The media and entertainment industry is under constant pressure to deliver quality results at high speeds. Video post-production and rendering require data-intensive environments commonly plagued by the slow data access of traditional hard disk drives (HDD). Fast data access and rapid processing time are essential to ensure a high-quality final product. These issues can be solved with a flash memory storage system that has the reliability, performance and economics necessary to be deployed as mission-critical primary storage in the most data-intensive video environments.
The Violin Memory 6000 is a fully-integrated, flash-based system for primary enterprise storage. With a “rack in a box” approach, the system paves the way for media and entertainment companies to leverage flash as the foundation for virtualized data centers. A single rack of 10 arrays can deliver 12 million IOPS and 48GB/s of bandwidth from 160TB of storage capacity.
The system optimally balances performance, reliability and cost to provide a high-performing solution for primary storage rather than traditional disk-based storage. This is achieved by taking a systems approach specifically designed to optimize flash performance and aggregation from the chip up. Compared to disk-based systems, Violin Memory arrays deliver low cost per bandwidth (for rendering) and latency (for post-production and streaming), and significantly lower both CAPEX and OPEX. Compared to other solid-state storage alternatives, the arrays deliver on all key enterprise requirements where alternatives are lacking. These requirements include:
- Shared network storage — Arrays attach to the network, providing a strategic resource that can be shared across thousands of servers and multiple applications. Network options are also flexible enough to allow for arrays to be directly connected to servers.
- Sustained performance — Unique vRAID technology, with its patent-pending erase-and-write hiding feature, ensures sustained performance without the read-and-write latency spikes experienced by other solid-state approaches.
- Mixed workload performance — The arrays deliver excellent performance over a wide range of read/write I/O profiles with predictable low latency.
- Enterprise high availability — There is no single point of failure. Two or more of every component exist, all of which are hot-swappable in order to ensure no downtime or loss of data in the event of failure. Flash chips are packaged into Violin Intelligent Memory Modules (VIMMs) that are organized into RAID groups (of five VIMMs) to ensure no loss of data or performance in the event of VIMM failure. Spare VIMMs are configured in the system, allowing for fail-in-place with no disruption of operation and hot-swap when convenient.
- Performance under failure — Performance must be maintained in the event of component failures, even if a RAID rebuild is required. Gone are the days when a failed hard disk took dozens of hours to rebuild, impacting application performance the entire time.
Other deployments
There are other approaches to deploying flash memory, all of which fall short on one or more dimensions:
- PCIe cards are essentially a form of Direct Attached Storage (DAS) as they are configured within a single server providing a memory extension or caching capability. This approach is poorly suited to sharing over the network across multiple applications and servers. Additionally, component failure prevents data access, cannot be hot-swapped and usually causes server failure. PCIe cards are also susceptible to potentially significant performance degradation from the flash “write cliff.”
- SSD-based arrays use off-the-shelf solid-state disks in place of Hard Disk Drives (HDD). These arrays lack the enterprise reliability capabilities required for primary enterprise storage. The use of commodity SSD components, in conjunction with standard disk interfaces and controllers, limits the ability to optimize for flash. Even though latency performance is faster, the controller becomes the bottleneck, limiting IOPS with only a few SSDs.
Violin arrays deliver high performance density. In addition to the 16TB Single Level Cell array described earlier, Multi Level Cell (MLC) systems are also available. These arrays can provide a higher density of up to 32TB per 3U system, while still delivering 500K IOPS with read/write latency under 200ms.
Summary
Violin Memory takes an integrated systems approach to deliver flash memory arrays for mission-critical shared network storage. The arrays are engineered from the chip up for optimal use of flash, sustained performance and high availability that cannot be achieved by using SSDs or PCIe cards for primary enterprise storage.
With video post-production, where a relatively small number of users are employing heavy-duty software in order to view and alter movies one scene at a time, it is absolutely critical to guarantee reliable, zero-jitter performance in order to provide a suitable product. HDD-based systems have an extremely difficult time guaranteeing isochronous delivery of video frames. Flash-based storage platforms eradicate these concerns where low latency is a given under all conditions.
Jonathan Goldick is CTO software, Violin Memory.
