Software-defined storage (SDN) startup Coho Data has hit the headlines again, teaming up with chip making giant Intel Corp. to search for a way to redefine how data center I/O can be managed in the age of containers.

Coho Data came out of stealth in 2013, and discussed how it integrates SDN with storage to automate scale on theCUBE at that year’s OpenStack Summit. Now, the company is working with Intel to develop a reference architecture that allows Docker containers running on Cloudera’s Hadoop distro to be run directly on a Coho storage cluster based on Flash.

In an interview with Data Center Knowledge, Coho’s CTO Andy Warfield said the idea was to scale container-based micro services across a storage cluster using Intel’s Data Plane Development Kit (DPDK). The kit has been built “from the ground up”, Warfield explained, with the Hadoop Distributed File System being uses as a protocol.

Warfield says the issue they’re trying to solve is that, with network bandwidth increasing and hand processors have become faster than ever, it’s necessary to boost the performance of storage I/O in order to eliminate bottlenecks in the data center.

Bottlenecks are an inevitable prospect as more primary storage moves into flash, Warfield believes. He says Flash offers advantages in that it’s both faster and much easier to optimize I/O performance. In addition, he thinks container technologies like Docker will further exacerbate the strain on I/O by increasing the number of apps attempting to access storage resources. Whereas before, we might have seen 30 VMs running on a physical server, it’s quite common to find up to 100 containers.

To that end, Coho’s flash storage arrays can be plugged directly into the PCIe slot, while its DataStream platform allows developers to access data over a linearly scalable NFS implementation that can scale from 180K IOPs across two Coho Data MicroArrays to 1.8 million IOPs across 20 MicroArrays.

That’s impressive, but Coho has gone further because it also allows multiple profiles to be defined, by using containers instead of NFS, by plugging each data profile into its DataStream DirectConnect API.

Intel’s DPDK is a set of libraries and network interface controller drivers that utilizes an Environment Abstraction Layer (EAL) to deliver faster packet processing. It provides a programming framework for x86 (Atom, Xeon) processors, and enables faster development of high-speed data packet networking applications.

The basic idea is that by better tuning its hardware and software to Intel’s tech, Coho should be able to develop much faster systems. It currently scales out to 64 nodes, but that’s likely to grow in the not so distant future. Coho explains more in its white paper here.

Coho Data came out of stealth in 2013, offering a sophisticated SDN storage solution based on commodity hardware. It’s chief backer is Andreessen Horowitz, and it’s raised some $25 million to date.

image: Stephen Foskett on Flickr