1. LightningDB is

A distributed in-memory DBMS for real-time big data analytics

  • Realtime ingestion and analytics for large scale data
  • Advantages in random small data access based on DRAM/SSD resident KV Store
  • Optimized for time series data and geospatial data

2. Architecture

Spark with Redis/Rocksdb key-value stores

  • No I/O bottleneck due to redis in DRAM and RocksDB in SSDs due to the small-sized key/value I/O and DRAM/SSDs’ short latency (~200us)
  • Filter predicates push down to redis and only associated partitions are chosen to be scanned

flashbase-architecture2

3. Features

  • Ingestion performance (500,000 records/sec/node)
  • Extreme partitioning (up-to 2 billion partitions for a single node)
  • Real-time query performance by using fine-grained partitions and filter acceleration (vector processing by exploiting XEON SIMD instructions)
  • Column-store / row-store support
  • DRAM - SSD - HDD Tiering
  • High compression ratio and compression speed (Gzip level compression ratio w/ LZ4 level speed)
  • Low Write Amplification for SSD lifetime