This week, Teradata introduced its in-memory capabilities.  The press release is here.   

In-memory is the current fastest commercial medium for data storage, upwards of thousands of times faster than HDD.  Naturally, it is expensive relatively speaking or it would be more pervasive now.  However, the price point has lowered recently and we are having an in-memory hardware renaissance.   As Tony Baer of Ovum said in a tweet response to me:  "w/DRAM cheap,takes SmartCaching to nextlevel."

Teradata takes in-memory a step further by considering the relative priority of data and, though all data is backed by disk, it will automatically transfer it in and out of memory as appropriate, in a feature it calls Intelligent Memory.  This is the data "80/20 rule" at work.   After 7 days of usage, Teradata says the allocation will be fairly settled.  Teradata has found that, across multiple environments, 85% of I/O uses 10% of the cylinders.  It's in-memory driven by data temperature (and this extends to HDD and SSD). 

You still need to allocate the memory (and SSD) initially but whatever you allocate will be intelligently utilized.  It will run will run across the Teradata family: the 670, 1650, 2700 and 6700 models. 

It behooves us to seek out the fastest medium for data that can be utilized effectively in the organization and delivers maximum ROI.  Those who understand the possibilities for data today should be sharing and stimulating the possibilities for information in the organization.  All productive efforts at information management go to the bottom line.

Couchbase announced the general availability of Couchbase Server 2.0 today.  This is the long anticipated release that carries Couchbase's capabilities over the threshold from a key-value store to document store. 

Key-value stores have little to no understanding of the value part of the key in the key-value. It is simply a blob, keyed by a unique identifier which serves a "primary key" function.  This key is used for get, put and delete.  You can also search inside the value, although the performance may be suboptimal.  The application consuming the record will need to know what to do with the blob/value.

Key-value stores are prominent today but with a few additional features they can become a document store or a column store and Couchbase has made their server a document store. When records are simple and lack strong commonality of look, a key-value store may be all you need. However, I don't see any of Couchbase's current workloads moving away from Couchbase due to the innovations.  These workloads include keeping track of everything related to the new class of web, console and mobile applications that are run simultaneously by thousands of users, sometimes sharing sessions - as in games.  They also include shopping carts.

Workloads involving relationships amongst data, complexity of the data or the need to operate on multiple keys/records at once need some of the recently added functions to 2.0 which include distributed indexing and querying, JSON storage, online compaction, incremental map reduce and cross data center replication.

Document stores tend to be excellent for logging online events of different types when those events can have varying characteristics.  They work well as pseudo content management systems that support website content.  Due to their ability to manipulate at a granular level, document stores are good for real-time analytics that change dynamically for which the application can take advantage of immediately.  

Key-value stores have historically (ok, it's all a short history) performed better than the more functional document stores.  Couchbase is navigating their capabilities forward in a very sensible manner.  There is no reason for performance degradation or lack of functionality and Couchbase is beginning to show this.  While there's still a release or two to go to be at a functionality standard for document stores, Couchbase is coming at it from a strong performance base.

The story goes that Willy Sutton robbed banks because "that's where the money is."  While this attribution appears to be an urban legend, it's no myth that Oracle has a lion's share of databases - both transactional and analytic.

IBM started an advanced land grab for Oracle customer conversions by bringing a high compatibility of PL/SQL into the DB2 database.

Now, Teradata has invested resources in facilitating the migration away from Oracle.  With the Teradata Migration Accelerator (TMA). structure and SQL (PL/SQL) code can be converted to Teradata structures and code. This is a different philosophy from IBM, which requires few code changes for the move, but also doesn't immediately optimize that code for DB2.

While data definition language (DDL) has only minor changes from DBMS to DBMS, such as putting quotes around keywords, Teradata's key activity and opportunity in the migration is to change Oracle cursors to Teradata set-based SQL. 

"Rule sets" - for how to do conversions - can be applied selectivity across the structure and code in the migration.  TMA supports selective data movement, if desired, with WHERE clauses for the data.  TMA also supports multiple users doing a coordinated migration effort.

TMA also works for DB2 migrations.

While it will not do the trick on its own, having these tools, which convinces a shop that the move could be more pain-free than originally thought, will support DBMS migrations.  

Teradata Aster demonstrates its graphical "pathing" capabilities very nicely by showing the relationships between tweeters and their tweets at events, like the Teradata Third-Party Influencers Event I attended last week. 

The demonstration shows how to produce some sentiment of the event, but more importantly demonstrates relationships and influence power.  Customer relationships and influence power are becoming part of the set of derived data needed to fully understand a company's customers.  This leads to identifying engagement models and the early identification of patterns of activity that lead to certain events - desired or otherwise.

One important point noted by Stephanie McReynolds, Director of Product Marketing, at Teradata Aster, was that the sphere of relevant influence depends on the situation.  You can retweet hundreds of tweets, many for which you do not even know the tweeter.  However, when buying a car, those who would influence you would be only a handful.

One would need to take some more heed of an influencer's opinion - or that of someone with a relationships to the influencer.  It can become quite a layered analysis and influence power is hard to measure.  Grabbing various digital breadcrumbs is relatively easy, but is it indicative of influence?  Likewise, is a tweetstream indicative of the sentiment of an event?  I'm not sure.  It may not even be indicative of the sentiment of the tweeters.  Digital is all a start.  The worlds of third-party data, real sentiment analysis and possibly sensor data are coming together.