[2657] in Kerberos_V5_Development
Re: Kerberos NT source code
daemon@ATHENA.MIT.EDU (Ken Raeburn)
Fri Oct 24 18:12:03 1997
To: krbdev@MIT.EDU
From: Ken Raeburn <raeburn@cygnus.com>
Date: 24 Oct 1997 18:11:54 -0400
In-Reply-To: "Theodore Y. Ts'o"'s message of "Thu, 23 Oct 1997 14:21:21 -0400"
"Theodore Y. Ts'o" <tytso@MIT.EDU> writes:
> I still think there are many problems for which it is technically
> possible to use threads, but it's not always clear that threads are the
> best way to solve that problem. For example, I recall Chris Provenzano
> once claiming that you might use threads as a way of structuring a
> program which computed the answer to a math problem. I don't remember
> enough of the details, but it was enough to convince me that for there
> were certainly folks for which a threads package was a hammer, and
> everything ended up looking like a nail.....
It does seem like overkill sometimes, but there are times it can be
useful. For example, for a large-scale calculation, operating on
large bodies of data, (a) perhaps you have multiple processors that
can work on mostly-independently parts of the problem, and (b) perhaps
the kernel can switch threads while you're waiting for data in some
part of your address space to get swapped in.
Shared memory could be used to address these problems too, though, and
I'm not familiar enough with implementations of either to know the
tradeoffs. But then you've got problems like not sharing file
descriptors...
