Dogecoin Tuning
This contains a handful of small optimizations following recent Dogecoin load:
- Based on testing, the increased batch size appears to result in faster completion of a given scan block even on daemons under load and considering the larger batches are heavier individually.
- The
dbcache
should provide some small improvement in disk utilization. - The
rpcthreads
increase has no observed effect, but there is testing in flight to add concurrency on top of batching in a future patch release where this may become relevant. - The
minrelaytxfee
will drastically decrease the node-local mempool count. Some small fee inbounds may be skipped until they are mined, but that should pose no problem. Separately all outbounds tend to have over an order of magnitude higher fee - but in any case, observing outbounds in the mempool should no longer matter, since the insta-observe now prevents cross-block reschedule.
The above tuning is likely insufficient to keep up with tip under current circumstance. While new UTXO package is significantly more efficient, Dogecoin transaction rate is significantly higher than other UTXO chains nearing all time highs, and this compounds with Dogecoin scanning being less efficient than the others in the first place due to the lack of a getblock
with verbose transaction support. There is work in flight to optimize this further in Bifrost, but that will not come until the next release.