Analysis of data warehousing giant Teradata. Related subjects include:
Ever more products try to integrate SQL with Hadoop, and discussions of them seem confused, in line with Monash’s First Law of Commercial Semantics. So let’s draw some distinctions, starting with (and these overlap):
- Are the SQL engine and Hadoop:
- Necessarily on the same cluster?
- Necessarily or at least most naturally on different clusters?
- How, if at all, is Hadoop invoked by the SQL engine? Specifically, what is the role of:
- HDFS (Hadoop Distributed File System)?
- Hadoop MapReduce?
- How, if at all, is the SQL engine invoked by Hadoop?
- If something is called a “connector”, then Hadoop and the SQL engine are most likely on separate clusters. Good features include (but these can partially contradict each other):
- A way of making data transfer maximally parallel.
- Query planning that is smart about when to process on the SQL engine and when to use Hadoop’s native SQL (Hive or otherwise).
- If something is called “SQL-on-Hadoop”, then Hadoop and the SQL engine are or should be on the same cluster, using the same nodes to store and process data. But while that’s a necessary condition, I’d prefer that it not be sufficient.
Let’s go to some examples. Read more
|Categories: Cloudera, Data integration and middleware, EAI, EII, ETL, ELT, ETLT, Hadapt, Hadoop, HBase, Hortonworks, MapReduce, Microsoft and SQL*Server, NewSQL, PostgreSQL, SQL/Hadoop integration, Teradata||36 Comments|
I talked tonight with Lee Edlefsen, Chief Scientist of Revolution Analytics, and now think I understand Revolution’s parallel R much better than I did before.
There are four primary ways that people try to parallelize predictive modeling:
- They can run the same algorithm on different parts of a dataset on different nodes, then return all the results, and claim they’ve parallelized. This is trivial and not really a solution. It is also the last-ditch fallback position for those who parallelize more seriously.
- They can generate intermediate results from different parts of a dataset on different nodes, then generate and return a single final result. This is what Revolution does.
- They can parallelize the linear algebra that underlies so many algorithms. Netezza and Greenplum tried this, but I don’t think it worked out very well in either case. Lee cited a saying in statistical computing “If you’re using matrices, you’re doing it wrong”; he thinks shortcuts and workarounds are almost always the better way to go.
- They can jack up the speed of inter-node communication, perhaps via MPI (Messaging Passing Interface), so that full parallelization isn’t needed. That’s SAS’ main approach.
One confusing aspect of this discussion is that it could reference several heavily-overlapping but not identical categories of algorithms, including:
- External memory algorithms, which operates on datasets too big to fit in main memory, by — for starters — reading in and working on a part of the data at a time. Lee observes that these are almost always parallelizable.
- What Revolution markets as External Memory Algorithms, which are those external memory algorithms it has gotten around to implementing so far. These are all parallelized. They are also all in the category of …
- … algorithms that can be parallelized by:
- Operating on data in parts.
- Getting intermediate results.
- Combining them in some way for a final result.
- Algorithms of the previous category, where the way of combining them specifically is in the form of summation, such as those discussed in the famous paper Map-Reduce for Machine Learning on Multicore. Not all of Revolution’s current parallel algorithms fall into this group.
To be clear, all Revolution’s parallel algorithms are in Category #2 by definition and Category #3 in practice. However, they aren’t all in Category #4.
|Categories: Greenplum, Hadoop, MapReduce, Netezza, Parallelization, Predictive modeling and advanced analytics, Revolution Analytics, Teradata||Leave a Comment|
Relational DBMS used to be fairly straightforward product suites, which boiled down to:
- A big SQL interpreter.
- A bunch of administrative and operational tools.
- Some very optional add-ons, often including an application development tool.
Now, however, most RDBMS are sold as part of something bigger.
- Oracle has hugely thickened its stack, as part of an Innovator’s Solution strategy — hardware, middleware, applications, business intelligence, and more.
- IBM has moved aggressively to a bundled “appliance” strategy. Even before that, IBM DB2 long sold much better to committed IBM accounts than as a software-only offering.
- Microsoft SQL Server is part of a stack, starting with the Windows operating system.
- Sybase was an exception to this rule, with thin(ner) stacks for both Adaptive Server Enterprise and Sybase IQ. But Sybase is now owned by SAP, and increasingly integrated as a business with …
- … SAP HANA, which is closely associated with SAP’s applications.
- Teradata has always been a hardware/software vendor. The most successful of its analytic DBMS rivals, in some order, are:
- Netezza, a pure appliance vendor, now part of IBM.
- Greenplum, an appliance-mainly vendor for most (not all) of its existence, and in particular now as a part of EMC Pivotal.
- Vertica, more of a software-only vendor than the others, but now owned by and increasingly mainstreamed into hardware vendor HP.
- MySQL’s glory years were as part of the “LAMP” stack.
- Various thin-stack RDBMS that once were or could have been important market players … aren’t. Examples include Progress OpenEdge, IBM Informix, and the various strays adopted by Actian.
I coined the term schema-on-need last month. More precisely, I coined it while being briefed on JSON-in-Teradata, which was announced earlier this week, and is slated for availability in the first half of 2014.
The basic JSON-in-Teradata story is as you expect:
- A JSON document is stuck into a relational field.
(Oddly, Teradata wasn’t yet sure whether the field would be a BLOB or VARCHAR or something else.)Edit: See Dan Graham’s comment below.
- Fields within the JSON document can be indexed on.
- Those fields can be referenced in SQL statements much as regular Teradata columns can.
You have to retrieve the whole document.Edit: See Dan Graham’s comment below.
- To avert the performance pain of retrieving the whole document, you can of course copy any particular field into a column of its own. (That’s the schema-on-need part of the story.)
JSON virtual columns are referenced a little differently than ordinary physical columns are. Thus, if you materialize a virtual column, you have to change your SQL. If you’re doing business intelligence through a semantic layer, or otherwise have some kind of declarative translation, that’s probably not a big drawback. If you’re coding analytic procedures directly, it still may not be a big drawback — hopefully you won’t reference the virtual column too many times in code before you decide to materialize it instead.
My Bobby McFerrin* imitation notwithstanding, Hadapt illustrates a schema-on-need approach that is slicker than Teradata’s in two ways. First, Hadapt has full SQL transparency between virtual and physical columns. Second, Hadapt handles not just JSON, but anything represented by key-value pairs. Still, like XML before it but more concisely, JSON is a pretty versatile data interchange format. So JSON-in-Teradata would seem to be useful as it stands.
*The singer in the classic 1988 music video Don’t Worry Be Happy. The other two performers, of course, were Elton John and Robin Williams.
|Categories: Data models and architecture, Data warehousing, Hadapt, Schema on need, Structured documents, Teradata||3 Comments|
Much of modern analytic technology deals with what might be called an entity-centric sequence of events. For example:
- You receive and open various emails.
- You click on and look at various web sites and pages.
- Specific elements are displayed on those pages.
- You study various products, and even buy some.
Analytic questions are asked along the lines “Which sequences of events are most productive in terms of leading to the events we really desire?”, such as product sales. Another major area is sessionization, along with data preparation tasks that boil down to arranging data into meaningful event sequences in the first place.
A number of my clients are focused on such scenarios, including WibiData, Teradata Aster (e.g. via nPath), Platfora (in the imminent Platfora 3), and others. And so I get involved in naming exercises. The term entity-centric came along a while ago, because “user-centric” is too limiting. (E.g., the data may not be about a person, but rather specifically about the actions taken on her mobile device.) Now I’m adding the term event series to cover the whole scenario, rather than the “event sequence(s)” I might appear to have been hinting at above.
I decided on “event series” earlier this week, after noting that: Read more
|Categories: Aster Data, Business intelligence, Data warehousing, EAI, EII, ETL, ELT, ETLT, Platfora, Predictive modeling and advanced analytics, Teradata, Vertica Systems, Web analytics, WibiData||19 Comments|
Teradata Aster 6 has been preannounced (beta in Q4, general release in Q1 2014). The general architectural idea is:
- There are multiple data stores, the first two of which are:
- The classic Aster relational data store.
- A file system that emulates HDFS (Hadoop Distributed File System).
- There are multiple processing “engines”, where an engine is what occupies and controls a processing thread. These start with:
- Generic analytic SQL, as Aster has had all along.
- SQL-MR, the MapReduce Aster has also had all along.
- SQL-Graph aka SQL-GR, a graph analytics system.
- The Aster parser and optimizer accept glorified SQL, and work across all the engines combined.
There’s much more, of course, but those are the essential pieces.
Just to be clear: Teradata Aster 6, aka the Teradata Aster Discovery Platform, includes HDFS compatibility, native MapReduce and ways of invoking Hadoop MapReduce on non-Aster nodes or clusters — but even so, you can’t run Hadoop MapReduce within Aster over Aster’s version of HDFS.
The most dramatic immediate additions are in the graph analytics area.* The new SQL-Graph is supported by something called BSP (Bulk Synchronous Parallel). I’ll start by observing (and some of this is confusing):
- BSP was thought of a long time ago, as a general-purpose computing model, but recently has come to the fore specifically for graph analytics. (Think Pregel and Giraph, along with Teradata Aster.)
- BSP has a kind of execution-graph metaphor, which is different from the graph data it helps analyze.
- BSP is described as being a combination hardware/software technology, but Teradata Aster and everybody else I know of implements it in software only.
- Aster long ago talked of adding a graph data store, but has given up that plan; rather, it wants you to do graph analytics on data stored in tables (or accessed through views) in the usual way.
Use cases suggested are a lot of marketing, plus anti-fraud.
*Pay no attention to Aster’s previous claims to do a good job on graph — and not only via nPath — in SQL-MR.
So far as I can infer from examples I’ve seen, the semantics of Teradata Aster SQL-Graph start:
- Ordinary SQL except in the FROM clause.
- Functions/operators that are the arguments for FROM; of course, they output tables. You can write these yourself, or use Teradata Aster’s prebuilt ones.
Within those functions, the core idea is: Read more
|Categories: Application areas, Aster Data, Business intelligence, Data models and architecture, Data warehousing, Hadoop, Parallelization, Predictive modeling and advanced analytics, RDF and graphs, Teradata||4 Comments|
I recently wrote (emphasis added):
My clients at Teradata Aster probably see things differently, but I don’t think their library of pre-built analytic packages has been a big success. The same goes for other analytic platform vendors who have done similar (generally lesser) things. I believe that this is because such limited libraries don’t do enough of what users want.
The bolded part has been, shall we say, confirmed. As Randy Lea tells it, Teradata Aster sales qualification includes the determination that at least one SQL-MR operator — be relevant to the use case. (“Operator” seems to be the word now, rather than “function”.) Randy agreed that some users prefer hand-coding, but believes a large majority would like to push work to data analysts/business analysts who might have strong SQL skills, but be less adept at general mathematical programming.
This phrasing will all be less accurate after the release of Aster 6, which extends Aster’s capabilities beyond the trinity of SQL, the SQL-MR library, and Aster-supported hand-coding.
Randy also said:
- A typical Teradata Aster production customer uses 8-12 of the prebuilt functions (but now they seem to be called operators).
- nPath is used in almost every Aster account. (And by now nPath has morphed into a family of about 5 different things.)
- The Aster collaborative filtering operator is used in almost every account.
- Ditto a/the text operator.
- Several business intelligence vendors are partnering for direct access to selected Teradata Aster operators — mentioned were Tableau, TIBCO Spotfire, and Alteryx.
- I don’t know whether this is on the strength of a specific operator or not, but Aster is used to help with predictive parts failure applications in multiple industries.
And Randy seemed to agree when I put words in his mouth to the effect that the prebuilt operators save users months of development time.
Meanwhile, Teradata Aster has started a whole new library for relationship analytics.
|Categories: Application areas, Aster Data, Data warehousing, Predictive modeling and advanced analytics, Teradata, Text||1 Comment|
In a general pontification on positioning, I wrote:
every product in a category is positioned along the same set of attributes,
and went on to suggest that summary attributes were more important than picky detailed ones. So how does that play out for investigative analytics?
First, summary attributes that matter for almost any kind of enterprise software include:
- Performance and scalability. I write about analytic performance and scalability a lot. Usually that’s in the context of analytic DBMS, but it also arises in analytic stacks such as Platfora, Metamarkets or even QlikView, and also in the challenges of making predictive modeling scale.
- Reliability, availability and security.* This is more crucial for short-request applications than analytic ones, but even your analytic systems shouldn’t leak data or crash.
- Goodness of fit with legacy systems. I hate that one, because enterprises often sacrifice way too much in favor of that benefit.
- Price. Duh.
*I picked up that phrase when — abbreviated as RAS — it was used to characterize the emphasis for Oracle 8. I like it better than a general and ambiguous concept of “enterprise-ready”.
The reason I’m writing this post, however, is to call out two summary attributes of special importance in investigative analytics — which regrettably which often conflict with each other — namely:
- Agility. People don’t want to submit requests for reports or statistical analyses; they want to get answers as soon as the questions come to mind.
- Completeness of feature set — for a particular use case, that is. There’s no such thing as an investigative analytics offering with a feature set that’s close to complete for all purposes; even SAS, IBM and other behemoths fall short.
Much of what I work on boils down to those two subjects. For example: Read more
|Categories: Aster Data, Business intelligence, Data warehousing, KXEN, Predictive modeling and advanced analytics, SAS Institute, Teradata||11 Comments|
I talked with Teradata about a bunch of stuff yesterday, including this week’s announcements in in-database predictive modeling. The specific news was about partnerships with Fuzzy Logix and Revolution Analytics. But what I found more interesting was the surrounding discussion. In a nutshell:
- Teradata is finally seeing substantial interest in in-database modeling, rather than just in-database scoring (which has been important for years) and in-database data preparation (which is a lot like ELT — Extract/Load/transform).
- Teradata is seeing substantial interest in R.
- It seems as if similar groups of customers are interested in both parts of that, such as:
This is the strongest statement of perceived demand for in-database modeling I’ve heard. (Compare Point #3 of my July predictive modeling post.) And fits with what I’ve been hearing about R.
|Categories: EAI, EII, ETL, ELT, ETLT, Parallelization, Predictive modeling and advanced analytics, Revolution Analytics, SAS Institute, Telecommunications, Teradata||1 Comment|
When we scheduled a call to talk about Sentry, Cloudera’s Charles Zedlewski and I found time to discuss other stuff as well. One interesting part of our discussion was around the processing “frameworks” Cloudera sees as most important.
- The four biggies are:
- MapReduce. Duh.
- SQL, specifically Impala. This is as opposed to the uneasy Hive/MapReduce layering.
- “Math” , which seems to mainly be through partnerships with SAS and Revolution Analytics. I don’t know a lot about how these work, but I presume they bypass MapReduce, in which case I could imagine them greatly outperforming Mahout.
- Stream processing (Storm) is next in line.
- Graph — e.g. Giraph — rises to at least the proof-of-concept level. Again, the hope would be that this well outperforms graph-on-MapReduce.
- Charles is also seeing at least POC interest in Spark.
- But MPI (Message Passing Interface) on Hadoop isn’t going anywhere fast, except to the extent it’s baked into SAS or other “math” frameworks. Generic MPI use cases evidently turn out to be a bad fit for Hadoop, due to factors such as:
- Low data volumes.
- Latencies in various parts of the system
HBase was artificially omitted from this “frameworks” discussion because Cloudera sees it as a little bit more of a “storage” system than a processing one.
Another good subject was offloading work to Hadoop, in a couple different senses of “offload”: Read more