Calpont’s InfiniDB – Another ADBMS Insurgent Arises
November 8, 2010 Leave a comment
Calpont, rapidly emerging as yet another contender in the ADBMS sweepstakes, has announced version 2.0 of InfiniDB, its columnar MPP offering over shared storage. The value proposition hits now-familiar themes: high-performance query, fast data loading, data compression, and parallelized user defined functions (UDFs), all of which are becoming key checkoff capabilities. InfiniDB also hits hard on pricing, which it says dramatically undercuts that of its competitors. And a 30-day free trial of the enterprise edition sweetens the offer. For those comfortable with open source, the 2.0 release of the community edition is available as well. Calpont says the community edition (which is limited to a single server but is otherwise database feature-complete) has had 15,000 downloads. But the company’s relationship with Oracle for its MySQL components must be considered a risk going forward.
InfiniDB, like Infobright, is built atop Oracle’s MySQL. (I posted about Infobright last year, and it also has made significant progress, drawing favorable comment in the open source community for its continuing maturation.) Calpont’s relationship with Oracle must be seen as a risk factor..Oracle’s recent decisions about support raise questions about its interest in supporting anyone who is not an enterprise-class user of the Oracle-branded MySQL offering. Calpont has a deal through 2012 that includes an OEM license to integrate and use MySQL as the InfiniDB branded solution, and access to the MySQL channel. What will happen beyond that is clearly a concern.
InfiniDB Release 2.0 answers some prior concerns about the maturity of the product. My friend and BBBT colleague Jos van Dongen posted a look at InfiniDB earlier this year after looking at release 1.5 , in which he noted some challenges, including some missing SQL capabilities. MySQL consultancy Percona also raised questions in its MySQL Performance Blog about the absence of some data types, including Year, Time, Tinytext, and NOT NULL. These have evidently been addressed in version 2.0, although I have not confirmed the specifics point by point.
Version 2.0 also adds automatic data partitioning, another familiar theme. InfiniDB partitions data in both column and row dimensions. Dropping unneeded partitions can be automated, a nice feature for data whose horizontal partitioning is based on date. InfiniDB continues to operate over shared storage, but Calpont says shared-nothing is on the roadmap for 2011. Data compression has been enhanced, and Calpont says data can be decompressed “while being read from disk.” I didn’t dig into the deep discussion of this that will clearly be required to separate the players in the year ahead. Approaches vary, and a great deal of innovation is happening in the space.
InfiniDB executes what it calls map reduction (NOT an Apache MapReduce implementation, but a “similar” architectural approach.) The InfiniDB MPP architecture employs User modules and Performance modules. User modules host MySQL instances, keep an extent map of storage, enable failover, and manage distribution and final completion of work. Performance Modules, when operating as Workers, execute the distributed SQL and manage the distributed caches, shipping results to the controlling User module for aggregation and final processing. InfiniDB says filters, expressions, inner joins, outer joins, multi-table joins, correlated and non-correlated subquery, group by, and aggregation behaviors operate as fully distributed “reduction operations” within its framework.
Performance Modules also also serve when needed as Data modules to handle parallelized bulk data loading and DBMS management activities provided in DDL. All these may occur in a single server, where InfiniDB’s Community Edition will make use of all available CPUs and cores, or in a scaled out multi-server deployment, which requires the Enterprise Edition. Scale-out can be performed in real time without taking the system down – simply register the new module online and it will be used.
Version 2.0 participates in another ADBMS theme: UDFs. These are fully parallel and distributed, running as “an integrated operation within the InfiniDB storage engine.” InfiniDB will use all available cores within the distributed layer. Only UDFs written in C++ are supported at this point.
Given its relatively recent arrival, it’s not surprising that InfiniDB only boasts a couple of named customers. The ASI Group, which tells of a POC win based on load time deployment speed (also cited by named customers CaringBridge and OneView), is in the airline industry, an unusual vertical so far for ADBMSs. SamKnows, a broadband performance testing provider, touts the price performance and also cites the ease and speed of deployment. Cognitive Match treads more familiar territory: real-time behavioral analytics.
Price is a key issue. Calpont asserts that its prices are dramatically lower than its competitors, and that its 3-year TCO reflects that. It prefers pricing by hardware configuration as shown in its table shown here at right (click to enlarge); it says that per-TB pricing effectively penalizes the customer for using the product. The table shows Calpont’s read on the pricing of its competitors. Take it with a grain of salt – all list pricing from any vendor is subject to discount, of course. If previous experience is any guide, some of the mentioned vendors will offer some thoughts – corrections or otherwise – on this information in comments to this post; come back and check if you’re looking at this early. But whether you accept the table as representative of the costs you would see or not, it’s fair to say that Calpont will be significantly competitive on price.
Disclosures: Neither Calpont, Infobright nor Microsoft are 2010 clients of IT Market Strategy. Oracle is, as are IBM, Teradata, SAP, Sybase, and Greenplum.