The main idea, learned from the Internet giants, is to provide an enterprise class storage that resides on-premise with all cloud benefits. It reminds me what I preached and changed a few years ago at a storage software company that tried to compete against SAN and NAS, their initial message was a real failure. This Igneous model means remote management operated by Igneous, autonomy and automation, high scalability, high availability and data durability, failure domain optimization, ease of deployment and ease for applications but you, the user, access your data at LAN or local speeds, within your datacenter perimeter and without any access fee.
I was intrigued for a few years by Igneous' project, I met them finally at NAB 2016 last April where they displayed a version of their 4U system and then wrote 2 shorts posts about them (1 and 2). The second post written 6 months ago explained already approximately everything Igneous announced recently. Of course more details are given now with officially the Marvell ARM processing board embedded on each drive delivering the next phase of a sort of Ethernet Drive. This nanoserver is based on the Marvell Armada 370 processor, 32-bit processor with two Cortex-A9 cores running at up to 1 GHz with an Linux instance, and the board has 2 x 1Gb/s Ethernet ports. The RatioPerfect concept introduced by Igneous refers to the 1:1:1 from CPU to I/O to disk. It also has the beauty to reduce a failure domain to one disk so very manageable instead of 60 to even disks attached to one x86 server.
Globally Igneous is a 6U offering with 2 x 1U for the dataRouters - exposing the access methods - connected to a 4U system named dataBox - representing the content store - with all drives vertically organized and plugged. A 4U system offers 212TB usable built with 60 drives of 6TB organized in 2 x (20 + 8) erasure coded drives sets. Today the interface exposed is an Amazon S3 compatible API and we can imagine that this current object access mode is a just the beginning... we can anticipate other services with computing... Two dataRouters are preferable for HA and globally dataRouter and dataBox scale independently but when dataBoxes are multiplied, the 1:2 (ratio dataBox:dataRouter) is optimized. As an example with 4 dataBoxes (848TB), 3 x 1U dataRouters are generally deployed and recommended, still offering a N+1 redundancy. The dataRouter instance, stateless by nature, is also responsible for the management and remote control of the deployment.
This system is priced at less than $40k per year meaning that 1$ spent each year offers 5.3GB of usable capacity or 1 usable TB costs $188 per year. Super idea, very impressive and well developed.
We'll learn more about Igneous very soon as we'll meet the team during the 21st edition of The IT Press Tour in December.