Hi, This is long but I wanted to provide as much information as possible.
There appears to be an interesting "feature" with NIS client support under FreeBSD 3.2-STABLE (as at Tuesday 24 August). This was a new installation onto a Dell Precision 410-MT (single 400MHz CPU, single SCSI disk, on-board fast ethernet) into an existing environment (the first FreeBSD host in this environment which has mainly Suns, a few Alpha and SGIs and an increasing number of Linux boxes). The machine is part of a group that does telecommunications and networking research - they already have Suns and Linux boxes. Essentially the install went smoothly direct from the 3.2 CDs, and subsequent post-installation configuration involved making a custom kernel, setting up NIS client support, configuring AMD, installing some key packages (e.g. "ssh") and installing and configuring an Efficient ATM card (using HARP included in the release). No real problems with any of this. NIS client support uses "ypbind" in broadcast mode - the NIS master server is a SunOS 4.1.3_U1B machine and three other NIS slave servers are all running Solaris 2.6. This environment has served the site well for a number of years and is VERY stable. We make extensive use of various NIS maps including the automounter for home directories. What appears to happen with the FreeBSD host is that after a reboot, it correctly binds to one of the NIS servers BUT whenever, for whatever reason, it is forced to re-bind it ends up never re-binding, creates something like 600kbit/s of broadcast traffic across our LAN and forces the "rpcbind/portmap" and "ypserv" process on the NIS servers to become CPU bound - it appears to also cause problems for some other broadcast-type applications such as DHCP. Interestingly, the broadcast traffic slowly decreases but probably only by say 100kbit/s in a period of more that 12 hours. Some info: FreeBSD host: ~~~~~~~~~~~~ % ypwhich ypwhich: can't yp_bind: reason: Domain not bound USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND root 147 44.7 1.5 1288 924 ?? Rs Tue08PM 394:11.05 ypbind root 132 1.6 0.9 820 576 ?? Ss Tue08PM 23:43.76 syslogd daemon 143 1.1 1.0 836 600 ?? Ss Tue08PM 22:50.86 /usr/sbin/portmap NIS master (SunOS 4.1.3_U1B): ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ USER PID %CPU %MEM SZ RSS TT STAT START TIME COMMAND root 60 34.0 0.3 68 236 ? R 20:27 280:09 portmap root 65 33.6 0.4 172 388 ? R 20:27 280:06 ypserv NIS slave (Solaris 2.6): ~~~~~~~~~~~~~~~~~~~~~~~ USER PID %CPU %MEM SZ RSS TT S START TIME COMMAND root 221 17.6 0.3 2264 1216 ? R Jul 29 386:43 rpcbind root 234 4.4 0.3 2368 1400 ? S Jul 29 474:11 ypserv In addition, from "syslog" on the FreeBSD host (the broadcast problem started at around 16:41 so the earlier messages may not be significant): Aug 26 05:19:24 rational xntpd[139]: time reset (step) 0.370413 s Aug 26 14:25:03 rational portmap[3024]: connect from 130.155.194.226 to callit(ypserv): request not forwarded Aug 26 14:25:03 rational portmap[3025]: connect from 130.155.194.226 to callit(ypserv): request not forwarded ... Aug 26 16:41:15 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp .csiro.au" not responding Aug 26 16:41:21 rational last message repeated 2684 times Aug 26 16:41:21 rational /kernel: xl0: command never completed! Aug 26 16:41:21 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp.csiro.au" not responding Aug 26 16:41:21 rational last message repeated 68 times Aug 26 16:41:21 rational /kernel: xl0: command never completed! Aug 26 16:41:21 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp.csiro.au" not responding Aug 26 16:41:21 rational /kernel: xl0: command never completed! Aug 26 16:41:21 rational last message repeated 3 times Aug 26 16:41:21 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp.csiro.au" not responding Aug 26 16:41:52 rational last message repeated 13943 times I thought it was interesting to see the messages from the "xl0" driver - this machine has an on-board 3Com ethernet (as do most newish Dell's) and is connected to a 10mbit/s hub: xl0: <3Com 3c905B-TX Fast Etherlink XL> rev 0x00 int a irq 14 on pci0.17.0 xl0: Ethernet address: 00:c0:4f:68:e1:6a xl0: autoneg complete, link status good (half-duplex, 10Mbps) The problem is repeatable and as a test I tried to force the host to bind to particular servers using the "-S" option. It works after a reboot but no subsequently with similar problems on the NIS servers. I didn't see the "xl0" messages in this case though: Aug 28 13:08:48 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp.csiro.au" not responding Aug 28 13:09:19 rational last message repeated 14143 times Aug 28 13:11:20 rational last message repeated 55641 times Aug 28 13:21:21 rational last message repeated 261703 times Aug 28 13:31:22 rational last message repeated 253355 times Aug 28 13:34:19 rational last message repeated 75019 times Aug 28 13:34:20 rational login: ROOT LOGIN (root) ON ttyv0 Aug 28 13:34:20 rational ypbind[147]: NIS server [130.155.194.25] for domain "rp.csiro.au" not responding Aug 28 13:34:25 rational last message repeated 2403 times So I think this is a "real" problem and for some time I have been pushing the use of FreeBSD because of its stability and heritage as well as the excellent networking features but there is no way I can push the case with this sort of problem - I am sure you know what I mean :-) My only other experience with NIS on FreeBSD is on my home network (two machines, NIS master on 2.2.8-STABLE and NIS client on 3.2-STABLE (SMP) where it works fine). I have a quick search of the mailing list archives but I didn't see anything similar to this. If I can provide any further information just let me know. Many thanks, Shaun. To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-hackers" in the body of the message
