@samoht thanks for the blog post, I have a few questions:
- when will this land in octez?
- I am looking for ways to improve performance and reduce cost of serving octez archive RPC data. The introduction of lower storage appears to make it practical to have a read-only lower storage that serves archive RPC up to a certain level. This would allow a single lower storage to be shared among many nodes, reducing cost. Could this be supported?
- what are you working on next? May I recommend support for warp-sync? This would allow a node to download its storage from other nodes with p2p, getting as quickly as possible to a state where it can validate new blocks, and then filling in backwards. This would eliminate the need for snapshot services. Some competing blockchains offer this.
Hey, thanks for your questions! I can try to answer briefly.
- This is planned to be release in octez v18. You can follow the integration on this MR
- The new design will indeed allow to store the lower and upper layer on different disks so in theory you could put the lower layer on a shared disk. However, the architecture you suggest is a bit tricky to support, as the current design is “single-writer, multiple-writes”, ie. only one node should be allowed to write in the lower, but at the same time, you want the upper layers to be in sync. I guess it’s possible to find a scheme to make this work but we’ll need to think a bit more about it. In any case, thanks for the idea, it’s an interesting scenario!
- Our next focus is on improving performance (in terms of disk space, and I/O access). We planning to start benchmarking and experimenting with various potential improvements very soon. I’ll add Warp-Sync-like to our todo list
1 Like