Not all machines have comparable resources and deciding on an a priori "optimum" for all peers would be foolish. It would be nice to expand and contract the max peer queue size based on machine performance. Not sure yet what stats are exactly relevant to resize, but I'm thinking:
1). Avg. num of peers contacted for a given query
2). Avg. depth of Domain Tree
3). Peer Queue churn rate
4). Avg. Pending List length vs. pending list churn rate
5). Peer to Domain ratio
6). If it's possible, memory pressure
Not all machines have comparable resources and deciding on an a priori "optimum" for all peers would be foolish. It would be nice to expand and contract the max peer queue size based on machine performance. Not sure yet what stats are exactly relevant to resize, but I'm thinking:
1). Avg. num of peers contacted for a given query 2). Avg. depth of Domain Tree 3). Peer Queue churn rate 4). Avg. Pending List length vs. pending list churn rate 5). Peer to Domain ratio 6). If it's possible, memory pressure