French open source object storage startup OpenIO is talking about having its software do work above and beyond basic storage and operating on, or rather, inside Kinetic drives.
OpenIO received seed funding earlier this year and is hoping to have A-round venture funding in a few months.
The additional functions are things like video transcoding, analytics and protection (cf erasure coding). OpenIO software can run with 1,000s of nodes and load balances across them. It sees the real possibility of thousands of nodes which could contain as few as a single Kinetic drive.
These drives are directly accessed over Ethernet, are key:value stores, and implement Get and Put operations on data. They have been driven by Seagate; OpenIO and Toshiba have set up KOSP, the Kinetic Open Storage Project, along with others in order to build standards and an ecosystem.
OpenIO CEO Laurent Denel described new Kinetic drives coming with 64-bit ARM CPUs onboard, a gigabyte of RAM (or more we guess), and OpenIO software running in them*. The general benefit is that you need fewer servers and their system software. The corollary is that you need a whole lotta nodes, very smart nodes.
V1.0 Kinetic was maybe not so successful, Denel says, but he's optimstic about v2.
The nodes communicate with a peer-to-peer, fully meshed networking architecture. Gigabit Ethernet between the drives/nodes has enough bandwidth for this, according to Denel.
Denel says cloud suppliers like Amazon and Google are going beyond basic storage and offering things like transcoding. He wants OpenIO to do this and offer it to customers who don't want to go to the public cloud.
Noting that object storage-based Exablox co-founder Tad Hunt spent time as a governing board member with KOSP before joining Google in February to work on adding things to its App Engine, we might ponder on Google working in the same area as OpenIO.
The company will be at the next KOSP plugfest but OpenIO is not an official KOSP member. At least, not a member "yet", we might think. ®
* Why stop there? Let's have a hypervisor running in these CPUs and system apps running as containers/virtual machines under them. In fact,let's turn disk drives into servers and ... (That's enough! Ed)