Introduction

Relational databases (RDBMS) like SQL Server, Oracle, MySQL and even Access all store tabular data row-by-row.  This structure is best for transactional/operational systems that require large numbers of concurrent insertions.  With indexes, it can also provide realistic query response times for row-based queries that do not frequently require aggregations or joining of many tables.

Data analysis often requires aggregation of data as well as merging of data located in multiple disparate tables.  When dealing with these types of queries, relational databases reach their limits pretty quickly. The only way to extend these limits is by putting in stronger hardware and pre-aggregating data to reduce the amounts of calculations that occur in real time.

The ElastiCube Columnar Database

ElastiCube data is held in a Columnar Database Management System (CDBMS) that stores data field-by-field.  Each field is individually stored in a memory-mapped file, the same mechanism the Windows operating system pagefile uses for memory dumping and loading.

When a query is executed over an ElastiCube, only fields referenced in the query need to be loaded into memory.  This leaves enough space for actually processing the query entirely in memory without any read/write to the hard-drive – the prime reason for poor performance of queries.  Once a field is no longer used, it is removed from memory and its consumed space is freed.

This approach has several advantages:

  • Query Response Time
    Queries over data sets containing millions of rows of data return in seconds even under modest hardware configurations such as desktop computers.
  • Materialization Time
    ElastiCubes do not require pre-aggregations and/or creation of indexes to assure fast query response, therefore the actual creation of an ElastiCube takes a fraction of the time of a data mart or an OLAP cube.
  • Storage Space
    Pre-aggregations and the creation of indexes are not needed to assure fast query response, making an ElastiCube’s size significantly smaller than a datamart or an OLAP cube.
  • High Compression
    This columnar storage strategy makes the data much more suitable for high levels of compression, without loss of detail or accuracy. This means less hardware is needed; less disk space and less RAM than for an equivalent-sized, traditional Business Intelligence DB.
  • 64-bit Support
    Written and designed to natively support 64-bit processing, it vastly increases the amount of memory the system can address at any given time. 64-bit architecture means you can work with virtually unlimited amounts of data.
  • True Multi-User, Multi-Application Architecture
    ElastiCubes are not tightly coupled with the application layer of the system. This frees up a single ElastiCube to handle multiple applications and users. Not having to reproduce your data model for every application saves significant time developing and maintaining your dashboards and reports.

Just-In-Time, In-Memory Processing

  • Smart Cache and Instruction Recycling
    CPU cycles and RAM space are the two most precious resources in any computer, and ElastiCube is designed to use both as efficiently and speedily as possible. Using our sophisticated caching algorithm, the data is only loaded into memory when it’s needed. As part of this algorithm, compute- and time-intense calculations are also intelligently cached to further reduce I/O calls.
  • Cache-aware Algorithm
    Additional sophisticated algorithms further increase Sisense’s performance. Once data is loaded into memory, the main performance bottleneck becomes CPU cache misses that naturally come with random access. The ElastiCube is specifically designed to minimize these errors by employing a unique cache-aware algorithm, further increasing Sisense’s performance by an additional order of magnitude.
  • Compressed Calculations
    Every DB compresses data to save disk space and RAM. ElastiCube is designed to work directly on this compressed data, so that the need for decompression is virtually eliminated, further increasing ElastiCube’s performance.

Designed with Standard Hardware in Mind

Just about every new computer on the market—even portables like iPhones and iPads—are built with very powerful multi-core processors, putting several CPUs into one. ElastiCube was built specifically to take advantage of these powerful CPUs, further increasing Sisense’s performance on standard hardware, enabling you to run multiple applications and support multiple users.

  • Highly RAM-efficient
    The thing we know for sure about DBs is that they grow. Fast. So no matter how much fancy footwork is done with completely in-memory DBs, eventually you run out of RAM space and need to upgrade—at least your RAM (best case) or your entire hardware platform (worst, very expensive, case). At Sisense we know this, so we spent years designing the ElastiCube to be able to handle terabytes—billions of rows—of data efficiently and quickly, even on standard PC hardware.

Unified Analytics Engine

Sisense can execute queries against a wide variety of data sources as if they were all of the same type, essentially making the individual characteristics of each physical data source unimportant. Our Unified Analytics Engine is what makes this possible.

When Sisense imports data, the Unified Analytics Engine creates a metadata layer, or abstraction layer, which is then used to formulate queries across any number of tables from any number of data sources in any number of formats. It even supports the combined querying of resident and external (live) database sources without first loading data into the database!

These capabilities provide the user with unparalleled flexibility and speed in creating, executing and sharing highly complex reports, dashboards, and analytic applications, with any number and variety of data sources.

Compliant with Industry Standards

  • Supports SQL-92 Standard
    Even with all this advanced technology, we knew that none of it would be any good if our users couldn’t access their existing data. So, we built in an SQL layer to the system, which allows users to integrate Sisense to external applications without needing to learn new scripting languages.
  • Seamless Integration with Existing Data Sources
    Got an ODBC/OleDB compliant DB today? Great, we built in the ability to access those, too. ElastiCube will seamlessly connect to those data sources so, again, there is no need to learn a new language or write special code to connect to your existing data. With ElastiCube there’s no need to start over, you just get faster, easier, and more scalable, with minimal need for IT.