SiSense Touts 3G Discovery and Analysis
SiSense is taking its pitch directly to business users -- much like insurgent BI specialists QlikTech and Tableau did half a decade (or more) ago.
- By Stephen Swoyer
- November 27, 2012
SiSense doesn't propose to take the data warehousing world by storm. No, it plans to parachute in.
It describes its flagship product, Prism, as a "third-generation" business intelligence (BI) platform designed to scale from an individual user's laptop up to and across the enterprise. Even though SiSense positions itself as an "enterprise-grade" BI platform, it's making its pitch directly to business users -- much like insurgent BI specialists such as QlikTech Inc. and Tableau Software Inc. first did half a decade (or more) ago.
"Instead of taking this to the big enterprise, [SiSense's strategy is] let's give users the opportunity to use this technology. Small departments can take down a huge amount of data and crunch it via [Prism's] elasticubes [technology]. They can't do that with Tableau or QlikTech," argues Eldad Farkash, SiSense founder and CTO
We'll come to elasticubes in a moment. Prism itself bundles a hardware-optimized columnar engine, a drag-and-drop data integration studio (Elasticube Manager), and a development environment aimed at technical and non-technical users alike (BI Studio).
Prism's column-store engine exploits CPU-based vector processing and SIMD capabilities; like other hardware-optimized columnar engines -- a category that includes Vectorwise (marketed in the U.S. by Actian Corp.) and SAP AG's HANA, to name just two. It's able to break down queries and to "parallelize" them into hundreds or thousands of CPU-specific instructions. "Every one of our queries is ... structured into thousands of instructions, and all of those are CPU-optimized," Farkash says.
Prism's BI Studio component is designed for either traditional DM practitioners or for a class of users that Farkash describes as "data heroes:" IT analysts, data-savvy business users, and other high-powered users.
Then there's Elasticube Manager. The basic unit of analysis in the SiSense world is the "elasticube." Unlike conventional OLAP cubes, elasticubes don't have to be pre-calculated or pre-aggregated; Farkash claims that they can likewise support an unlimited number of physical dimensions. Based on SiSense's description, elasticubes sound like "virtual cubes," such as those supported by vendors including Teradata Inc. and Kognitio.
One area of difference is that SiSense uses a "just-in-time" (JIT) in-memory implementation, which -- by virtue of its ability to dynamically load data into and out of memory -- gives an "elasticube" a good bit of its elasticity.
In terms of the kind of problem it's trying to solve, the elasticube is also similar to the ROLAP model, which MicroStrategy Inc. helped make famous. The idea is that the work associated with preparing data for access, to say nothing of the actual building of cubes, can be eliminated -- or minimized to some extent. Cubes don't have to be pre-computed or pre-aggregated, but can instead be generated on-the-fly.
In the Prism model, a data hero uses the Elasticube Manager to build access to data sources. Prism itself has OLEDB and ODBC connectivity, along with native connectors for SQL Server, Oracle, MySQL, and Hadoop. SiSense also offers a connectors for SMB- or Web 2.0-oriented offerings, including QuickBooks, QuickBase, Access, Google Analytics, Salesforce, ZenDesk, and other sources.
It isn't clear just what kinds of out-of-the-box data preparation tasks -- e.g., transformations, profiling, cleansing, or data quality routines -- Elasticube Manager can actually perform: on the crowded Hadoop + Strata show floor, there wasn't time enough, room enough, or audibility enough to learn from a deep-dive demo.
In any case, once a data hero has provisioned elasticube views in the Elasticube Manager, he/she can build dashboard views (mash-ups of data sources) in BI Studio. (The latter tool is designed for both data heroes or for less-specialized user constituencies.) Like discovery-oriented software from QlikTech Inc., Prism emphasizes information sharing and collaboration: Farkash says users can share particular views -- right down to the drill path -- with colleagues, for example.
3G versus 2G Connectivity
SiSense, in a sense, seems like a mashup of Tableau, with its focus on end-user-oriented visualization and discovery, and QlikView, with its emphasis on workgroup-level collaboration, rapid application development, and visual discovery.
Farkash, however, disputes both of these characterizations. He contrasts Prism – as a "third-generation" BI platform – with both offerings.
"Tableau and QlikTech are second-generation BI. Their designs address some of the problems of first-generation [BI], but they're limited in the amount of data they can handle or the complexity of the data [they can manage]," he claims. "With Tableau, if you have more than one data source, it becomes an IT problem; with QlikTech, if you're looking at large data sets, it's RAM-based [i.e., in-memory], so that becomes a problem."
With its JIT in-memory implementation, SiSense claims to deliver in-memory-like performance without being constrained by RAM limitations. Given the market segment it's targeting -- business users or business workgroups -- this makes sense. SiSense's show-floor demo at Strata + Hadoop World consisted of a laptop (populated with 8 GB of RAM) crunching a 1 TB dataset. JIT and elasticubes make this possible, Farkash claims.
"We don't need to squeeze everything into memory. What we're saying is, if you're using a third-generation columnar database [such as Prism], you can scale [to larger data sets] than just what your memory gives you," he concludes.