Next Generation POSIX Threads

Posted Sep 20, 2002 18:05 UTC (Fri) by Nero (guest, #2085)
In reply to: Next Generation POSIX Threads by Shewmaker
Parent article: Native POSIX Thread Library 0.1 released

If you have a look at the discussion that followed the original post (you can find it on groups.google.com), this question was asked, Robert Love replied with something along the lines of "The glibc people never intended for NGPT to replace the old pthread library."

[though, I can't think why not..]

to post comments

Next Generation POSIX Threads

Posted Sep 21, 2002 10:33 UTC (Sat) by Peter (guest, #1127) [Link] (2 responses)

"The glibc people never intended for NGPT to replace the old pthread library."

[though, I can't think why not..]

My theory: NIH. glibc, being a GNU project, requires copyright assignments; perhaps the NGPT people were less than forthcoming at giving them.

Alternate theory: as Uli notes in his announcement, they concluded that the complexity of MxN threading does not pay off in performance for realistic workloads, compared to simply optimising the kernel to handle large numbers of truly lightweight threads. MxN was all anyone talked about a few years ago, but more than one person in this thread (no pun intended) on linux-kernel has said 1x1 makes more sense. One such person is Larry McVoy, whom I'll trust on scalability issues any day. (Well, until he disagrees with Dave Miller, like on the numa clusters thing - hard to know which horse to bet on there!)

Speaking of GNU NIH, does anyone know how does the gnutls project plans to differentiate themselves from openssl?

gnutls

Posted Sep 26, 2002 4:24 UTC (Thu) by joey (guest, #328) [Link]

Having a license that lets you link it to GPL'd code and distibrute the result is a good start..

Next Generation POSIX Threads

Posted Sep 26, 2002 12:34 UTC (Thu) by da4089 (subscriber, #1195) [Link]

re: gnutls ...

imo, the best way to distinguish gnutls from openssl is to make the programming api clear, simple, consistent, documented, etc.

openssl is a bit of a mess ...