Splice Machine revs up Derby for SQL-on-Hadoop system
Are you sitting down? What if you could replace Oracle for Hadoop. WHAT THEN?
Startup Splice Machine has revealed its plan to take on database incumbents by letting people run a relational database management system on top of Hadoop's resilient free file system.
The company believes that where other SQL-on-Hadoop companies, like DrawnToScale, have failed, it will succeed. With its launch this week of a free and a paid-for version of its tech it hopes to convince people that by pairing the Hadoop File System (HDFS) with the Apache Derby Java-based database and the HBase Hadoop database and some other tech it can show a convincing speed-up over typical databases while costing less than typical providers.
But, given the large sums of money that have been poured into traditional paid-for databases like DB2, Oracle's suite, and SQL Server, along with powerful open source systems like MySQL and PostgreSQL, we find ourselves wondering what Hadoop can add.
"With the massive growth in data across all companies, traditional RDBMSs like MySQL and Oracle are either hitting the wall and having to prune data or becoming prohibitive to scale up on servers that can cost millions of dollar each," explained Splice Machine's chief executive Monte Zweben in an email with El Reg.
"Companies need to replace these struggling databases with a scale-out RDBMS. Scale-out based on Hadoop technology is very attractive because it has been proven to scale out to petabytes, has an extremely active developer community, has become the de facto standard for a distributed file system, and has an extensive ecosystem of analytical and data management tools."
The company is running Apache Derby on top, which is a "mature and stable" relational database management system with ANSI-99 SQL support. The tech was first developed by a company named Cloudscape which was bought by Informix in 1999, which was then bought by IBM in 2001, and IBM released Derby as open source in 2004.
"We do not see it as a fringe Apache project at risk of losing community support," Zweben explains. "In any case, Splice Machine's dependence on the Apache Derby community is minimal. As mentioned above, we have significantly modified the planner, optimizer, and excecutor to take advantage of the distributed infrastructure of HBase."
The Splice Machine system has a bevy of desirable characteristics, such as ACID transactions. It is designed to deal with both Online Analysis Processing (OLAP) and Online Transaction Processing (OLTP) tasks.
As for performance, Zweben says "on simple reads and writes, we provide performance in the low 10s of milliseconds, which is more than fast enough to power real-time applications and websites. We have seen over 10x price/performance improvement over existing RDBMSs such Oracle and MySQL."
"We can't reveal the name of the database, but we recently increased query performance against a clustered 20TB RDBMS (e.g., Oracle RAC, IBM DB2) by 3x-7x. For instance, one query that used a where clause of insert over 3M rows took 180 sec on the RDBMS, but it took only 26 sec or 7x faster on Splice Machine. These results were achieved on hardware that cost 75% less than the RDBMS cluster," he said
The free version of Splice Machine gives companies all the tech features but no support other than access to forums and upgrades. To try it out, companies must be less than five years old and have less than $10m in revenues.
The paid version, the "Enterprise Edition" comes with broad, round-the-clock support options and a guarantee to respond to problems in as little as one hour from when they are reported. It costs $5,000 per node, per year. ®
Sponsored: Benefits from the lessons learned in HPC