|
Reads:34287Replies:0
X-Engine Compactionhttps://www.alibabacloud.com/blog/when-databases-meet-fpga-%E2%80%93-achieving-1-million-tps-with-x-db-heterogeneous-computing_594147?spm=a2c5t.10695662.1996646101.searchclickresult.30f57c84HIgYFb X-Engine maintains multiple memtables(->immutable) in its memory -> flushing to persistent storage SSTables+ regular compaction+ latest KVs. Intelligent tiering of data + compression based on temperature. CPU-intensive ops on compaction addressed by VT-tree, bLSM, PE, PCP, and dCompaction for LSM-tree based engine. Hybrid design , FPGA is integrated as a real co-processor( Host connected to FPGA via PCIe) is used rather than bump-in-the-wire( Not much details n explanation though) usage of asynchronous hardware device FPGA, rather than the CPU to complete the compaction operation. Not sure of ECC, CRC , how fast and accurate the back and forth computing . It leads me to think of another design aspect as L0, L1,L2 can engender intensive resource usage.   |
|
|
