned yesterday, were  answered today as well. It will support pre-12.1 databases (only 11.2.0.3 is my guess and based on the slides used) which will plug into a 12.1 container database (the housing or hypervisor database if you will). All databases can be backed up as one and recovered separately including point-in-time (PIT). PDB can also be run in standby setups though I'm still left wondering how PDB works exactly in a RAC environment. Migration into this architecture appears to be done via Data Pump (I'm guessing TTS since otherwise it would be long migration). Resource utilization is handled by Resource Manager (DBRM) so processor (and other settings?) usage can be allocated to each database. Patching/upgrades can be done separately to each database though I'd imagine the container DB must always be at the highest release (similar to Grid Infrastructure). A question, out of many I have, is how does affect Oracle VPD and will it be a paid feature? (we all know the answer is yes)

Another very interesting and immediately usable feature is "Automatic Usage Based Compression". Essentially a heat map of table partitions is used to compress various partitions based on usage/activity (INSERT, UPDATE and DELETE statements) using some user defined policies. Compression is done online, in the background. Does this mean HCC is open to all now? Is this using DBMS_REDEFINITION under the covers for the online compression change? What about compressing table blocks and not just partitions? Will the threshold for hot, warm and cold be adjustable (there is always some hidden parameter)? Is this part of the compression package/option and how much will it cost?

Redaction of Sensitive Data is another big feature. This moves the masking of sensitive data from the application level to the database level where the DBA does this change online and immediately (no logoff/logon required) based on set policies. I'm left wondering how this affects the Data Masking Pack and Label Security? (and again, how much?)

A new feature (which is actually available now supported for Exadata) is RMAN Cross Platform  Incremental Backups. This is using RMAN to do a platform conversion from a big endian platform such as Solaris SPARC, IBM AIX and HP-UX to Linux using backups/restores where the incrementals can be applied to the target until ready to switch platforms at which point the actual switch over is considerably less time (and effort). Note 1389592.1 explains this with greater details.

The Jimmy Cliff Experience

Whomever thought up the Oracle Music Festival deserves a raise, and even better, the party responsible for bringing in Jimmy Cliff deserves a promotion! The man, Jimmy Cliff, must be in his 60's at least but has more energy than a 20 year old! He completely rocked the house with his energy, song arrangements and charisma. The crowd (including myself) was completely involved and so much so enjoyed his performance that an encore was demanded, and graciously accepted.

I had another big day ahead of me and so left during his encore. I can already feel the pain in various body parts following my "dancing" (I use the term loosely). I good end to another great day at OpenWorld...

Oracle OpenWorld 2012: Sunday (continued...)

Sessions and Keynote

I attended a few sessions which for the most part were informative in some way. One take-away was that Oracle Enterprise Manager 12c Cloud Control is very popular as it is on a lot of minds (mine including since I'm trying to get my company to bypass our implementation of 11g and go straight to 12c but I digress). The few sessions I attended had quite a lot of questions. I was also able to meet a few of my Twitter contacts (@dbakevlar @aakela and @fuadar) which was awesome!

An interesting nugget in the session "Will it blend? Verifying Capacity in Server and Database Consolidations" was that of 'Consolidated DB Replay'. This is feature introduced in a patch (13947480) for 11.2.0.2+ which (as the name suggestions) allows for the concurrent running or replay of multiple captured workloads. The use case for this, of course, is capturing workloads from multiple source database systems and replaying on a single target system which is intended as a consolidation database. Ideally the capture time and periods across the multiple source databases should be the same to get the best picture of what the consolidated workload would look like on a single consolidate database. This feature would replace (or minimize) manual efforts involving visually analyzing workload graphs using  OEM (as an example) for each database or looking at consolidated/merged AWR information for the multiple source systems.

Keynote

There have been many rumors surrounding a new Exadata 1/8 rack configuration, Exadata hardware upgrades (would there be an X3-2) and database 12c (if it would be announced or not). I did not attend Enkitec Extreme Exadata Expo (E4) 2012 but did read some of the Tweets and postings concerning the sessions which did point to such announcements. As it turned out, the Larry's Keynote did not disappoint and confirmed the rumors with the announcements of Exadata X3-2 (including an Exadata X3-2 1/8 rack), Oracle Database 12c release sometime in 2013 (my guess is some features will be cut for a January/February released in 2013), Oracle Private Cloud and IaaS.

Oracle Private Cloud is an offering for companies needing their own private infrastructure which can either run externally at Oracle facilities or inside the companies own data center but managed completely by Oracle. Having experienced various Oracle Support Services (OCS, OCMS or Oracle On-Demand) I can say the success for this offering will depend heavy on improvement for these support offerings and clear understanding between all involved parties as to what is meant by "managed". Oracle Cloud is Oracle's Infrastructure as a Service or IaaS offering composed of Exadata, Exalogic, Oracle Linux, Oracle VM, Oracle Storage, and InfiniBand (IB) components.

Oracle Database 12c which will be released sometime in 2013 (January/February is the whisper) will have some interesting features (not sure if I disclose at this time other than what has just been announced) such as Pluggable Databases. This is essentially multiple databases sharing the same server using containerization at the database level therefore being more efficient (so not complete separation in terms of processes and memory) and not requiring any application changes. For those familiar with SQL Server, PostgreSQL (including Netezza) and other such database platforms this is not anything knew. It is however, in the context of Oracle databases and has several benefits in the area of consolidation and hosting.

Exadata X3-2 was announced as the hardware refresh for the previous generation X2-2 along with a new 1/8 rack deployment option (starting price for negotiations is $200,000, nice!). Strangely enough, there is also an Exadata X3-8 as the refresh for Exadata X2-8, but this got no recognition (perhaps these do not sell as well and are a niche offering?). A few quick overview specifications are below:

Database Nodes

• up to 8 x Oracle/Sun X3-2 servers
• up to 2 TB RAM or 256 GB/node
• up to 128 cores using 2x8-core Intel E5-2690 (2.9 GHz) per node

Storage Nodes

• Up to 14 x Sun X3-2L
• Up to 168 cores using 2x6-core Intel E5-2600 series per node
• Up to 22 TB Flash memory
• Up to 168 x 600 GB 15K rpm HP or 168 x 3 TB 7.2K rpm HC HDD

In terms of performance a full rack X3-2 should scream with:

• ~50K IOPS using 8K IO requests (most vendors use 2K so be careful doing comparisons)
• 100 GB/s bandwidth taking into consideration HP HDD and Flash
• 16 TB/hour data loads (from past experience this is within the same array so again, be careful and ask specific questions).

A software upgrade to the platform (this means available now w/o upgrading to X3-2) brings Cached Writes along with the previous Cached Reads. So maximum IOPS for 8K IO requests involving Flash is ~1,500,000 for read and ~1,000,000 for write. With compression (your mileage will vary according to your data) numbers should be better but again, test, test and test again. Your workload was not used when obtaining these benchmark figures. Usable disk capacity is ~45 TB for HP and ~224 TB for HC HDD. Also, Oracle Cloud and Private Cloud will start with Exadata X3-2 systems.

For me, here is what is missing from Exadata or what I'd like to see:

• A more appliance-centric approach where even the ASM and DB configuration is standard and factory setup (sorry, but OCS involvement/engagement would be minimized)
• More work being pushed down to the storage level (more analytics, more parallel processing, more "transparent" indexing so I don't have to create and maintain)
• Automatic data compression (can still provide better/advanced compression levels at cost)
• Built-in Hadoop integration (storage nodes as data nodes and a dynamic compute as the named node?)
• Integrated monitoring via included OEM appliance as either an included 2x1U server in the rack or external servers. You can argue you can just use existing or build your own but wouldn't it be nice to have this option? Quite frankly I'm puzzled as to why Oracle has not come out with an OEM appliance yet and have suggested as such to some powers in the OEM team (also wondering about a MySQL engineered system, ExaSQL :-))

Fujitsu Keynote
Moving on (or back) to the Fujitsu portion of the keynote, I found it most interesting since they are doing very similar work with their "Fujitsu Agricultural Cloud Service" to that of my company (though not quite as full-featured a service if I do so say). The services gathers data collected by farmers via various devices and runs analysis which will aid in providing information to improve yields. This is being done very cost effectively and near real time. Then there is project "Athena" which is the merging of hardware and software (OS and database if I understood correctly) to bring forth a new processing model which will far surpass anything currently available. Leveraging knowledge and technology from the K supercomputer, Liquid Loop Cooling (LLC), 512 GB per socket (32 TB per system), 4 CPUs w/2TB each scaling/connecting in a building block fashion (up to 16 blocks?) and software on chip (database software also in silicon) the SPARC64 X was/will be born in 2013. Testing has shown a 2x increase in performance over IBM Power7 though no specifics were given (it was just a keynote). I do love  how they showed real world type scenarios and business usage instead of just pure tech.

So far that has been my OpenWorld 2012 experience to date. Sunday down, next up Monday to Thursday.

Oracle OpenWorld 2012: Sunday

MySQL Connect

First up today was a visit to the MySQL Keynote featuring some speakers from Twitter, Paypal and Verizon Wireless. I've been interested in MySQL for sometime but never really played with it for myself. My company is seeking to investigate it further since we've got over 50 instances of it running, but also to look into open source alternatives (cost reduction mainly).

Very interesting what these large companies are doing with MySQL and how they are using the technology. Twitter uses it extensively because:

• It is fast, even compared with NoSQL alternatives (depending on what you are doing with it of course)
• It has very low latency
• It scales well
• It has a large ecosystem
• The safety of InnoDB (i.e. it does not lose data)

It is not all roses however, as there were some words of caution::

• It is not yet optimized for SSD (which they are working on)
• There is room for a configuration management tool
• It is not a complete solution and should be used as a building block
• It is not a purpose-built key-value (KV) store (so try other NoSQL if that is your true requirement)
• It is not schema-less (both pro and con)
• There is a need for better performance/response time metrics
• There is a need for better monitoring

What they could share of there environment is also very interesting:

• 25 traditional master-slaves
• 3-100 machines with 6 DBAs and 1 developer
• > 6 million queries per second (qps) w/400 million tweets per day (+ metadata)

The talk from Verizon Wireless was also  interesting  as they apparently use MySQL for their intranet with a customized landing page for each user (based on location, function, etc.) which the user can further customize. It is all highly performant (since that was a main criteria) and scalably. Their "Verizon Infrastructure as a Service (IaaS) Group" was inspirational! An internal group which provides IaaS to the rest of the company.