The rise of scale-out network-attached storage
A comprehensive collection of articles, videos and more, hand-picked by our editors
Panasas today unveiled its new ActiveStor 16 high-performance computing scale-out NAS appliance that supports HGST UltraStar He6 6 TB helium drives. To handle the larger capacity drives, Panasas developed per-file triple-parity RAID that it said becomes more reliable and rebuilds drives faster as the system scales.
Like previous Panasas appliances, ActiveStor 16 consists of director blades and storage blades. The director blades handle metadata access and system management. The storage blades contain hard disk drives, solid-state drives (SSDs) and cache.
Each storage blade can hold 122.4 TB of hard drives per 4U shelf and scales to a total of 12 PB with the 6 TB helium drives. The shelves also support 1.5 GBps throughput and scale to 150 GBps in a global namespace. Each storage blade contains 240 GB of SSDs used to store metadata and small files and 8 GB cache.
The latest PanFS storage operating system performs RAID 6 triple-parity data protection that increases reliability by 150% compared to dual-parity RAID. The RAID uses erasure codes and rebuilds only the affected files instead of entire drives. An Extended File System Availability feature keeps the namespace online for business continuity during a catastrophic failure.
Geoffrey Noer, Panasas' senior director of product marketing, said ActiveStor's per-file distributed RAID architecture turns files into a collection of objects distributed across all blades. Small files are triple mirrored, while large files are striped. The more drives that are added, the less likely that three drive failures will affect a file.
Unlike with traditional RAID, Panasas said its drive rebuilds actually get faster and the percentage of file restores decreases as more drives are added. With 100 drives, you would have to restore only one out of every 200,000 files, Noer said.
"Our system, a system that is 10 times larger, has rebuild times that are 10 times faster," Noer said. "That's because of the way we distribute data across the entire system. So when there is a rebuild, you are using every drive in the system. And the more distributed you get, there is less risk to any particular file."
Noer said the RAID has extra protection for directory data, so if there are three simultaneous failures the system knows which files need to be restored.
"The design claim here is counterintuitive. The larger the system, the more reliable it gets rather than less reliable, which is the norm," said Steve Conway, IDC's research vice president for high-performance computing (HPC) and data analysis. "If it does what [Panasas] claims, it's a breakthrough. No one else is claiming the bigger the system, the higher the reliability."
Conway said HPC, originally used primarily for modeling and simulations, is going mainstream as commercial companies dive deeper into big data and data analytics. HPC systems now have an entry price that is less than $50,000 compared to $1 million. However, as they take on greater capacities and drives, the risk of failures increases.
"Some of these supercomputers have over 1 million cores per cluster," Conway said. "It's the Sword of Damocles -- people know it's hanging over their heads. There's a high level of concern. It's not unusual to see online capacities in the 5 PB to 25 PB range. So if Panasas lives up to its claims, it will be a breakthrough. It does not mean others won't do it, but this is the first announcement of someone actually implementing this."
The new platform will become available this month. An ActiveStor 16 with 82.4 TB of 4 TB hard drives and 240 GB of SSDs has a list price of $130,000; a configuration with 122.4 TB of 6 TB helium drives and 240 GB SSDs is $160,000.