Coho Data emerged from stealth mode Tuesday and launched Coho DataStream, its scale-out, object-storage platform...
that combines its software-controlled switch with PCI Express flash and spinning disk.
Designed to target storage scalability, the Coho DataStream hybrid 3U device includes a 2U storage "micro array" with two 800 GB multi-level cell (MLC) PCI Express (PCIe) cards, six 3 TB spinning disk drives and two 10 Gigabit Ethernet (GbE) ports. The third rack unit houses a 52-port 10 GbE switch.
Each node includes 39 TB of raw capacity, 19.5 TB of usable capacity, and can produce up to 180,000 IOPS.
Micro arrays replicate data between two nodes for high availability. Because each node is equipped with compute, networking and storage capacity, adding nodes means linear capacity and performance scaling. Customers can add a second switch for higher availability. Each switch works in active-active mode in regular operation.
"The software layer we've built acts similarly to what hardware virtualization does to the CPU," said Andy Warfield, founder and chief technical officer at Vancouver, British Columbia-based Coho Data. "It's really decoupling the management and presentation of the data from the underlying physical asset management."
Warfield and Keir Fraser, another Coho Data founder, were part of the group at the University of Cambridge Computer Laboratory that authored the Xen virtual machine monitor that Citrix Systems Inc. acquired and turned into XenServer. The company formerly went under the name of Convergent.io and has flown under the radar for the past two-and-a-half years with funding from Andreessen Horowitz.
"The premise of what we're doing is basically that storage in the enterprise has for 20-plus years really been about selling washing machines," Warfield said. "We've set out to build a storage system that you can deploy inside an enterprise that has a lot of the properties in terms of adoption and scale that you get with a hosted offering."
As a node is added to the DataStream array, the management system recognizes it and starts migrating data to the new node to balance the workload. Warfield said the system overcomes traditional NAS bottlenecks and storage scalability issues.
"By doing NFS integrated with the switch and the scale-out nodes, we don't have the historical NFS bottleneck problem," Warfield said. "As you add more nodes, the NFS protocol implementation scales out with the system. That's how we get linear scaling. We're not stuck bottlenecked on a single NFS controller."
Mark Peters, a senior analyst with Milford, Mass.-based Enterprise Strategy Group Inc., said he sees similarities between DataStream and cloud-based services. "The attraction is scalability," he said. "The cost, of course, comes down because in essence what they are doing is converging networking and storage together. And it's all standard components."
Device management is done through a Web-based graphical user interface with virtual machine-level granularity and reporting. It has hot-swappable physical components, and the physical servers that house the PCIe cards are also hot-swappable.
Warfield said the Coho DataStream roadmap includes adding deduplication in the 1.5 release, possibly next year, as well as remote replication. Pricing starts at $2.50 per gigabyte.