Matthew Toseland

Fix burst only

12

Various cleanups, mostly directed at fixing logging errors. Relatively low risk but could conceivably change connectivity behaviour on non-NATed nodes.

Update the tray (and possibly the wrapper) on Windows

14

This allows updating FreenetTray.exe and could also be used for updating the wrapper binaries. That is important because we may lose large numbers of Windows nodes due to Oracle moving...

Faster simulations (summer project)

3

This will likely need significant refactoring. The main goals are to enable fast unit tests involving multiple actual nodes and to demo load management simulations; to this end I implemented...

Async chkinsertsender riskier cleanup

7

This branch makes CHKInsertSender and CHKInsertHandler largely asynchronous, and greatly reduces the number of threads needed on high bandwidth nodes. It has been tested reasonably thoroughly, both on my local...

Fix synchronization on _matched

4

This might have been valid originally but broke because of adding asynchronous filters.

Random initial routing

15

It looks like this should actually improve performance, at least for inserts. It should also improve anonymity. It should be tested; any performance effects will be visible on a local...

Throttle unmatched scan

2

This needs more testing and probably more work. However review would be helpful. See https://bugs.freenetproject.org/view.php?id=6773

Lazy start RequestStarter's

9

This saves a few threads at the cost of some complexity. Its main purpose is to make it easier to simulate really big networks for one-request-at-a-time simulations, e.g. for extensive...

Largish changes to in-RAM temp buckets

1

Major changes here: - Track last access to any temporary bucket/RAF. - Always migrate buckets to disk [configurable period] after last access. - Use less memory for temp buckets by...

Load management send slow down when close to limit (without strict fairness)

2

This makes fred send a slow-down signal when load is above 75% of the node's capacity, rather than only sending slow-down signals when we reject requests. The idea is to...