• Just use the global flood control variables directly rather than calling the global from a function in net.cpp.
  • Change the names of the global vars to indicate they are MAX values.

• In net_processing.cpp we shoul use TxMayAlreadyHave() as it is much more performant that TxAlreadyHave(), since there is no need to take locks on the underlying data unless we at least suspect that we have it.

• cleanup mempool sync code
• fix use of the message cookie

Enter the issue description

• With the recent bug fix to the net code where under heavy load the network was getting deadlocked, we can not remove the requestPacer which was acting as a kind of band aid, preventing that network bug from getting triggered. Instead we can just limit the request manager maps from getting too big...for transactions we can limit to 1Million and for blocks just 2 times the block download window should suffice.

• Limit the number of sources we can add to 5, for transaction requests. It's very rare to get a re-request for a transaction so there's no need to use up so much memory for nodes that have many peers connected.

• If a peer is misbehaving then set the "desirability" of the peer to zero when adding to the source list.