mleku on Nostr: i'm gonna be using this on my fiatmine server build as well, because it was blowing ...
i'm gonna be using this on my fiatmine server build as well, because it was blowing over 1gb of utilization and heroku servers were killing it for this
similar reason as the import function blowing out 2gb of memory on default GC and memory settings - doing too much work for the space to do it in
in the fiatmine server this means the initial sync can now be capped and maintained under 700mb of usage and normally it will run at under 200mb of memory
i will be looking at figuring out a divide and conquer strategy for the computation it does when doing a massive set of comparisons, currently it just traverses a graph left-to-right and then down meaning it has to decode everything as many times as there is things to decode
i think the better traversal path is to go across the diagonal like a 45 degree line wiping across from one corner to the other, and the peak memory utilization when it gets to the either side of the center will be the square root of 2 times the number of elements, roughly 1.5, this will be huge
similar reason as the import function blowing out 2gb of memory on default GC and memory settings - doing too much work for the space to do it in
in the fiatmine server this means the initial sync can now be capped and maintained under 700mb of usage and normally it will run at under 200mb of memory
i will be looking at figuring out a divide and conquer strategy for the computation it does when doing a massive set of comparisons, currently it just traverses a graph left-to-right and then down meaning it has to decode everything as many times as there is things to decode
i think the better traversal path is to go across the diagonal like a 45 degree line wiping across from one corner to the other, and the peak memory utilization when it gets to the either side of the center will be the square root of 2 times the number of elements, roughly 1.5, this will be huge