newscast: diffuse our ID with a higher timestamp during the bootstrap periodIn typical applications, peers bootstrap from the same
small set of nodes. Therefore, at the beginning, their
cache is not a uniform random sample. As a consequence,
their messages could hit bootstrap nodes aggressively,
also modifying the randomness of those caches.
We avoid this by implementing an initial grace period
when the entering peer's cache is randomized without
disturbing the system...
newscast extension: adding reply tokensWe deal with the following problem: our ID is not diffused well in the system
- problem: we do not receive queries, thus, we do not send replies, so the situation gets even worse!
- indicator: not receiving queries
- action: taking an active role and sending out a query instead of a reply
- note: again, we work on averages here
handling of reply_tokens:
- when a reply is sent (a query was rece...
newscast extension: adding query tokensWe deal with the following problem: our query is lost (either due to packet loss or due to dead peer). In this case we try to send another query through
- loss indicator: not receiving a reply
- action: sending another query
- note: as you'll see, it is enough to keep this on average, no need to track sessions
- note2: we could try to assume symmetric packet losses and thus assume that half of ...
net-helper-ml: changing to another STUN servergit-svn-id: https://repository.napa-wine.eu/svn/napawine-software/branches/winestreamer-1.0-rc@2624 c54fac2e-1815-407e-9b96-088d06743913