- getFramedAveragePeerClockSkew() now returns a long (ms);
was a Long (s)
- Implement NTP-style clock slewing so the clock is adjusted
gradually
- Implement clock strata so we prefer better clocks
- Implement a timestamper in the transport so we will periodically
update the clock even if NTP is not working
This allows the router to converge the clock instead of simply
hoping the first connected peer is correct.
- Slow down NTP attempts after several consecutive failures
- Randomize Build Message expiration to make it harder to guess hop position
- Save expired tunnel build configs for a while, so that we will still use the tunnel
and update peer stats if the reply comes in late
- Don't update our own profile for Tunnel Build Replies
- Add getRecordCount() to TunnelBuildMessage and TunnelBuildReplyMessage
so they can be extended.
- New I2NP Messages VariableTunnelBuildMessage and VariableTunnelBuildReplyMessage,
which contain the number of request slots in them.
- Convert all static assumptions of 8 slots to getRecordCount()
- Use the new VTBM if all hops in the tunnel and the OBEP or IBGW of the reply tunnel
support it, and the tunnel is 4 hops or shorter.
- Reply to a VTBM with a VTBRM of the same size
- Make BuildReplyHandler static
- Convert the currentlyBuilding List to a ConcurrentHashMap to speed reply lookups
and eliminate a global lock; don't put fallback tunnels in there
- Add new tunnel.corruptBuildReply stat
- Various cleanups and javadoc
Tested as compatible with current network, new messages untested.
- Store and report UDP clock skews even for large values, so
a badly skewed local clock will be reported to the console
- Don't shitlist for NTCP clock skew if we don't know what time it is
- If NTP hasn't worked yet, have NTCP or SSU update the clock one time
- Include failed clock skew in NTCP skew vector if there aren't many connections
- Don't include NTCP clock skews for non-established connections
- Fix framed clock skew frame size
- Report framed clock skew even if for only one peer, if NTP hasn't worked yet
- Don't log RRD errors after clock adjustment
- Reduce min skew for console warning to 30s (was 45s)
- More Java 5 cleanups
- Enable multiple parallel job runners much sooner to speed startup
- Rearrange the startup order to get the long jobs started sooner
- Don't allow the netDb readin job to clog the job queue
on a stream... AKA setting keepalive. This should fix the stuck NTCP
issue that has been bothing zzz for years.
* Set keepalive on BOB connections too, since this will assist closing
the connections in the event of a crash on a client.
