On Sun, Apr 11, 2004 at 10:30:39PM +0200, martin f krafft wrote: > also sprach Henrique de Moraes Holschuh <[EMAIL PROTECTED]> [2004.04.09.1949 +0200]: > > It won't hurt. But nmi_watchdog is only for usermode and > > kernemode hangs. The NMI watchdog is useless against nasty bugs > > (hw or sw) that make the hardware unstable. > > Yes, I have just discovered that. > > (Note that my new thread is about a different machine. I can't try > the stuff you suggested until Wednesday, when I get back to Zurich. > > > I get about 101 NMIs per second on each CPU using nmi_watchdog=1. > > HZ=100 in this machine, I suppose... This is a not-really-SMP > > machine, with a single P4 HT processor. > > related, but not to the thread: what's that HZ stuff? I assumed it > to be something like Hertz, but I could be wrong. However, ever > since I switched to the 2.6 kernel, I get the "Wrong HZ, was 67; > should be 100" or something like that messages occasionally. >
HZ is the number of jiffies in a second in the kernel (a value for kernel time). In 2.4 under i386 platforms it is 100, it 2.6 under i386 it was changed to 1000 and a new value called USER_HZ was introduced for converting kernel time to usermode time for programs that use hardcoded values, although jiffies still slips through in several proc interfaces. > > The software watchdog will reboot your machine (and AFAIK it might > > very well be using NMIs to do it, too). If you can use a chipset > > watchdog, however, that's much better (e.g. the TCO timers in most > > Intel systems, and _especially_ the IPMI watchdog in servers with > > a baseboard controller worth something). > > I'll think about it. How much are these hardware watchdogs, and > which one would you recommend? > > -- > Please do not CC me when replying to lists; I read them! > > .''`. martin f. krafft <[EMAIL PROTECTED]> > : :' : proud Debian developer, admin, and user > `. `'` > `- Debian - when you have better things to do than fixing a system > > Invalid/expired PGP subkeys? Use subkeys.pgp.net as keyserver! -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]