This is the mail archive of the cygwin mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Hyperthreading Problem - suggestion


On Mon, Jan 10, 2005 at 09:53:55AM +0100, Volker Bandke wrote:
>I really want that "hyperthreading problem" with Cygwin resolved.  Does
>the following suggestion make any sense:
>
>(I assume cfg is in the US, if not, make the necessary changes)
>
>a: Someone (close by to cfg) demonstrates that the bug/problem/issue
>appears on his machine
>
>b: This someone determines what an identical replacement machine would
>cost now
>
>c: CFG (or the someone from a) sets up a paypal account for "HT Machine
>Donations"
>
>d: When the required amount is reached, the Someone from a: sends his
>machine to CFG and buys a replacement out of the paypal account
>
>I am willing to pay 100 USD into that account.  Knowing that my machine
>has cost about Euro 1000 I believe it should take a real short time to
>provide cfg with a machine to test / debug the hyperthreading issue .

http://www.cygwin.com/ml/cygwin/2004-09/msg00084.html

I was going to say that this was YA repeat observation, in keeping with
the need for people to send the same observations about hyperthreaded
behavior here.

The repeated observations are as follows:

  1) I just got a hang by doing X (where X is some variation of previously reported
     problems)!

  2) I just got this problem as reported in the mailing list thread here:
     *show a URL where someone is reporting hyperthread problems* (bonus points for
     showing URL with an entry from me talking about how I can't duplicate the
     problem)

  3) I can provide a system over the internet for use in debugging.

  4) I just noticed a hang on Windows 95!  This must mean that hyperthreading is
     not the problem!

Let me try to state this very clearly.

The last time I tried to duplicate the problem, I ran a test for three
days, tying up a machine that I use for other purposes.  I'd previously
tried other variations as well.  The system is an SMP system so there
should have been a good chance of duplicating the bug.  I'm not going to
try every variation of this problem when it shows up on the list.

Fixing this problem will be difficult.  Dealing with race conditions
like this always is difficult.  I don't want to do this over the
internet.  It would be too painful and there is no guarantee that I
could even do it at all.  It could involve machine reboots (I just had
to do one of these yesterday while debugging a non-thread race
condition).

No one has said that cygwin is bug free.  So, there is undoubtedly more
than one way to make it hang.  A hang that shows up on Windows 95 does
not necessarily translate to the same problem that people on
hyperthreading machines are seeing.  It may, but since the only Windows
9x type system that I have available for extended use is a vmware
emulation, it isn't likely that I'll be able to duplicate the problem in
Windows 9x.

I have seven machines in my office right now and don't really need another
one (I never thought I'd hear myself say that) but I will somehow make room
if I can get another hyperthreaded machine here.  If I do, I will dedicate
it to cygwin and cygwin testing and try to make sure that this problem never
crops up again.

Someone else has previously raised the issue of a paypal account and the
thread in the quoted URL above shows how to make donations.  However, I
am heartened to see that Volker is thinking like a scientist and has
noted a problem with the "get a new hyperthreaded system to cgf"
problem.  There is no guarantee that any *new* machine that I get will
evidence the problem.

Rather than do some complicated swap game on systems, however, I'd be
willing to chance that a system which is sufficiently similar to a
problem system will show the problem.  If/when I get enough money to
purchase one, I will check such a system out.  If I can't duplicate the
problem on that system, I'll be glad to then swap with someone else.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]