Open shenyaqi9527 opened 4 years ago
Node sync is inefficient when trying to download the full chain from scratch.
Gero bottleneck during node sync is disk I/O, not network.
~1000 files are modified per minute in chaindata folder (Google leveldb), resulting is disk use (SSD) almost 100%. Node sync speed decreases drastically over time:
There is a big scalability issue.
There should be a different way to download / verify / store the chain data when the client is very late behind the last block.
I would add that besides disk I/O, the CPU can also be a bottleneck. Recommend AMD Ryzen 3900x or higher. The constant disk IO requires many CPU threads and larger L3 cache.
My nodes are slow to synchronize. After syncing to the latest block, after a while, it is behind the latest block. my service is 4c16G. Do I need to upgrade my service?