Oracle Openworld 2013 Day 2: Exalytics, TimesTen and Essbase Futures

Monday's almost over at Oracle Openworld 2013 in San Francisco, and it started for me today with a presentation on Enterprise Manager 12c and the BI Management Pack, alongside Henrik Blixt (PM for the BI Management Pack) and Dhananjay Papde, author of a book on EM12c. I've covered EM12c and the BI Management Pack extensions quite a bit on the blog over the past few months so it was good to exchange a few ideas and observations with Henrik, and it was also good to meet Dhananjay, who's been working with EM for a long time and has particularly specialized in the configuration management, and SLA-monitoring parts of the app.

Similarly, I finished-up the day with another joint session this time on TimesTen for Exalytics, with Peak Indicators' Tony Heljula and Chris Jenkins, one of the TimesTen development PMs. As with all these sessions, it's the audience interaction that makes them interesting, and we had a number of excellent questions, particularly at the TimesTen one given the very interesting product announcements during the day - more on which in a moment.


Before I get onto those though, here's the links to today's RM presentation downloads, with presentations today given by myself, Jérôme Françoisse (with Gurcan Orhan) and Venkat:

So, onto the product roadmap sessions:

1) Oracle Exalytics In-Memory Machine

The first set of announcements was around Oracle Exalytics In-Memory Machine, which started off as a Sun x86_64 server with 1TB RAM and 40 CPU cores, then recently went to 2TB and SSD disks, and now is available in a new configuration called Oracle Exalytics T5-8. This new version comes with 4TB RAM and is based on Sun SPARC T5 processes with, in this configuration, a total of 128 CPU cores, and is aimed at the BI consolidation market - customers who want to consolidate several BI applications, or BI environments, onto a single server - priced in this case around the $350k mark excluding the software.

What's also interesting is that the T5-8 will use Sun Solaris on SPARC as the OS, giving it access to Solaris' virtualisation and resource isolation technologies again positioning it as a consolidation play rather than host for a single, huge, BI application. Given the price I can't quite see us getting one yet, but it's an obvious upgrade path from the X2-3 and X2-4 servers and something you'd want to seriously consider if you're looking at setting up a "private cloud"-type server infrastructure.

The Exalytics roadmap session also previewed other potential upcoming features for OBIEE, I would imagine earmarked for Exalytics given some of the computation work that'd need to go into the background to support them, including:

  • A "Google Search" /. "Siri"-type feature called BI Ask, that presented the user with a Google-style search box into which you could type phrases such as "revenue for May 2010 for Widget A", with the feature then dynamically throwing-up a table or graph based on what you've requested. Rather than attempting natural language parsing, BI Ask appears to work with a structured dictionary of words based on objects in the Presentation Services catalog, with choices available to the user (for example, lists of measures or dimensions) appearing under the search box in a similar manner to the "Google Suggest" feature.Although the demo was done using a desktop browser, where I think this could be particularly useful is in a mobile context, especially given most browsers' and mobile platforms' in-built ability to receive speed input and automatically pass that as text to the calling application. If you can imagine Siri for mobile analytics, with you holding your iPhone up and saying to it "revenue for southern region for past three months, compared to last year" and a graph of revenue over this period automatically appearing on your iPhone screen - that's what I think BI Ask is trying to get towards.
  • A user-driven data mashup feature that allowed the user to browse to a spreadsheet file on their local desktop, upload it to the OBIEE server (maybe to an Oracle Cloud database?), and then automatically join it to the main corporate dataset so that they could add their own data to that provided to the BI system. Clearly any setup like this needs to clearly differentiate between metrics and attributes uploaded by the user, compared to the "gold standard" ones provided as part of the RPD and Presentation Services Catalog, but this is potentially a very useful feature for users who'd otherwise export their OBIEE data to Excel, and then do the data combining there.
  • Probably more for "Exalytics v2", but a totally revamped aggregate refresh and reload framework, probably based around DAC technology, that would leverage the DAC's own data loading capabilities and tools such as GoldenGate to perform incremental refreshes of the Exalytics adaptive in-memory cache. No specific details yet but it's pretty obvious how this could improve over the current Exalytics v1 setup.
2) Oracle TimesTen for Exalytics

Yesterday of course had the big announcement about the new In-Memory Option for Oracle Database 12c, and this of course then led to the obvious question - what about TimesTen, which up until now was Oracle's in-memory database - and what about Exalytics, where TimesTen was the central in-memory part of the core proposition? And so - given that I was on the TimesTen Birds of a Feather Panel this evening and no doubt would need to field exactly those questions, I was obviously quite keen to get along to one of the TimesTen roadmap sessions earlier in the day to hear Oracle's story around this.

And - it actually does make sense. What's happening is this:

  • TimesTen's development team has now been brought under the same management as Oracle Database 12c's In-Memory option, with (over time) the same core libraries, and same performance features
  • TimeTen will get the ability to store its tables (which are already held in memory) in columnar format as well as the existing row format - the difference being that unlike the Oracle in-memory feature, this is not done through on-the-fly data replication - it's either stored row-store or column-store, something you decide when you create the table, and the only thing disk is used for is checkpointing and data persistence between reboots
  • TimesTen will also gain the ability to be set up as a grid of servers that provide a single database instance - a bit like RAC and it's single instance/cache fusion, and with support for replication so that you can copy data across the nodes to protect against machine failure. Currently you can link TimesTen servers together but each one is its own database instance, and you'd typically do this for high-availability and failover rather than creating one large database. What this grid setup also gives us though is the ability to do parallel query - Oracle didn't say whether this would be one slave per grid node, or whether it'd support more than one slave per node, but coupled with the in-memory column store feature, presumably this is going to mean bigger TimesTen databases and a lot faster queries (and it's fast already).
So what about the positioning of TimesTen vs. Oracle Database In-Memory Option - does one replace the other, or do you use the two together? Oracle's ideas on this were as follows:
  • Clearly the in-memory Oracle Database option is going to be a great query accelerator for large-scale data warehouses, but there's still a lot of value in having a mid-tier in-memory data layer that's under the control of the BI system owner, rather than the DBAs. You'll have control over the data model, you can implement it quicker than you'd be able to upgrade the whole data warehouse database, and its physically co-located closer to the BI Server, so you'll have less of an issue with network latency.
  • TimesTen's in-memory columnar storage technology will be based on a similar approach to that which is being taken by the database, and developed by the same overall team. But TimesTen most probably will have shorter development cycles, so new features might appear in TimesTen first, and it's also lower risk for customers to test out new in-memory approaches in TimesTen rather than trying to reconfigure the whole warehouse to try out a new approach
And I think this makes sense. Of course, until we actually get hold of the two products and test them out, and see how the pace of development works out over time, we're not going to fully know which product to deploy where - and of course pricing and packaging has yet to be announced; for example, I'd strongly predict that columnar storage for TimesTen will be an Exalytics-only feature, whilst the In-Memory Option for the database might be priced more like RAC than Partitioning, or even packaged up with Partitioning and OLAP as some sort of "data warehousing option". We'll just have to wait and see.

3) Oracle Essbase

The Essbase roadmap was the last session I managed to attend today, and again there were some pretty exciting new features announced or trailed (and it was made clear that the new features at the end of this list were more at planning or conceptual stage at the moment, and may well not make it into the product). Anyway, here's what was talked about in the session, for BI and Exalytics-type use cases:

  • Improved MDX query creation when working with the BI Server, including support for sub-selects - something that might help to reduce the number of separate MDX queries that OBIEE has to generate to work-out all the subtotals required for hierarchical column queries
  • Improvements to the MDX AGGREGATE function and a revamped cube spin-off feature for OBIEE, including a prototype new web-based MOLAP Acceleration Wizard for auto-generating Essbase cubes for OBIEE aggregate persistence
  • A new Cube Deployment Service private API, that's used by the MOLAP Aggregation Wizard (amongst others) to generate and deploy an Essbase cube within a cloud-type environment
  • A "renegade member" feature used for collecting in all the data load records for members that can't be located - aimed at avoiding the situation where totals in an Essbase cube don't then match the totals in the source system, because records got dropped during the data load
  • Very speculatively - a potential hybrid BSO/ASO storage mode, combining BSO's calculation capabilities with ASO's dynamic aggregation.
So - lots of potential new features and a peek into what could be in the roadmap for three key OBIEE and Exalytics technologies. More tomorrow as we get to attend roadmap sessions for OBIEE in the Cloud, and ODI 12c.