Open source object storage software startup OpenIO is easing into branded hardware.
The France-based vendor, which recently opened a North American headquarters in San Francisco, is previewing its dense ServerLess Storage (SLS) appliance to cloud service providers and hyperscale enterprises.
SDS also runs on industry-standard servers and uses a grid-based approach to run applications directly on storage infrastructure. The vendor said it has about a dozen hyperscale enterprise customers, including video-sharing site Dailymotion.com.
The OpenIO SLS hardware is based on a Marvell Armada 3700 system on a chip with a dual-core ARM v8 Cortex CPU. The chassis has two Marvell Prestera six-port 40 Gigabit Ethernet switches for front-end client connection and back-to-back chassis expansion.
OpenIO SLS packages 'scale-out infrastructure in a box'
The ARM-based CPU architecture is designed to boost power efficiency. Up to 96 hot-swappable nano nodes can be placed in a single SLS-4U96 chassis. The nodes connect via embedded Ethernet switches.
Each compute node is a dedicated domain that manages a single high-capacity disk, using a small amount of flash storage for fast metadata access. Onboard power management turns off disks when they're not in use. The vendor said an SLS appliance could withstand the loss of about 30 drives and still function.
Multiple SLS hardware devices may be combined and managed as a logical stretch cluster with synchronous replication between remote sites.
"In theory, placing their tin-wrapped software on lower cost ARM processors should reduce the cost of effective storage, hopefully without compromising on performance," said Greg Schulz, senior advisory analyst at Server StorageIO and UnlimitedIO in Stillwater, Minn. "OpenIO has some interesting capabilities to optimize its storage I/O personalities for different applications, [based on] performance, availability, capacity and economics."
The initial version of SLS supports 8 TB disks and scales to 768 TB of raw storage. Effective capacity is 595 TB with erasure coding. A 4U rack supports 10 chassis and more than 11 PB. Future SLS iterations are expected to accommodate 10 TB and 12 TB drives.
Each nano node is equivalent to a single disk and may be individually replaced during a failure. There is no consistent hashing algorithm or recalculation of key space. New nano nodes are discovered automatically and join the pool without the need to rebalance drives.
Redundant links, switches and power supplies prevent a single point of failure. A single SLS chassis includes four N+1 power supplies and five removable fan modules.
OpenIO SLS is similar to hyper-converged systems, except with ARM-based storage modules instead of HDDs. The biggest distinction is that OpenIO software allows applications to run where the data resides, said Enrico Signoretti, OpenIO's head of product strategy.
"SLS is scale-out storage infrastructure in a box. We sell it as a single appliance, but it has all the components and assembles into a full-fledged scale-out infrastructure. We have radicalized the scale-out idea (by dedicating) one disk per nano node for massive parallelization," Signoretti said.
OpenIO faces stiff competition in object storage
SDS open source object storage converts a rack of commodity servers into a large pool of compute and storage. Object-based REST APIs for Amazon Simple Storage Service and OpenStack Swift are standard, as is support for AFP, FTP, NFS and SMB file protocols.
A key software feature is OpenIO Conscience data awareness that collects service metrics from the nodes and provides a quality score to each service. Conscience distributes the scores to every client and node to optimize data placement.
OpenIO started out in 2006 to provide storage for email and messaging. The code was placed into open source in 2012 and forked as OpenIO SDS when the vendor launched in 2015.
The object storage market is packed with contenders. Signoretti said bundling OpenIO open source object storage on prequalified hardware allows it to hit a select portion of the market.
"We think we are moving in the direction of enabling our customers to do more data-intensive workloads, based on events and objects," Signoretti said.
Target use cases include backup repositories, collaboration, data lakes and storage consolidation. Signoretti said a fully populated SLS chassis costs about $170,000 and includes three years of advanced hardware replacement and SDS software support.
Get off to a good start with hybrid cloud
Will popularity of object storage make NAS obsolete?