Stellus Emerges From Stealth With Data Platform That Drives Massive Data Access Throughputs
Introduces Key-Value over Fabrics technology to accelerate performance for unstructured data.
This is a Press Release edited by StorageNewsletter.com on February 5, 2020 at 2:15 pmStellus Technologies announced its Data Platform, based onKey-Value over Fabrics technology, that can efficiently accelerate by 10x the storing and accessing of the world’s unstructured data.
The platform delivers groundbreaking read and write performance for file and object-based data, necessary for the most demanding data-intensive workloads and applications such as cognitive AI, machine learning, high-speed microscopy, genomic sequencing, and data streaming applications.
“The explosive pace of data creation is driving the need for massive storage and processing of structured and unstructured Big Data in cloud-native environments,” said Young Sohn, president and chief strategy officer, Samsung Electronics and chairman of Stellus. “The introduction of this innovative new-generation file system by Stellus will unleash the full performance of industry-leading memory and storage created by Samsung, and deliver the flexibility and scalability of cloud-native architectures for customers and partners.”
The Data Platform is a software as a service-based data system that leverages cloud, c ore and edge infrastructures to enable hyperscale throughput for digital enterprises. Key-Value over Fabrics technology can make unstructured data access faster and more resource efficient than other data storage architectures available today. It allows businesses to capture, store and process increasingly large data sets from an increasing number of data sources, which is critical to power new parallel data analysis workloads.
“We see significant demand from customers looking for the unique combination of Key-Value over Fabrics technology, high-throughput Key-Value Stores and algorithmic data locality,” said Jeff Treuhaft, CEO at Stellus. “Stellus Data Platform turns storage investments into a massively parallel ‘data computer’ that consistently delivers fast parallel data access
performance that scales efficiently for core, edge or cloud-native infrastructures.”
Unlike traditional file systems for unstructured data that support scale-up or scale-out storage strategies, the Stellus Data Platform allows for scale-through performance and efficiency. It uses Key-Value over Fabrics technology to scale-through the hybrid cloud, core and edge infrastructures struggling to keep unstructured data flowing quickly and efficiently enough to the most critical workloads.
ESG conducted a technical evaluation to validate the consistent performance of the scale-through architecture, which can be found here (video and whitepaper).
“Stellus gives companies a choice of throughput or capacity or both, all within the same key-value-based platform. It’s no longer about choosing to scale-up or scale-out. Stellus’s ‘Scale-Through’ technology provides digital enterprises with a clear and unprecedented path to efficiently and effectively manage unstructured data at massive throughput and scale. Stellus brings hyperscale capabilities to every digital enterprise, eliminating the requirement to compromise,” said Steve Duplessie, ESG founder and senior analyst.
Comments
Company's profile
Location
Jeff Treuhaft, president and CEO, held executive product management, marketing, and corporate development positions at Fusion-io, Zetta, Verisign, and Netscape.
- Bala Ganeshan, PhD - CTO and VP engineering, held the CTO roles at EMC and QLogic. Additionally, he has held research roles as assistant professor at Virginia Tech and architect role at AT&T and Delta Air Lines.
- Ken Grohe, CRO and SVP, was president of SignNow, SVP and GM at Barracuda, CRO at Virident (WDC acquired) and spent 25 years at EMC, culminating in running the flash division.
Our comments:
Stellus claims to play in high-performance data systems with a memory native file system delivering a new offering in file storage. Interestingly it received a significant investment of Samsung illustrating a form of system development try for the South-Korean company.
Via the launch of the company, we had access to the presentation and we're surprised by the file system listing the company promotes indicating a strange view of the market. It mixes disk file systems like ZFS and WAFL with parallel ones like GPFS and Lustre and OneFS as a distributed one (but not parallel). It also creates confusion as some of them are external file system (Lustre, GPFS, etc.), we mean used by applications servers directly and other internal (OneFS) i.e not seen and touched by these machines, in that case EMC Isilon exposes NAS protocols (NFS and/or SMB on top). Also, several high performance file systems (external and internal in our terminology) are missing such WekaIO WekaFS, Panasas PanFS, VAST Data 'Transaction File System' or Qumulo QFS, and this slide doesn't give us a good feeling. As Stellus promotes Key-Values, it would have been good to mention Pure Storage FlashBlade with its KV layer embedded below its NFS service. The comment on the right of the slide is incredible as WekaIO, Qumulo and VAST Data designed their file system with flash in mind and not HDD. Why this miss?
Stellus with their Data Platform (SDP) promotes a shared-everything architecture with Data Managers (DM) - their access node or compute server (1U) - able to access to Key Value Stores (KVS) nodes - their storage server (1U) - via NVMe-oF. DMs are stateless and the minimum configuration is 2 DMs coupled to 2 KVS nodes for a capacity of 369TB raw with 8TB NF1 NVMe SSD (we suppose). The maximum capacity is 1.4PB raw. A KVS node provides 32 SSDs for a native capacity of 256TB but 184TB effective and each SSD manages 4 KVS so in total 128 KVS per node. It shows an interesting disaggregated model than tends to grow on the market. Also the design offers a high degree of parallelism between DMs and KVS nodes and even to several KVS on each KVS node.
The web site and presentation don't mention access methods, try to find NFS for instance, on the web site. We understand that SDP supports NFS, SMB and S3 and we didn't get confirmation of a Posix client.
For data protection, SPD uses erasure coding model, a good fit when file system and file size are bigs.
In term of performance, SDP delivers above founders expectations with 80GB/s of throughput coupling an interesting more than 40GB/s of reads and writes at the same time. This performance is delivered with 4 Data Managers and the SDP440 model. We understand that performance direction is bandwidth and not IO/s, at least today, as only throughput numbers are given, meaning that the company targets some vertical use cases like M&E, oil and gas, life sciences, etc., and some HPC related applications with large file sizes and large I/Os. The system seems to be pretty linear as DMs are added (SPD 220 - 420 - 820) with a decoupling of performance and capacity (SPD 280 - 480 - 880), the first number representing the number of DMs and the two other the number of KV Store to offer the capacity.
In the middle, between DMs and KV Store, Stellus supports a NVMe-oF via Ethernet switch interconnect.
To illustrate such performance need, the company targets 2 use cases in M&E and life sciences but again the interface between applications servers and DMs are not mentioned. As a file storage vendor it could be NAS protocols such NFS and/or SMB but also a Posix client.
As Kinetic in the past and several other vendors' developments, we anticipate that Samsung will offer KV drives in the future able to connect directly in such Stellus architecture. We remember drives with embedded RocksDB or LevelDB.
Read also:
Stealthy Start-Up Stellus Technologies in SSDs for Servers
100% subsidiary of Samsung
May 5, 2016 | Press Release
A newcomer Stellus Technologies
August 23, 2019
Subscribe to our free daily newsletter
