Why is http://github.com/freenet/fred-official seven hours older (and
a Node version string of #1278 build01278-68-g3f9a3) than
http://github.com/freenet/fred-official/tree/build01280-real ?

On Fri, 1 Oct 2010 02:20:03 +0100, Matthew Toseland wrote:
> Freenet 0.7.5 build 1280 is now available, and will be mandatory on
> Saturday, please upgrade!
> Build 1278:
> - Fix a critical bug causing us to accept too many requests because
> of unlocking too early when turtling a request. This may be a
> significant factor in recent backoff problems.
> Build 1279:
> - Various bugfixes to the block transfer code, fix some stalling
> bugs, handle aborts (e.g. triggered by the upstream receive) better,
> synchronization fixes, don't abort twice, don't send abort message
> twice, but do send it even when the downstream node aborts by sending
> us one.
> - Prepare for getting rid of unnecessary block-level retransmissions
> (we have retransmission at the message layer), which waste a lot of
> bandwidth and may cause self-perpetuating big problems. E.g. don't
> timeout until some time after we have sent all the blocks in the
> transfer, rather than relying on frequent missing packet notification
> messages.
> Build 1280:
> - Fix a bug in 1279 causing block sends to stall when a node
> disconnects or something similar happens (the bug existed before but
> it didn't cause stalls until recently; it might affect some other
> places in the code though).
> Apart from 1280, the block transfer changes are in code that was
> written many years ago, but the problems may cause fairly significant
> problems (e.g. some block transfers use a *lot* of retransmits, and
> this tends to make the situation worse and worse). Some of the block
> transfer changes were developed while debugging the
> new-load-management branch, some were dealt with after testing the
> prerelease code. 1280 will include more low level code merged from
> the branch.
> Even if all the problems go away, I will continue with the new load
> management branch because the current architecture is fundamentally
> broken and vulnerable.
> THANKS! Please report bugs!
Support mailing list
Unsubscribe at http://emu.freenetproject.org/cgi-bin/mailman/listinfo/support
Or mailto:support-requ...@freenetproject.org?subject=unsubscribe

Reply via email to