Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
Thank you! Since it's tunable at runtime I just tested it, and -- sure enough -- no negative ping times. Ironically, it was the kernel that selected the ACPI timer, scoring it higher than the timestamp counter as a clock source. Perhaps code should be added to ensure that the timer is not chosen if it rolls over in less than a second, since this clearly leads to imprecision and missed rollovers. --Brett Glass At 11:04 PM 9/12/2011, Adam Vande More wrote: it's a runtime tunable so /etc/sysctl.conf ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
At 09:16 AM 9/13/2011, Dan Nelson wrote: It doesn't roll over in less than a second; it rolls over in 16777215 / 3579545 = 4.6 seconds. Your negative time delta problem isn't due to rollover. If that's indeed the case, the kernel must be doing the math wrong. I wonder how many other systems this is affecting. --Brett Glass ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
[ ...combining two emails... ] On Sep 13, 2011, at 9:49 AM, Brett Glass wrote: If that's indeed the case, the kernel must be doing the math wrong. While there have undoubtedly have been kernel bugs with timekeeping (and there may be more still present), it's not uncommon for hardware issues to cause one or more of the available time sources to be broken. On a good day, the clock source is broken obviously enough that the kernel notices it during testing during boot and gives it a negative quality score. Other times, the clock becomes broken only after the box suspends and resumes from an ACPI S# state, or does frequency changes for power/thermal management, etc. Ironically, it was the kernel that selected the ACPI timer, scoring it higher than the timestamp counter as a clock source. Perhaps code should be added to ensure that the timer is not chosen if it rolls over in less than a second, since this clearly leads to imprecision and missed rollovers. The kernel attempts to notice problems when it probes for the various clocks during boot (ie, dev/acpica/acpi_hpet.c, dev/acpica/acpi_timer.c, etc); for ACPI, see acpi_timer_probe() acpi_timer_test(). Regards, -- -Chuck ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
RE: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
More information regarding the odd behavior I'm seeing. Turns out that packets do not even need to leave the machine for it to report large negative ping times, on the order of more than half a second. (See below.) Clearly something is odd about timekeeping in this system (SiS motherboard chipset, PII-generation Celeron but still effectively a 686) which was not a problem when it was running FreeBSD 4.11-RELEASE (as it was before). What's more, it appears that the negative ping times being shown for pings of localhost are off by about -687 ms, consistently. Any ideas? I am wondering if perhaps some recent change to the kernel assumed that one would always have a faster CPU than the old Celeron this machine is running, and that there is a race condition or an error in the kernel code. --Brett Glass # ping localhost PING localhost (127.0.0.1): 56 data bytes 64 bytes from 127.0.0.1: icmp_seq=0 ttl=64 time=-0.148 ms 64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=-0.151 ms 64 bytes from 127.0.0.1: icmp_seq=2 ttl=64 time=-686.111 ms 64 bytes from 127.0.0.1: icmp_seq=3 ttl=64 time=-0.180 ms 64 bytes from 127.0.0.1: icmp_seq=4 ttl=64 time=0.110 ms 64 bytes from 127.0.0.1: icmp_seq=5 ttl=64 time=686.351 ms 64 bytes from 127.0.0.1: icmp_seq=6 ttl=64 time=-686.376 ms 64 bytes from 127.0.0.1: icmp_seq=7 ttl=64 time=0.121 ms 64 bytes from 127.0.0.1: icmp_seq=8 ttl=64 time=-686.402 ms 64 bytes from 127.0.0.1: icmp_seq=9 ttl=64 time=-686.105 ms 64 bytes from 127.0.0.1: icmp_seq=10 ttl=64 time=686.623 ms 64 bytes from 127.0.0.1: icmp_seq=11 ttl=64 time=0.107 ms 64 bytes from 127.0.0.1: icmp_seq=12 ttl=64 time=0.119 ms 64 bytes from 127.0.0.1: icmp_seq=13 ttl=64 time=0.418 ms 64 bytes from 127.0.0.1: icmp_seq=14 ttl=64 time=0.401 ms 64 bytes from 127.0.0.1: icmp_seq=15 ttl=64 time=-0.169 ms 64 bytes from 127.0.0.1: icmp_seq=16 ttl=64 time=0.113 ms 64 bytes from 127.0.0.1: icmp_seq=17 ttl=64 time=0.401 ms 64 bytes from 127.0.0.1: icmp_seq=18 ttl=64 time=-686.117 ms 64 bytes from 127.0.0.1: icmp_seq=19 ttl=64 time=0.115 ms 64 bytes from 127.0.0.1: icmp_seq=20 ttl=64 time=0.111 ms ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
On Sep 12, 2011, at 4:50 PM, Brett Glass wrote: What's more, it appears that the negative ping times being shown for pings of localhost are off by about -687 ms, consistently. Any ideas? Your system's timekeeping appears to be busted. Are you running ntpd with tinker step 0.0 or some home-grown mechanism which might be forcibly stepping the clock rather than skewing it, by any chance? Anyway, the output of: sysctl -a kern.timecounter ...is likely to be informative. Try switching to another clock type, especially ACPI-safe if it hasn't been chosen by default. Your CPU is probably too old to have a power-state invariant TSC, but if you disable SpeedStep, powerd and similar which might change the processor frequency, TSC might work OK also. Regards, -- -Chuck ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
I just put FreeBSD 8.1 up on an old (but good) 500 MHz Celeron with half a gig of RAM. Interfaces are classic xl (3Com) and dc (DEC tulip). Works quite nicely except for one quirk: ping times that ought to be positive (no more than 200 ms worst case) are coming out negative! Can't figure out what might be causing this. dmesg output is as follows: If you are just upgrading now, why not use 9 BETA? I think that your older machine will be much happier -- the new timer code in 9 has a bunch of bugfixes, allows for a wider choice of alternative timers, in case some are broken, and places lighter loads on the system, by allowing some (formerly periodic) timer use to be deferred. And then there is the host of other improvements... b. ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
At 06:15 PM 9/12/2011, Chuck Swiger wrote: Your system's timekeeping appears to be busted. Are you running ntpd with tinker step 0.0 or some home-grown mechanism which might be forcibly stepping the clock rather than skewing it, by any chance? Nothing like that. Anyway, the output of: sysctl -a kern.timecounter ...is likely to be informative. Here it is: kern.timecounter.tick: 1 kern.timecounter.choice: TSC(800) ACPI-safe(850) i8254(0) dummy(-100) kern.timecounter.hardware: ACPI-safe kern.timecounter.stepwarnings: 0 kern.timecounter.tc.i8254.mask: 4294967295 kern.timecounter.tc.i8254.counter: 5754 kern.timecounter.tc.i8254.frequency: 1193182 kern.timecounter.tc.i8254.quality: 0 kern.timecounter.tc.ACPI-safe.mask: 16777215 kern.timecounter.tc.ACPI-safe.counter: 7967112 kern.timecounter.tc.ACPI-safe.frequency: 3579545 kern.timecounter.tc.ACPI-safe.quality: 850 kern.timecounter.tc.TSC.mask: 4294967295 kern.timecounter.tc.TSC.counter: 4058536290 kern.timecounter.tc.TSC.frequency: 501141177 kern.timecounter.tc.TSC.quality: 800 kern.timecounter.invariant_tsc: 0 This is very instructive. I didn't know that FreeBSD used the Pentium internal timestamp counter for anything but profiling. I am noticing here that the mask (which I assume is the maximum value just before a rollover) for the ACPI-safe timer is very small. Maybe it's rolling over very frequently and/or the system is missing some of the rollovers. This would cause it to calculate negative times, of course. Try switching to another clock type, especially ACPI-safe if it hasn't been chosen by default. No docs on how to do this. Is this done by, for example, setting kern.timecounter.hardware=TSC in loader.conf? Your CPU is probably too old to have a power-state invariant TSC, but if you disable SpeedStep, powerd and similar which might change the processor frequency, TSC might work OK also. I've already turned off all power saving mechanisms listed in the BIOs setup, including clock speed modulation. So, the TSC ought to be pretty stable. At least it's worth a shot. --Brett Glass ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
At 06:54 PM 9/12/2011, b. f. wrote: If you are just upgrading now, why not use 9 BETA? Production machine. Also, whenever we create a new production box, we normally pick the release (not beta; we need to be able to do binary upgrades and this is only supported from one release to another) with the EOL that's the farthest out. We'll retire the hardware before we will run non-release code on a production box. --Brett Glass ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: Negative ping times with FreeBSD 8.1-RELEASE on older Celeron system
On Mon, Sep 12, 2011 at 10:30 PM, Brett Glass br...@lariat.net wrote: At 06:15 PM 9/12/2011, Chuck Swiger wrote: sysctl -a kern.timecounter No docs on how to do this. Is this done by, for example, setting kern.timecounter.hardware=TSC in loader.conf? it's a runtime tunable so /etc/sysctl.conf and the sysctl -a kern.timecounter doesn't need a -a -- Adam Vande More ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
