> -----Original Message----- > From: [email protected] [mailto:rsyslog- > [email protected]] On Behalf Of [email protected] > Sent: Monday, June 15, 2009 8:27 PM > To: rsyslog-users > Subject: Re: [rsyslog] high-performance, low-precision time API under linux? > > On Mon, 15 Jun 2009, Rainer Gerhards wrote: > > >> -----Original Message----- > >> From: [email protected] [mailto:rsyslog- > >> [email protected]] On Behalf Of Thomas Mieslinger > >> Sent: Monday, June 15, 2009 7:58 PM > >> To: rsyslog-users > >> Subject: Re: [rsyslog] high-performance, low-precision time API under > > linux? > >> > >> BTW: In the Informix DBMS the Software Architects decided to implement an > > OS > >> independent timer thread to work arround (very costly in the late 90s) > >> gettimeofday() calls. > >> > >> I've run the code on some Solaris 10 boxes, because they were known for > > their > >> slow gettimeofday() implementation > >> > >> S11 snv_105 Pentium 4 3GHz > >> time ./a.out 100000 > >> real 0m0.087s > >> user 0m0.082s > >> sys 0m0.004s > >> > >> S10 Generic_138889-03 Athlon 2,2GHz > >> time ./a.out 100000 > >> real 0m0.028s > >> user 0m0.026s > >> sys 0m0.002s > >> > >> ok, lets try this on an older boxes... > >> suse70 2.2.21 libc-2.1.3-141 2xP3 800 > >> time ./a.out 100000 > >> > >> real 0m0.100s > >> user 0m0.080s > >> sys 0m0.020s > >> > >> So, what are you using as a development box? > > > > I should have mentioned that I am typically running inside a virtualized > > environment as I routinely need to run three to four machines in parallel > for > > testing. > > ahh, virtualized boxes are especially bad for gettimeofday(). > which virtualization are you using?
This one is using vmware, but that doesn't really matter. Even on a real box, as you said, the toll to pay for gettimeofday() is high compared to the rest of the workload. So I actually do not care how much the absolute time is but rather how it goes into the overall mix. The context switch itself is expensive enough, so this makes everything that goes to kernel space rather expensive. As I had written earlier, I will probably give up the approach of a generalized timeout handler in favor of many targeted implementations. It would have been nice to have this callout-tablish facility, but it looks like it is too expensive to implement. Rainer > > David Lang > > > If anyone is up for hardware contributions, just let me know ;) > > > > Rainer > >> _______________________________________________ > >> rsyslog mailing list > >> http://lists.adiscon.net/mailman/listinfo/rsyslog > >> http://www.rsyslog.com > > _______________________________________________ > > rsyslog mailing list > > http://lists.adiscon.net/mailman/listinfo/rsyslog > > http://www.rsyslog.com > > > _______________________________________________ > rsyslog mailing list > http://lists.adiscon.net/mailman/listinfo/rsyslog > http://www.rsyslog.com _______________________________________________ rsyslog mailing list http://lists.adiscon.net/mailman/listinfo/rsyslog http://www.rsyslog.com
