Startup InterModal Data today introduced its distributed storage system designed to run on commodity-based hardware...
that scales out into thousands of nodes for Web-scale storage configurations.
InterModal's storage software allocates and load balances capacity and I/O performance to specific workloads. Its goal is to provide the type of storage used by companies, such as Amazon, Google and Facebook, at a price considerably lower than traditional networked storage. The startup has not disclosed pricing for its systems, however.
The company, which has raised $4 million in funding, physically separates the controllers from the storage into a storage pod and a data director. The two are connected via a private, high-speed Ethernet network with parallel paths.
The storage pods contain CPU, solid-state drives (SSDs) and hard disk drives. Each pod has two storage nodes that can be all-flash or hybrid storage. All pods have RAM, and flash cache is optional.
The data directors hold controllers for stateless data access, and contain CPU, RAM and flash cache for file and block data stores via NFS or iSCSI. Each data director has two highly available data nodes that share access to write-ahead logs. A storage pod provides capacity to any data director, and a data director can consume storage from any storage pod.
InterModal Data software runs on bare metal, virtual machines or containers, and allows administrators to scale I/O resources, RAM cache, SSD cache and capacity independently. Resources can be intelligently allocated and rebalanced according to application needs.
The startup claims its software can support millions of workloads and billions of files, and allocates resources on a per-workload basis. The software creates storage pools from physical drives on each storage pod node, using the nodes' RAM and flash caches. It virtualizes each pool and divides them into logical unit chunks.
Software is sold as a perpetual license, so it can be used at no additional charge if customers upgrade their hardware from qualified suppliers.
"We have two levels of resiliency," said Dave Kresse, CEO of InterModal Data. "We can mirror data on the I/O node, and we do triple parity, double parity and mirroring on the storage node. The software creates a virtual chunk of storage that is served up to the data node and reaggregated in pools at the controller level. So, you can scale performance and capacity independently."
Kresse said if any node is oversubscribed, the logical pool can be moved to a less busy controller node.
"It's a scale-out approach, but it has two dimensions," said Jeff Kato, senior storage analyst at Taneja Group Inc., in Hopkinton, Mass. "They have capacity nodes that can scale out, and then on the front end, you have scale-out performance that can be tuned for I/Os. They are saying they can do storage at true Web scale -- we are talking thousands of nodes -- for the enterprise space."
Kato said the company is not targeting the highest performing applications that would run on all-flash systems.
"They are going after everything else," he said. "They want dramatic consolidation of all file data, big data and so on."
Although the technology is not similar, the InterModal team is heavy with former Coraid executives. Kresse, InterModal CTO Nakul Saraiya, VP of sales Keith Carpenter and VP of support Brandt Mackey all came from Ethernet SAN vendor Coraid, which folded in April.
Determine whether scale-up or scale-out is right for you
Everything you need to know about NAS systems
Examining the evolution of scale-out NAS