Use Cases

Pliops Multiple Performance in Real-World Use Cases

The performance gap between compute and fast storage technologies is rapidly expanding. Therefore, infrastructure that balances fast access to storage with reasonable compute costs is increasingly rare. The decades-old paradigm of waiting for the next generation of processors to solve performance problems has met its match in modern NVMe flash storage. The need to divert CPU power to low-level data access makes the equation less and less sustainable.

The Extreme Data Processor (XDP) technology by Pliops answers these challenges. Designed from the ground up for modern SSD storage, the unique XDP cost-effectively offloads and accelerates data access and processing functions from the general-purpose CPU for a broad variety of use cases.

Cloud Service Providers (CSPs) tap into Pliops XDP acceleration for relational SQL databases (including MySQL, PostgreSQL, and MariaDB) and NoSQL databases (such as MongoDB, RocksDB, and Redis). With XDP, CSPs can:

  • Accelerate data ingestion and query processing rates
  • Save storage space
  • Optimize performance of cheaper SSDs
  • Achieve higher system reliability

XDP Block Device Accelerates SQL Databases

Transform inefficient random write access into SSD-friendly sequential patterns with complete application transparency. XDP eliminates wasteful double-write protocols with internal non-volatile memory. And built-in compression increases capacity and boosts performance by minimizing reads and writes: The industry-standard TPC-C benchmark runs up to 3x faster on XDP.

XDP KV Store Powers NoSQL Database Performance

Completely eliminate intermediate block-and-file abstraction layer overhead to slash data access latencies. XDP accelerates key-value performance benchmarks more than 10x as compared to the state-of-the-art RocksDB KV storage engine.

With high-bandwidth and bursty data generation patterns, HPC applications typically bounce between compute (model update) and store (model backup) phases. The sooner the model is saved to persistent storage, the faster computation can proceed to the next iteration and the job completes. For HPC, XDP can:

  • Quickly absorb high-intensity random write spikes and transform them to steady sequential writes to maximize SSD write bandwidth.
  • Saturate multiple SSD drives in parallel for linear scalability, which is vital to end-to-end application performance.
  • Increase scalability by relieving workload pressure on parallel filesystems by completely bypassing the traditional storage stack.

Server architectures have largely remained unchanged despite the demands of the rise of big data and today’s data-intensive applications and use cases. Even high-performance NVMe SSDs can experience lower performance, higher latencies, and poor Quality of Service (QoS) in the face of varying application needs. As the number of applications and the amount of data grows, data access patterns change. Simply adding more nodes can be highly inefficient. To meet the needs and challenges of this on-demand, fast data era, enterprise IT and private cloud systems require a new data architecture. With its breakthrough design, Pliops XDP can help:

  • Eliminate bottlenecks to gain full utilization of all CPU and GPU cores
  • Offload complex data processing and storage management tasks
  • Enable traditional scaleup and modern scaleout database applications
  • Reduce TCO by delivering reliable performance of high-capacity, lower-cost QLC SSDs with no tradeoffs

Pliops white paper thumbnail

White Paper

Multiply Your Data Center to the Power of X

Emerging 5G networks and IoT devices are huge catalysts for flash storage deployment at the network edge. The modern connected car alone is poised to produce dozens of TBs of data every day. With more and more AI-driven applications at the edge, storage will need to change. XDP can help:

  • Save space and offload the CPU to minimize the impact of less co-located CPU power
  • Prolong storage lifetimes with native data fault protection and reduced write amplification
  • Improve end-to-end system autonomy
  • Drive lower TCO by enabling reliable performance using cheaper flash technologies

As AI-driven applications become real-time—with the window between new data arrival and model updates contracting to zero, and as web-scale models with hundreds of billions of parameters challenge DRAM—storage systems must adapt. Real-time stream processing platforms like Flink and Ray use embedded KV stores for in-place training and serving of AI models. XDP effectively optimizes performance to:

  • Accelerate KV store and maximize overall system throughput
  • Provide high-speed random read and write access by parameter key with acceleration of 2x+ using the native key-value API
  • Align with advanced RocksDB capabilities, such as atomic snapshots, used by Flink to create consistent global state backups