TheOtherPhil
03-16-2003, 10:39 AM
Login: acidblood Name: Décio Luiz Gazzoni Filho
Directory: /home2/acidblood Shell: /bin/tcsh
Last login Sat Mar 15 21:36 (CST) on ttyp1
No Mail.
Plan:
:: 16-Mar-2003 04:07 GMT (Sunday) ::
Andreas Beckmann and I have traced another bug on the new SS 2-pipe core for
x86 processors (currently present on the pre-release 2.9003.481/481a clients
only), in addition to the reentrancy issues already known for some time. From
http://n0cgi.distributed.net/bugs/show_bug.cgi?id=3187:
``Aside from the reentrancy issue, this core has a problem whereby signal
handlers on Unix systems can overwrite the core's local variables, rendering
the results incorrect. A fix brings its performance down to ca. 94% of
DG 2-pipe, and thus the core is being removed from the client until the
author provides a reentrant and signal handler safe version with improved
performance.''
This bug affects uniprocessor machines as well. Since we have no guarantees
about the correctness of results produced by this core, every result returned
by an x86 machine running the 481/481a clients will have to be rechecked. Rest
assured that no stats are going to be rolled back because of this; however,
we do ask that any installations of this pre-release client on x86 machines
be removed. The buggy clients will soon be taken down from the pre-release
download page, and we will be refusing results from these clients after some
time.
Directory: /home2/acidblood Shell: /bin/tcsh
Last login Sat Mar 15 21:36 (CST) on ttyp1
No Mail.
Plan:
:: 16-Mar-2003 04:07 GMT (Sunday) ::
Andreas Beckmann and I have traced another bug on the new SS 2-pipe core for
x86 processors (currently present on the pre-release 2.9003.481/481a clients
only), in addition to the reentrancy issues already known for some time. From
http://n0cgi.distributed.net/bugs/show_bug.cgi?id=3187:
``Aside from the reentrancy issue, this core has a problem whereby signal
handlers on Unix systems can overwrite the core's local variables, rendering
the results incorrect. A fix brings its performance down to ca. 94% of
DG 2-pipe, and thus the core is being removed from the client until the
author provides a reentrant and signal handler safe version with improved
performance.''
This bug affects uniprocessor machines as well. Since we have no guarantees
about the correctness of results produced by this core, every result returned
by an x86 machine running the 481/481a clients will have to be rechecked. Rest
assured that no stats are going to be rolled back because of this; however,
we do ask that any installations of this pre-release client on x86 machines
be removed. The buggy clients will soon be taken down from the pre-release
download page, and we will be refusing results from these clients after some
time.