- /// If we've already processed a node as the best node, we shouldn't process it again. Normally
- /// we'd just ignore it if we did as all channels would have a higher new fee, but because we
- /// may decrease the amounts in use as we walk the graph, the actual calculated fee may
- /// decrease as well. Thus, we have to explicitly track which nodes have been processed and
- /// avoid processing them again.
- was_processed: bool,
+
+ // The last 16 bytes are on the next cache line by default in glibc's malloc. Thus, we should
+ // only place fields which are not hot there. Luckily, the next three fields are only read if
+ // we end up on the selected path, and only in the final path layout phase, so we don't care
+ // too much if reading them is slow.
+
+ fee_msat: u64,
+
+ /// All the fees paid *after* this channel on the way to the destination
+ next_hops_fee_msat: u64,
+ /// Fee paid for the use of the current channel (see candidate.fees()).
+ /// The value will be actually deducted from the counterparty balance on the previous link.
+ hop_use_fee_msat: u64,
+