Raidix Interstellar Software Scales to 512PB and Achieves Productivity of 2TB/s

Raidix, LLC is expanding its product line with a software platform for building storage clusters – codenamed ‘Interstellar’ – that it expects to launch late this year.

The product brings essential HPC storage to enterprise customers. Interstellar blends the company’s HPC RAID engine and cluster-in-a-box technology with Intel Enterprise Edition for Lustre software, revealing a turnkey software platform for building and managing high performance storage clusters.

Sergey Platonov, owner, Raidix product said Interstellar will scale to 512PB  and achieve 2TB/s and ultra-high throughput and IO/s.

“It’s a high-performance platform, so it gives customers a way to slice and dice IO/s any way they want. Raidix is literally stuffed with features and is powered by its own patented RAID algorithms and calculation methods, delivering productivity of 37GB/s (RAID-6) and 25GB per second (Raid 7.3) per processor core,” Platonov said. “The hybrid Raidix Interstellar storage platform can accommodate several hundreds of HDD drives and SSDs, allowing a large RAM cache and extra-large SSD cache”

As for the Interstellar appliances, Platonov outlines that the software runs on commercial off-the-shelf (COTS) hardware. The 42U storage rack stably drives out up to 70-80GB/s in performance.

“You can get the storage cluster up and scale it out using SSDs or spinning disks as your needs evolve”, he said. “Moreover, Lustre object servers, metadata servers and targets are installed directly on storage nodes, saving on infrastructure and allowing more budget for compute nodes.”

Equipped with the HPC RAID engine, Interstellar boasts fast and reliable failover, high-performance data processing, and broad functionality for data integrity and monitoring.

The clustering system of company’s HPC creates a fault-tolerant high-performance cluster (by configuring dual-controller mode) and places RAIDs asymmetrically on the nodes. Each RAID can be accessed through either of the nodes.

Also, the company’s solutions uses erasure coding based on proprietary patented RAID algorithms optimized for high performance tasks. Supported RAID levels include classical RAID-6, RAID 7.3 and RAID N+M, customizable to allow from two to a fully configurable number of drive failures. Choosing any of the RAID levels for storage not only secures its integrity, but also guarantees high sustainable performance while in the degraded mode – the system does not experience any performance losses.

Besides RAID levels, two additional features ensure data integrity and storage productivity: Silent Data Corruption Detection and Correction (SDC), and Partial Reconstruction and Advanced Reconstruction. SDC forward silent error correction algorithm analyzes RAID metadata in the background mode, and detects and fixes corruptions and soft errors without impacting the performance rate.

The Advanced Reconstruction mechanism accelerates storage performance by eliminating the slowest HDDs from read operations. Partial Reconstruction allows restoring only a particular area containing corrupted data on a HDD, reducing the array recovery time. Partial reconstruction is extremely effective for large arrays.

Intel Enterprise Edition for Lustre software OSS+OST (Active-Active) and MDS+MDT (Active-Passive) are installed directly on the storage nodes, saving on infrastructure and allowing more budget for compute nodes. Failover of Lustre OSS and MDS is processed by company’s cluster (based on storage events). Lustre* services work with volumes locally which mean low IO latencies and increased performance.

Intel Enterprise Edition for Lustre software PFS helps eliminate risks associated with limitations of an open source solution, enabling simplified installation and configuration, consolidated view with global namespace, comprehensive monitoring tool (Intel Manager for Lustre) and superior enterprise technical support from Intel.

“We are pleased to welcome Raidix to the Lustre community,” said Brent Gorda, GM, high performance data division, Intel Corporation. “They bring a deep understanding of high performance storage software and its tailored integration into industry-specific solutions.”

Enterprises with high performance workflows encounter the storage ceiling of traditional SAN and NAS architectures. Reservoir and production engineering in oil and gas, financial engineering in the financial sector, collision simulation in machine manufacturing all require massive throughput and highly parallel access to data. Standalone parallel file systems offer a higher performance alternative, but those brave ones who attempt to deploy a four-square storage riveted up with open source file systems are bumping into deployment, management, performance, scaling and support issues.

The company has tested Interstellar software on Echostreams, Supermicro and AIC hardware.