Author Topic: Stable Build 5090; weniger RNFs  (Read 1923 times)

0 Members and 1 Guest are viewing this topic.

Offline Nemo

  • Global Moderator
  • Elite
  • *****
  • Posts: 1303
  • Karma: +27/-0
    • View Profile
Stable Build 5090; weniger RNFs
« on: August 07, 2004, 11:06:49 AM »
Quelle: http://article.gmane.org/gmane.network.freenet.devel/14974

Quote

From: Toad <toad@...>
Subject: Stable build 5090
Newsgroups: gmane.network.freenet.devel, gmane.network.freenet.support
Date: Fri, 06 Aug 2004 19:49:50 +0100

Stable build 5090 is now available. The snapshots have been updated.
Please upgrade ASAP, unless you are using the unstable branch. You can
do this on Windows by using the update option on the start menu. On
Linux, OS/X etc, you can update by using the update.sh script (and
restart the node). You can alternatively shut down your node and
download http://freenetproject.org/snapshots/freenet-latest.jar over
your existing freenet.jar, then restart your node. If you are currently
on a build prior to 5089, then you will also need to reseed.

CHANGELOG:
Routing:
* Fixed a major 4 month old bug that was causing Route Not Found errors
  to happen more quickly and more often than they should. This probably
  did other bad things too.
* Most of the global estimators were wiped on startup resulting in
  less accurate routing and possibly rate limiting.
* When reseeding, don't include nodes so old that we won't talk to them.
Rate limiting:
* Improvements to globalQuota (rate limiting's "how many total requests
  should we aim for?" function): Average globalQuota, not the load put
  into it (otherwise we have inertia in the wrong place which makes
  oscillations worse), calculate the maximum number of requests we can
  handle based on bandwidth usage and probability of success and don't
  let globalQuota get any higher than that (should smooth rate
  limiting).
* Use a longer time period for the various averages and counters used
  by rate limiting. This should reduce the effects of long-lived
  requests on rate limiting.
* Add a new form of high-level bandwidth limiting, which predicts the
  expected outbound bandwidth usage based on current request rate and
  probability of an incoming request causing a transfer. Should smooth
  out rate limiting.
Padding:
* Don't send RequestInterval's on FNP messages. Don't parse them if we
  receive them. They are unnecessary as all nodes on the new stable
  network understand the new (small, fixed size) MRI messages. We can
  therefore reduce the padding interval from 200 bytes to 160 bytes.
UI changes:
* Two new values on the General infolet.
* Minor HTML fixes on the Open Connections page.
Misc:
* Fix runaway receiving transfers count on the Open Connections page.
  This was a reporting bug, we were not really receiving hundreds (in
  some cases) of files.
* If a config file parameter is misspelled or otherwise corrupt, we used
  to throw a ClassCastException and refuse to start. This is a fairly
  common support problem. Now we provide a much more informative message
  to stderr and the log file, and attempt to use the default value for
  the param.
* Accept multipliers (kKmMgG) on integers and shorts (different kinds of
  whole number options) in the config, as well as on longs. Practical
  upshot: You can now do "outputBandwidthLimit=10k".
* Don't start the node if another node is running. This is enforced
  using locking a lock file.
* Don't change the node file unless the address detected has actually
  changed, and add synchronization to prevent race in the address
  redetect function. Both of these fix the node renaming error messages
  sometimes seen shortly after startup.
* Get rid of WriteSelectorLoop.fixKeys, cause of the fixKeys error
  messages. Hopefully this is unnecessary. It was a workaround for
  perceived issues in NIO, which probably don't exist. :)
* Fix some NullPointerException's.
* Delete some useless code, including code for backwards compatibility
  with 5084. Some of the nastier bugs in 5085-5088 came from this code.
* Minor bug fixes (hopefully) in diagnostics.
* Bug fix relating to message construction, probably wouldn't cause
  problems in this build but did in 5088.
* Minor improvements to the Distribution Servlet: Use the right size
  bucket (slight efficiency improvement), support temp dir and distrib
  dir on different volumes.
* Logging improvements
* Update copyright notices on freenet/support/CPUInformation/*CPUInfo.
New diagnostic variables:
* outputBytesMRI (bytes used for sending low level MRI messages)
* outputBytesPaddingOverhead (bytes used only for padding messages out
  to a "safe" size)
* outputBytesTrailerChunksInsert (bytes used to send files for inserts)
* requestTransferTime (time taken by successful file transfers)
* requestCompletionTime (total time taken by successful requests)
--
Matthew J Toseland - toad@...
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.

_______________________________________________
Devl mailing list
Devl@...
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl


MfG,
Nemo.