Enter The Freenetrix
The Freenet Help Site
Enter The Freenetrix
Licences used on this wiki
Welcome admin to Freenet WikiServer
Edit Page - Diff - Revisions - Title Search - Preferences - Bottom
Unstable build 60257 is now in CVS. The snapshots will be updated soon.
  • Disable primary failure table. Reasons for this have been discussed at some length previously; essentially, it's quite possible for the failure table to deny access to content either maliciously or accidentally, when there is no good reason for it. Worse, it can probably propagate and self-sustain this denial... Until recently the network wasn't specialized enough to get lots of FT hits; it seems that stable at least is getting many FT hits now, judging from posts on support, and that the FT is getting in the way. Some of this is guesswork but the theory behind removing the FT is well established.
  • Add a parameter for the maximum number of retries for fproxy splitfile fetches. Default is 50. This is buggy, will be fixed at some point, produces an error on startup.
  • Simulator changes: implement "backpassbug", an emulation option for an old simulator bug.
  • Transport: minor fixes, logging, minor constructor refactoring.

Unstable build 60251 is now available from CVS. Snapshots will be
updated later, when we have sorted out the situation with stable. If you
can upgrade, please do so.

  • Fixed yet another insert deadlock in Insert Awaiting Store Data. This was causing (some of?) the insert deadlocks.
  • Add a Reason field to the FCP Restarted events. This is displayed by freenet.client.cli.Main so far.
  • 60250 (didn't commit :)):
    • Major simulation fixes
    • Fixed the big problem with restarting a simulation from a checkpoint; this was simply that the Keyspace Estimator Factory was initted too early with the result that restarted sims were 10x slower because they used slow estimators instead of fast estimators (if specified).
    • Only snapshot every 30 minutes.
    • Minor bugfixes in simulation. Logging. Trivial non-simulation logging.

Unstable build 60243 is now available. The snapshots are updating. Please upgrade, test, report bugs. The major change in this build is:
Inserts now will continue for the HTL even if the data is in the store. This allows inserts to be more effective, and paves the way for the elimination of HTL. It does not make it significantly easier to flood the network, because if you want to get a single file onto as many nodes as possible for as little effort as possible, you would insert it, then request it from lots of different nodes.

Unstable build 60241 is in CVS. The snapshots are being updated. This build does even more changes to the timeouts. Hopefully no more work on this area will be needed for a while. Please upgrade, test, report bugs.

Unstable build 60237 is now available. The snapshots are being updated. This is mostly minor fixes derived from log messages.

Unstable build 60236 is now available. The snapshots are being updated. Please upgrade.
  • Made 60234 mandatory. This is primarily to test the new timeout changes, which should make inserts work better.
  • Fixed an NPE in the Current Downloads infolet.
  • Paranoia re HTLs on messages sent with HTLs.
  • Uncomment the interpolation code in Optimizing SBKE. We don't use this any more for reasons explained in a comment in the corresponding method in SBKE.
  • Logging etc.

Unstable build 60232 is in CVS. Changes:
  • Handle Transfer Failed properly in fproxy.
  • Lots of changes to timeouts. Mostly the intention is to include the queueing time in the timeout (multiplied by the HTL). This should help to prevent various bad things and may help with inserts. Also we send slightly more Query Restarted's. Minor related changes.
  • Refactoring and a whole bunch of extra options for the simulator, and some supporting scripts (which will probably need to be adapted before they are used, but have a README saying this).
  • Minor refactorings.

Unstable build 60231 is in CVS. The snapshots are updating. Please download it, test it, and report bugs. Changes:
  • New "fast estimators". Config option useFastEstimators={true |false}. Default true. If enabled, we use a slightly modified keyspace estimator implementation (important part of NGRouting) which uses doubles (64-bit floating point value, 53 bit mantissa) instead of Big Integers (full 160 bits key precision) to represent keys. This yields a speedup of 3-4 times in simulations, or similar reduction in CPU usage of routing. Simulations show it producing similar results either way, and although the simulations are on a relatively small network, I see no reason to believe it shouldn't work.
  • Another config option, doEstimatorSmoothing={true |false}. If true, use the estimator smoothing algorithm I implemented some months ago with ian's help. If false, don't. Simulations are unclear.

Unstable build 60230 is in CVS and the snapshots are being updated. This build reverts the estimator implementation to its state before the Bucket Store refactoring. According to simulations, there is a serious problem with Bucket Store, but I don't know what exactly, as it does not seem to produce any obvious error messages; I may be mistaken in that, but in any case, Bucket Store is not ready. Simulations with (Float)Bucket Store produce significantly lower success ratios. So I have reverted it. The previous code works, despite being, by popular belief, ugly. I suggest that major routing changes (which can be - for example changes to the estimators) in future should be tested using the simulator code, at least before being merged to stable.

Unstable build 60226 is in CVS. The snapshots are being updated.
  • Lots of changes to the simulator code, detailed in CVS log message
  • Turned off SBKE smoothing, may be a problem

Unstable build 60220 is in CVS. The snapshots are updating.
  • Fix some NPEs
  • Try to catch cause of one in Mux Connection Handler (link ID was null for some reason..)
  • Fixes, synchronization, logging, trying to find bug in Simple Binary Running Average/EdgeKludgingBinaryRunningAverage: totalZeros was negative in a report on support, try to recover from this, and record more logging to find out why.
  • Logging to catch infinite loop in Queue Manager if it happens again.
  • Logging, delete old commented-out code.

Unstable build 60218 is now available.
  • Fixed an infinite loop in Queue Manager, related to invalid MRIs. Don't send invalid MRIs. Minor fix in Queue Manager for another bug that could have caused an infinite loop.
  • Improved start-freenet.sh. It was setting LD_ASSUME_KERNEL only on redhat, suse and gentoo.

Unstable build 60216 is in CVS. The snapshots will be updated soon.
  • Transfer failures have been fatal for a while, and this was causing deadlocks (freeze-ups anyway... no locking involved, but it was waiting for a notification that never came) in code e.g. Internal Client that assumed that there would be retries. This may also have affected FCP. One definite effect was that Fproxy would block threads forever sometimes. I saw this before, before queueing, but couldn't reproduce it. Now I have, and I think it is fixed.