The "NewSQL" category of databases targets the same scale-out capabilities and performance levels of NoSQL databases, while maintaining the relational model and database consistency of mainstream databases like Oracle, SQL Server, and MySQL. One of the products in the NewSQL category is NuoDB, a company I've written about before.
NuoDB explicitly targets cloud workloads, as the scale-out capabilities NuoDB offers align well with the elasticity characteristics of cloud computing. Under the latter, cloud infrastructure can be dynamically provisioned or de-provisioned, based on the ebb and flow of user demand, and NuoDB fits itself in just such scenarios.
Cloud via Moon(shot)
Enter project Moonshot, HP's new server initiative that focuses on high-density concentration of servers, each with very low-energy consumption. HP announced availability of its Moonshot offering, based on Intel S1260Atom processors, on April 8. With that product, a single 4.3U system contains 45 discrete physical servers.
I spoke with Seth Proctor, NuoDB's chief architect, who explained that NuoDB wanted to see how many databases it could get running on a single Moonshot box, in terms of total databases hosted on the system and the number of databases concurrently active at any given time.
This structure of the metrics is based on a multi-tenant hosting scenario, where multiple blogs are hosted on a single Moonshot system, and only a fraction of them encounter page view traffic in a discrete instant. The kicker: NuoDB can shut down the databases that are not in active use.
The result: NuoDB said it can serve 7,200 active databases and 72,000 total databases, on a single Moonshot system. Even under that load, the company said the server encounters only 70 percent utilization, so the numbers could theoretically go higher. NuoDB said the screenshot below shows a management dashboard readout from a Moonshot test run. The numbers are slightly lower than quoted above, but with server utilization at 67 percent, that seems OK.
Going beyond the high-density scenario is important, too. If a given blog hosted on the Moonshot system sees a significant spike in traffic, it may make sense to move that database to some bigger iron, with the option of moving it back to the Moonshot box if and when the spike subsides. NuoDB claims it can perform such a migration while the database remains live, online, and hot. The company calls this database "bursting", and if it works as advertised, it would seem a sensible companion to the high-density capability.
NuoDB has submitted a patent application for the bursting capability, along with the ability to "hibernate" databases when they're not in use and wake them up on demand. Proctor told me the latency involved in waking databases is only about 200 milliseconds on a Moonshot system, and significantly less on servers based on higher-powered CPUs.
The NewSQL movement is intriguing, as it seeks to forge a consensus between the decades-old client-server relational database stalwarts and the bright, young NoSQL upstarts. It's not clear whether the products in this category will carve out their own long-lasting market share, or if the old-line relational vendors will adopt some NewSQL architectural ideas. Either way, it's fascinating to see innovation in the relational space, which had matured and been rather static for the first decade of this new century. Together, NuoDB and HP seem to be pushing the envelope.