The Schwartz Cloud Report

Blog archive

Startup NuoDB Launches Relational Database for Cloud

One of the key inhibitors to using cloud infrastructures for applications that process thousands or even millions of transactions per second is the latency associated with those architectures. Storage, security and the network can all impact cloud performance. Another variable is the database, which is pivotal to modern systems that process real-time data that can range from the processing of financial transactions to feeds that populate social media networks.

Startup NuoDB, financed and operated by veterans of the database industry, hopes to solve the issue of processing and transacting real-time data in public, private and hybrid cloud architectures. At a lavish event in Boston on Tuesday, the company hosted a coming-out party where officials and some early customers explained how the company hopes to re-invent the database for the cloud era.

The Cambridge, Mass.-based company claims its Cloud Data Management System (CDMS) called NuoDB Starlings 1.0 has unique technology (and patents) that address the issue of scaling out while also supporting traditional SQL commands and reliable ACID (Atomicity, Consistency, Isolation and Durability) transactions.

In addition, Starlings runs on commodity hardware, handles unstructured data and non-SQL models and most importantly is designed for elastic scale-out and scale-down cloud architectures. It is also designed to ensure reliability in cloud scenarios where network latency is an issue.

On Monday I spoke with NuoDB CEO and Co-Founder Barry Morris, who described Starlings as a single logical relational database management system that can scale by simply adding multiple servers. Morris argued that traditional SQL DBMSes from the likes of Oracle, IBM, Microsoft and SAP's Sybase unit can't do that at Web scale (they would beg to differ). He also claimed Starlings can process 1 million transactions per second on 24 hosts running on commodity hardware costing about $50,000, according to a Yahoo Cloud Serving Benchmark (YCSB).

"All we are doing is right-clicking and adding another machine. We are not doing any clever partitioning of this data or building of caching systems or anything," Morris said. "You just add more machines and it goes faster."

It's also designed to address the network latency issue associated with cloud computing. "The network behaviors on these clouds are highly unpredictable, in terms of latency," Morris said. "If you've got 20 to 30 to 50 machines all taking part in a database and you don't know how long it will take to send a message across the network in the Amazon cloud, that's a problem for database systems. Our system is carefully designed so all those communications are asynchronous, so we are actually quite tolerant of these cloud systems."

Pointing to a published benchmark of one of the world's fastest transaction-oriented database architectures, an Oracle Database 11g R2 Enterprise Edition with RAC and partitioning running on SPARC SuperCluster servers could cost more than $2 million per year to process 500,000 transactions per second, Morris reasoned. He cited TPC-C benchmarks (though I must admit I don't give a whole lot of credence to such comparisons since they are simulations).

At any rate Morris argued the Starlings database would cost 20 percent of a traditional RDMS topology, though he acknowledged the 1.0 release hasn't been tested in scenarios with petabytes of data. He estimates a 500 gigabyte Starlings database would cost $15,000. It can run on premises or in public clouds. The company offers up to 4 GB of data running on up to two hosts free of charge indefinitely. Add two more hosts and 16 GB or more and pricing starts at $1,200 per year, according to the company's Web site. NuoDB offers unlimited capacity free of charge to developers.

About 20 percent of the company's 3,500 early customers are running Starlings in public clouds, primarily using Amazon Web Services EC2 and Simple Storage Service (S3), though plans call for testing on other services, Morris said. Among the customers using Starlings are auto parts retailer AutoZone, IT service provider NorthPoint Solutions, and food service and hospitality supplier Compliance Metrix.

Time will tell if NuoDB shakes up the database market as it claims it will. But given the credentials and brain trust behind the NuoDB team, the company is certainly worthy of attention.

The inventor of the NuoDB Emergent Architecture that powers Starlings is Jim Starkey, who in the late '70s developed Rdb at Digital Equipment Corp., the core relational database engine that DEC ultimately sold to Oracle in 1994. Starkey later went on to found Interbase Software, which provided popular RDMBSes for engineering workstations. In 2000, he founded Netinfrastructure, where developed a Web-scale relational engine that was later acquired by MySQL. At the end of last year, his work done at NuoDB, he retired.

Morris' credentials include CEO stints at Iona Technologies, a provider of popular transaction-oriented middleware that was acquired by Progress Software, and later Streambase Systems, a popular supplier of complex event processing software.

Also worth noting, as Morris pointed out, is all of the code was developed by the company. For better or worse, it's not based on open source code. If you're testing Starlings, drop me a line at jschwartz@1105media.com or share you observations in the comments section below.

Posted by Jeffrey Schwartz on 01/15/2013 at 12:48 PM


comments powered by Disqus

Virtualization Review

Sign up for our newsletter.

I agree to this site's Privacy Policy.