TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu (09/10/90)
Can Unix sleep in terms of mili or mirco second? I am aware that sleep() can only sleep in terms of second. Please specify the Unix Dialect when u reply. Thanks. - Beng Hang (email: taybengh@nusdiscs)
tli@phakt.usc.edu (Tony Li) (09/10/90)
In article <24437@adm.BRL.MIL> TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes:
Can Unix sleep in terms of mili or mirco second? I am aware that
sleep() can only sleep in terms of second. Please specify the Unix Dialect
when u reply. Thanks.
BSD derived systems provide usleep(3) which does what you want.
--
Tony Li - USC Computer Science Department
Internet: tli@usc.edu Uucp: usc!tli
Thus spake the master programmer: "A well written program is its own
heaven; a poorly-written program its own hell."2307GILLV%MUCSD.BITNET@cunyvm.cuny.edu (09/10/90)
The nap () system all in Xenix 2.3.2 can sleep in terms of
milli seconds. Caveat Emptor This function is linked with
the system clock, which is, in most has a granularity of
tens of milli seconds.
-I compute, Therefore I VAX
-Dicky Gill at MUCSD.EDUtomw@orac.esd.sgi.com (Tom Weinstein) (09/10/90)
In article <11899@chaph.usc.edu>, tli@phakt.usc.edu (Tony Li) writes: > In article <24437@adm.BRL.MIL> TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that > sleep() can only sleep in terms of second. Please specify the Unix Dialect > when u reply. Thanks. > > BSD derived systems provide usleep(3) which does what you want. Yuck. Ick. Ptui. usleep is really pretty icky because it uses itimers, and the system call overhead (4 of them!) can seriously destroy the accuracy of your timing. A much better idea is to use something like this: { static struct timeval tv; tv.tv_sec = usecs / 1000000; tv.tv_usec = usecs % 1000000; select(0, 0, 0, 0, &tv); } This uses only one system call instead of the 4 of usleep, and it doesn't use itimers so it won't screw up programs that use them for other things. -- Tom Weinstein Silicon Graphics, Inc., Entry Systems Division, Window Systems tomw@orac.esd.sgi.com Any opinions expressed above are mine, not sgi's.
boissier@irisa.fr (franck boissiere) (09/10/90)
From article <11899@chaph.usc.edu>, by tli@phakt.usc.edu (Tony Li): > > BSD derived systems provide usleep(3) which does what you want. > On some manuals for usleep it is mentionned that you may use select with no descriptor to use to do the trick. -- Franck BOISSIERE boissier@irisa.irisa.fr Prototyping Lab Manager boissier@ccettix.UUCP C.C.E.T.T. B.P. 59 boissier%irisa.irisa.fr@uunet.uu.net 35512 CESSON SEVIGNE CEDEX FRANCE
nanook@rwing.UUCP (Robert Dinse) (09/11/90)
In article <24437@adm.BRL.MIL>, TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that > sleep() can only sleep in terms of second. Please specify the Unix Dialect > when u reply. Thanks. > > - Beng Hang (email: taybengh@nusdiscs) Two provisions I'm familiar with: Under Xenix there is a nap() system call. It takes an argument in milliseconds, but the granularity depends on the system tick rate. To use it the -lx (to link the xenix library) is required. Under Sun OS, usleep(), which takes an argument in microseconds is available. I don't remember having to do anything special to use it but it's been awhile.
yukngo@obelix.gaul.csd.uwo.ca (Cheung Yukngo) (09/11/90)
I remember reading somewhere that the select() function can be used to do this. Anybody knows exactly how it is done? Thanks.
james@dlss2.UUCP (James Cummings) (09/11/90)
In article <24437@adm.BRL.MIL> TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that >sleep() can only sleep in terms of second. Please specify the Unix Dialect >when u reply. Thanks. How 'bout: main() { sleep_less_than_sec(25); exit(0); } sleep_less_than_sec(x) int x; { int i; for (i = 0;i <= x;i++) ; } I'd think that would give you approx. 25 clock cycles, dependent on the machine, the load on the machine, etc. -- =+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+ |Disclaimer: | James Cummings | | You can't blame me! | UUCP: | | I'm ignorant! | ..swblat!{texbell!texnet.. | |+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+| swgate!dlss1..}!dlss2!james | |Send flames to: | NET: | | sowc@devnull.com | jc@smunews | | | | =+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
mutchler@zule.EBay.Sun (Dan Mutchler) (09/12/90)
In article <84@dlss2.UUCP> james@dlss2.UUCP (James Cummings) writes: In article <24437@adm.BRL.MIL> TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that >sleep() can only sleep in terms of second. Please specify the Unix Dialect >when u reply. Thanks. How 'bout: main() { sleep_less_than_sec(25); exit(0); } sleep_less_than_sec(x) int x; { int i; for (i = 0;i <= x;i++) ; } I'd think that would give you approx. 25 clock cycles, dependent on the machine, the load on the machine, etc. Unless a context switch occurs, which is quite possible. That may still be okay if times much greater than 25 clock cycles are okay. That is you need a sleep that sleeps at least x milliseconds. Looping is in general a bad delay because different machines require different CPU time to perform the loop. Intelligent use of macros can help this, but unless it is in the kernel where a context switch cannot occur you can only guareentee that you will not return before the required delay. -- Dan Mutchler | ARPA/Internet: mutchler@EBay.Sun.COM Sun Federal System Engineer | UUCP: ...!sun!mutchler -------------------------------------------------------------------------- Work hard. Play hard. Eat hard. Sleep hard. Grow big. Wear glasses, If you need them. -The Webb Wilder Credo
morrell@hpcuhb.HP.COM (Michael Morrell) (09/13/90)
/comp.unix.internals/james@dlss2.UUCP (James Cummings)/5:11 am Sep 11, 1990/ >> Can Unix sleep in terms of mili or mirco second? > sleep_less_than_sec(x) > int x; > { > int i; > > for (i = 0;i <= x;i++) > ; > } > I'd think that would give you approx. 25 clock cycles, dependent on >the machine, the load on the machine, etc. ---------- This is not a very good idea. Any decent optimizer will simply optimize away your code loop and you won't get any delay (although if you declare "i" as volatile, it shouldn't be removed, but it's still very dependent on the machine and its current load). Michael
cpcahil@virtech.uucp (Conor P. Cahill) (09/13/90)
In article <MUTCHLER.90Sep11140510@zule.EBay.Sun> mutchler@zule.EBay.Sun (Dan Mutchler) writes: >In article <84@dlss2.UUCP> james@dlss2.UUCP (James Cummings) writes: > >Looping is in general a bad delay because different machines require >different CPU time to perform the loop. Intelligent use of macros can >help this, but unless it is in the kernel where a context switch >cannot occur you can only guareentee that you will not return before >the required delay. Plus it has a bad little side effect of eating up gobs and gobs of cpu time. -- Conor P. Cahill (703)430-9247 Virtual Technologies, Inc., uunet!virtech!cpcahil 46030 Manekin Plaza, Suite 160 Sterling, VA 22170
sxm@bebop.Philips.Com (Sandeep Mehta) (09/13/90)
james@dlss2.UUCP (James Cummings) writes: TAYBENGH%NUSDISCS.BITNET@cunyvm.cuny.edu writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that >sleep() can only sleep in terms of second. Please specify the Unix Dialect >when u reply. Thanks. How 'bout: sleep_less_than_sec(x) int x; { int i; for (i = 0;i <= x;i++) ; } For many (obvious) reasons a dummy loop is not the best idea. Besides you may not have a real time clock in your UNIX box, e.g., Sun 3's (at least the ones I use) use the Intersil 7170 "real-time" clock in 100 Hz periodic mode, giving a resolution of only 10 milliseconds. Using the clock(3C) function yields CPU time used only up to a 16.667 milliseconds accuracy. Sun 3 (SunOS 4.0) getrusage() assumes tick intervals of 20msec, for Sun 4's 10msec. I know that struct timerval' has a "usec" field but of course the value is hardware dependent. You might write your own little assembler routine to use a spare counter or something (not advisable :-)) which will suffer from all the restrictions of the process that uses it. I know that some folks @ Berkeley have a real-time timer facility for Sun 3's under SunOS and it's (h'ware + s'ware) pretty cheap. But then again, is UNIX/SunOS real-time ? Don't know if I answered the question either ? sandeep -- sxm@philabs.Philips.Com "Jazz is Happiness" - Le Sony'Ra
johnv@metaware.metaware.com (John Vinopal) (09/14/90)
I seem to recall mutchler@zule.EBay.Sun (Dan Mutchler) saying: > james@dlss2.UUCP (James Cummings) writes: > In article <24437@adm.BRL.MIL> TAYBENGH%NUSDISCS.BITNET writes: > > Can Unix sleep in terms of mili or mirco second? I am aware that > >sleep() can only sleep in terms of second. Please specify the Unix Dialect > >when u reply. Thanks. > > main() > { > sleep_less_than_sec(25); > exit(0); > } > > sleep_less_than_sec(x) > int x; > { > int i; > > for (i = 0;i <= x;i++) > ; > } > > I'd think that would give you approx. 25 clock cycles, dependent on > the machine, the load on the machine, etc. You assume that you have a remarkably dumb compiler. How is that code any different to an optimizing compiler than say: #define DEBUG 0 foobar(int x) { int i; for (i = 0; i <= x; i++) if ( DEBUG ) { crash_n_burn(); } } You wouldn't want THAT code stuck in your executable now would you? Both come out to be NULL and both should be removed! Optimize, optimize, optimize! -- The Wailer at the Gates of Dawn | johnv@metaware.com | #include <vomit.h> | uunet!metaware!johnv | void puke(struct dinner *p) { free(p); } | banshee@ucsc{b,f}.UCSC.EDU | 2,3,5,7,13,17,19,31,61,89,107,127,521,607....| banshee@ucsc{b,f}.BITNET |
vs_subbu@apollo.HP.COM (V. S. Subrahmanyam) (09/14/90)
In article <37960001@hpcuhb.HP.COM> morrell@hpcuhb.HP.COM (Michael Morrell) writes: >/comp.unix.internals/james@dlss2.UUCP (James Cummings)/5:11 am Sep 11, 1990/ >>> Can Unix sleep in terms of mili or mirco second? > >> sleep_less_than_sec(x) >> int x; >> } >---------- > >This is not a very good idea. Any decent optimizer will simply optimize away >your code loop and you won't get any delay (although if you declare "i" as >volatile, it shouldn't be removed, but it's still very dependent on the >machine and its current load). > > Michael You can try using "blocking I/O" : 1. Select a tty line that will not be used by any other person. 2. Open the tty line for read 3. ioctl that tty line , with vtime field in the c_cc array to the desired sleeping period. 4. Read the fd associated with the tty line. Since nobody is writing any data to that fd, and since we are using blocking I/O, the read call will block for the duration specified in the vtime field. Subbu.
rowe@cme.nist.gov (Walter Rowe) (09/14/90)
To quote a SunOS man page:
======================================================================
NAME
usleep - suspend execution for interval in microseconds
SYNOPSIS
usleep(useconds)
unsigned useconds;
DESCRIPTION
Suspend the current process for the number of microseconds
specified by the argument. The actual suspension time may
be an arbitrary amount longer because of other activity in
the system, or because of the time spent in processing the
call.
The routine is implemented by setting an interval timer and
pausing until it occurs. The previous state of this timer
is saved and restored. If the sleep time exceeds the time
to the expiration of the previous timer, the process sleeps
only until the signal would have occurred, and the signal is
sent a short time later.
This routine is implemented using setitimer() (see getiti-
mer(2)); it requires eight system calls each time it is
invoked. A similar but less compatible function can be
obtained with a single select(2); it would not restart after
signals, but would not interfere with other uses of setiti-
mer.
SEE ALSO
getitimer(2), sigpause(2V), alarm(3V), sleep(3V), ualarm(3)
======================================================================
wpr
---
Walter P. Rowe ARPA: rowe@cme.nist.gov
System Administrator, Robot Systems Division UUCP: uunet!cme-durer!rowe
National Institute of Standards and Technology LIVE: (301) 975-3694sgf@cs.brown.edu (Sam Fulcomer) (09/18/90)
In article <YUKNGO.90Sep11095526@obelix.gaul.csd.uwo.ca> yukngo@obelix.gaul.csd.uwo.ca (Cheung Yukngo) writes: >I remember reading somewhere that the select() function can be used to Well, there's an ugly little BSD function called usleep(3) which uses the itimer stuff. You can also use select(2) as in: select(0,0,0,0,&sleepy_time); (handle signals however you'd like...) Many Un*xen will have select and perhaps the itimer stuff. Fewer will have usleep (SYSV jocks have an easier time keeping up with the BSD system calls...). Bear in mind that this is all pretty grainy generic stuff. If you want "real time" control you should look at something marketed for it. -s
lef@dogwood.atl.ga.us (Lawrence E. Freil) (09/21/90)
For System V.3 or greater why not use the following code:
/**************************************************************************
* Function Name : nap
* Argument Description : time to sleep in milliseconds
* Global References : (a % by the name indicates an alteration occurs)
* Author : Lawrence Freil
* Function Description :
* This routine uses the poll system call to delay a specified number
* of milliseconds. Poll called without any file descriptors but with
* a timeout specified works for this purpose.
*
**************************************************************************/
int
nap(int timeout)
{
struct pollfd foo[1];
int status;
status = poll(foo, 0L, timeout);
return status;
}
You can also do the same thing for berkley with the select call (no files,
just the timeout.)
Lawrence Freil Usenet/DDN:lef@dogwood.atl.ga.us
National Science Center Foundation Phone:(404)-828-8459
P.O. Box 1648
Augusta, Ga 30903
--
Lawrence Freil Usenet/DDN:lef@dogwood.atl.ga.us
National Science Center Foundation Phone:(404)-828-8459
P.O. Box 1648
Augusta, Ga 30903