Friday Oct 02, 2015

Questions you asked: What happens if a column is not In-Memory?

When we're talking to customers or giving presentations questions often get asked that seem simple, but could have an answer that is worth showing with an example rather than just saying, "yes, it works this way". So I'm going to start a series of posts that will address those questions. I will continue the series as I come across what I think are interesting questions.

This first post in the series of "Questions You Asked" will be "What happens if a column is not In-Memory?"

[Read More]

Saturday Sep 12, 2015

Oracle Open World 2015 Time to plan your schedule!

There are only 6 weeks to go until Oracle Open World, the largest gathering of Oracle customers, partners, developers, and technology enthusiasts, which begins on October 25th in San Francisco.

Of course the In-Memory development group will be there and you will have multiple opportunities to meet up with us, in one of our technical sessions, our hands-on-labs or at the Oracle demogrounds.

This year the In-Memory team has 4 technical sessions and there are also 5 excellent customer sessions you shouldn't miss.

[Read More]

Monday Sep 07, 2015

Star Schema Challenge - Part 4

In my previous post Star Schema Challenge - Part 3, I reported the first of the in-memory performance results with 25 users querying the star schema in the In-Memory column store (IM column store). In Star Schema Challenge - Part 3.1 I did the same for the de-normalized fact table in the IM column store.  This serves as a comparison between the row format in-memory (buffer cache) and IM column store. For this comparison I stopped at 25 users because it became clear this was enough users for the row store table.  Now it is time to start working towards 100 concurrent users.

[Read More]

Friday Aug 28, 2015

Popular Statistics with Database In-Memory

Throughout our previous posts we have mentioned various session level statistics that are available to help identify what is actually going on with Database In-Memory. Since these statistic definitions didn't make it into the Reference manual I was asked recently if there is a list anywhere that defines what these statistics are. Since there isn't I thought I'd post a summary of the popular statistics that we've mentioned in the blog and that you might want to know about.

[Read More]

Friday Aug 21, 2015

Understanding new In-Memory notes in an execution plan

If you have started to play around with Database In-Memory, chances are you have been examining execution plans to see if the optimizer has chosen to use the In-Memory column store (IM column store) or not. In addition to the execution plan itself, you should also check out the NOTE section under the plan, which contains more information about how a query was executed, such as if dynamic sampling was used during the query optimization.

[Read More]

Friday Aug 14, 2015

Star Schema Challenge - Part 3.1

In my previous post (Star Schema Challenge – Part 3) I revealed the first of the Database In-Memory results, 25 users querying a 500 million row star schema with tables in the In-Memory column store(IM column store) on a commodity Intel-based server with 60 cores and 1TB of DRAM.

In this post I will reveal the first of the in-memory results for the de-normalized version of the fact table.

[Read More]

Wednesday Aug 05, 2015

New White Paper on using Database In-Memory with the E-Business Suite Released!

We would like to draw your attention to a new white paper has just been released titled "Using Oracle Database In-Memory with Oracle E-Business Suite".

The white paper details how Database In-Memory works, what the requirements are to use it with the E-Business Suite and works through some use cases and examples.  More details can be found in the My Oracle Support (MOS) Note 2025309.1.

The EBS team worked on hard on this paper and we think it's a great source of information for E-Business customers who want to take advantage of Oracle Database In-Memory. Enjoy the read!

Tuesday Aug 04, 2015

Oracle Database In-Memory on RAC - Part 4

Setting Up Independent In-Memory Column Stores

In previous posts we've talked about how to use RAC services to enable the IM column store to be run on a subset of nodes in a RAC environment. We also mentioned that it is possible, using RAC services and the DUPLICATE sub-clause on engineered systems, to enable rolling patches and upgrades when running the IM column store on a subset of nodes.

In this article we're going to talk about how to set up independent IM column stores on a RAC cluster using services and some database initialization parameters.

But first let's ask and answer the question, why would you want to do this? This might be a good idea if you are trying to enforce application affinity at the node level and don't want to allow inter-instance parallelism.

[Read More]

Friday Jul 24, 2015

What to do with optimizer statistics when upgrading to 12c to take advantage of Database In-Memory

Before most customers can take advantage of Database In-Memory they will need to navigate the tricky terrain of a database upgrade. One of the most challenging aspects of an upgrade is figuring out how to minimize performance regressions due to execution plan changes.

And if that wasn’t enough to handle, the introduction of Database In-Memory into a 12c environment has the potential to change even more of the execution plans. So, what should you do?

[Read More]

Tuesday Jul 21, 2015

Oracle Database In-Memory Bundle Patch 10 Released

The latest Bundle Patch for Database In-Memory has been released. The Bundle Patch is 21188742 or Bundle Patch 10 for Engineered Systems and DB In-Memory (July2015)). This Bundle Patch improves the performance of mixed workload environments (OLTP & DW workloads), as well as enhancing the performance of analytic queries with aggregation. More information on the latest Bundle Patch can be found in the MOS note 21188742.8 or in the Mos note: Bundle Patches for Engineered Systems and DB In-Memory (Doc ID 1937782.1).

[Read More]

Friday Jul 17, 2015

Star Schema Challenge - Part 3

In my previous posts on the Star Schema Challenge, I established baseline results for query performance with row store tables (that is, without using Database In-Memory). A 9 dimensional, 500 million row star schema supported 25 users with a median query time of 16.9 seconds. (With a median of 16.9 seconds, there was not much point in increasing the number of users with the row store tables). The same data in a 500 million row de-normalized table supported 25 users with a median query time of 33.5 seconds per query.

Remember all of these tests are being conducted on a 60 core Intel server with commodity disk and 1 TB of DRAM.

Row Store Query

Median query performance (in seconds) of the query workload with 25 users on row store tables.

[Read More]

Friday Jul 10, 2015

Star Schema Challenge - Part 2.1

In Star Schema Challenge – Part 2 I revealed baseline results for my query workload running on a 500 million row star schema with all tables marked NO INMEMORY and In-Memory Aggregation prevented using the NO_VECTOR_TRANSFORM hint. With a median of 16.9 seconds per query (allowing for an average think time between queries of 5 seconds), I decided that 25 users were enough for a baseline.

After my first post in this series someone suggested that the best implementation might be a de-normalized table. I decided to satisfy this user’s curiosity and test a de-normalized table also, starting with NO INMEMORY test. (Vector Transform is not applicable to the de-normalized table because it does not join to another table.)

[Read More]

Friday Jul 03, 2015

Oracle Database In-Memory Bundle Patch 9 Released

The latest Bundle Patch for Database In-Memory has been released. The Bundle Patch is 21053000 or Bundle Patch 9 for Engineered Systems and DB In-Memory (June2015)). More information on the latest Bundle Patch can be found in the MOS note: Bundle Patches for Engineered Systems and DB In-Memory (Doc ID 1937782.1), and for for specific details on Bundle Patch 9 see MOS note 21053000.8.

[Read More]

Friday Jun 26, 2015

Do I really have to drop all of my reporting indexes?

I'm back on the road this month, meeting with customers to discuss their initial impressions and experiences with Oracle Database In-Memory. During one such discussion, I got asked a very peculiar question. The question was, "Do I really have to drop all of my reporting indexes if I use Database In-Memory?"

I have to admit I was a little taken aback by this question. After all, I thought most folks would be delighted to have an opportunity to give up the majority of their indexes, not just because of the space savings and DML performance benefits but also the maintenance nightmare that indexes can sometimes become.

Assuming this was a trick question, I deployed the standard stalling technique of answering a question with a question, “Can you tell me a little more about your situation?”

To which the system architect explained that they were in production with Oracle Database In-Memory on a 2 node RAC cluster running on commodity servers and a crap IO subsystem (his words, not mine). They had a snowflake schema, and had enough memory to accommodate all of their dimension tables but only the last 3 months of data in their two fact tables. Following my guidelines, they had kept their primary key indexes but dropped the rest of their indexes. He assured me that the performance of most of their queries had improved 100X and their ETL jobs were finishing 2X faster without the indexes but there were some queries that accessed more than just the last 3 months worth of data in the fact table and their performance had gotten worse, a lot worse.

It was in that moment that I realized that our guidance on dropping all reporting indexes with Database In-Memory had been based on an assumption that was not always true. The assumption I had been working under was; all of your performance critical data resides in memory or you have a good IO sub-system (engineered system etc.)

[Read More]

Friday Jun 19, 2015

Star Schema Performance Challenge - Part 2

In my previous post, Star Schema Performance Challenge – Part 1, I outlined a challenge to support 100 concurrent users querying a 9 dimensional, 500 million row star schema using a Sun X4-4 (with 60 cores, 1 TB DRAM and commodity storage).  The users, of course, expect great query performance.  The challenging part of this is 100 active users on a 60 core machine.  With far fewer or less active users this might not be so challenging, but 100 users on this size machine is a different story.

In this post I’ll talk more about the workload and share some baseline results.

[Read More]

The Oracle Database In-Memory blog is written by the Oracle product management team and sheds light on all things In-Memory.


« October 2015