The IBM General Parallel File System (GPFS) is a high performance file system designed to scale to multiple Petabytes (PB). GPFS is typically used in high performance computing (HPC) and cloud-scale file storage environments. As the volume of unstructured data continues to explode, it’s no longer uncommon for companies to be looking for solutions that require support for billions of files in a single volume. One of the unique challenges of GPFS is the storage and management of metadata. A fast reliable metadata server is required to store the index information for each of those billion files so they can be quickly located and accessed.

Violin Memory Arrays are low latency, high bandwidth storage systems based on flash memory. When used in conjunction with GPFS to store metadata overall storage performance is dramatically increased. With billions of files, the task of managing back-up, replication, and migration across performance tiers is not possible with existing disk-based solutions. Violin memory arrays provide the consistent performance and high availability to achieve this scale today. A single 3U enclosure with 10TB of capacity can deliver over 220,000 IOPS with latency under 250 µsec, supporting 10 billion files without reaching performance limits. Redundancy and Violin’s unique vRAID technology ensure high availability and data protection to keep critical services running.

Violin Memory Arrays offer the following benefits:

• Low latency, high throughput access to data and metadata
• Scalability to support Petabyte-scale deployments
• Network-based storage for efficient collaboration
• Data protection and seamless fit with backup and restore

Performance and Scalability Characteristics:

• Up to 40TB in a single 3U enclosure, over one half a petabyte in a rack
• 250,000 IOPS from a single Memory Array
• Low latency – under 250 µsec
• Network attached via Fiber Channel (FC) or 10 Gigabit Ethernet (GbE)
• Unique flash vRAID for reliability and data protection

Violin Memory in Action: Breaking GPFS Performance Record by 37x

In July 2011 the IBM Advanced Storage Lab conducted performance tests to simulate policy-guided storage management tasks such as file selection for backup and migration for a 10 billion file environment. The test was run on a cluster of ten IBM xSeries servers with GPFS; the metadata was stored on four Violin Memory Arrays. The results were dramatic, shattering the previous world record also set with IBM. The active ten billion-file file system was scanned in 43 minutes, shattering the previous record by a factor of 37x.

As noted in the IBM whitepaper:
“SSDs on a PCIe card will not address the high availability requirement and many SSDs we evaluated didn’t sustain performance”

Read the IBM Whitepaper