Allows modern NPTL threading
implementation to achieve better performance
and much closer POSIX compliance
than LinuxThreads.
New arguments and additional flags for clone(2).
New system calls:
exit_group(2),
tgkill(2),
get_thread_area(2),
set_thread_area(2),
set_tid_address(2).
New futex(2)
system call supports faster implementation of
mutexes and semaphores.
Signal-driven I/O possible on pipes (2.6.0).
See fcntl(2) manual page.
semtimedop(2) == semop() with timeout,
for System V semaphores (2.6.0)
For applications that make heavy use of semaphore operations
with timeouts (some database management systems).
Subsequently backported into 2.4.22.
Not specified in SUSv3; available on Solaris 9 and HP-UX 11.
Hard CPU affinity: sched_setaffinity(2) (2.6.0).
Used on multiprocessor system to
confine process to subset of the CPUs.
Can be used to maximize throughput and response
of a single process, by placing
all other processes on another CPU.
Can prevent performance cost of cache invalidation that occurs if
process times out on one CPU and is rescheduled on another.
Even without sched_setaffinity(2),
Linux kernel attempts to provide soft CPU affinity
to reduce performance chances of cache invalidation.
/proc/sys/kernel/core_pattern
specifies a template used to name all core dump files (2.6.0).
Core file name can include: name of executable file,
time of dump (seconds since Epoch), PID,
real UID/GID of process, number of terminating signal,
name of host system (sethostname(2)).