Date: Wed, 10 Apr 1996 14:33:04 -0700 From: "Jin Guojun[ITG]" <jin@george.lbl.gov> To: joerg_wunsch@uriah.heep.sax.de Cc: freebsd-bugs@FreeBSD.ORG, gibbs@freefall.freebsd.org, jin@george.lbl.gov, jkh@time.cdrom.com Subject: Re: Unsuccessfully to install 2.2-960323 SNAP Message-ID: <199604102133.OAA21067@george.lbl.gov>
next in thread | raw e-mail | index | archive | help
} > A little more information. The FTP installation also pauses at this point } > for a second, and then continues. There must be some negotiation going on } > at this point and it is broken for NFS: } > } > ... } > stand/ee } > stand/df } > stand/cat } > stand/ls } > hanging } } Well, it's the first ``real file'' that is being fetched from the } server at this moment: } } j@uriah 94% gzcat < root.flp | cpio -ivt } drwxr-xr-x 3 root wheel 0 Feb 14 11:09 . } drwxr-xr-x 3 root wheel 0 Feb 14 11:09 stand } drwxr-xr-x 11 root wheel 0 Feb 14 11:09 stand/info } drwxr-xr-x 2 root wheel 0 Feb 14 10:02 stand/info/bin } -rw-r--r-- 1 root wheel 1799 Feb 14 10:02 stand/info/bin/bin.inf } ... } -r-xr-xr-x 59 root bin 0 Feb 14 09:34 stand/ee } -r-xr-xr-x 59 root bin 0 Feb 14 09:34 stand/df } -r-xr-xr-x 59 root bin 0 Feb 14 09:34 stand/cat } -r-xr-xr-x 59 root bin 0 Feb 14 09:34 stand/ls } -r-xr-xr-x 59 root bin 811008 Feb 14 09:34 stand/root_crunch } } (End of archive.) } } All the ``files'' above are actually hard links to root_crunch. So it } really looks like the generic NFS hang problem. For some reason, your } NFS server is pumping packets quicker than your ethernet card can pick } them up from the wire. However, the default NFS block size is 8 KB, } and NFS can only ``think'' in full NFS blocks, so it retransmits the 8 } KB block again. Apparently, this will hang the system. } } The ``slow NFS'' was supposed to get around it by reducing the block } size to 1 KB, so it always fits into a single UDP packet. It should } always be possible to retransmit it. If your ethernet card is very } lossy for some reason, the UDP/timeout/retransmit negotiation of NFS } will however drastically slow down the overall data rate, quite more } that TCP will. } } I've also noticed that my installation attempts required ``slow NFS'' } where i can't seem to remember that i had to specify this before. } (Server and client are both sluggish 3C503 8-bit cards.) } } -- } cheers, J"org I am not using any the old 3Com ethernet card. I am using 3C509B which can run at 10 MBps. Also, the server and the client are using the same type of cards, also, the server and the client are both running FreeBSD, also, the FreeBSD 2.1.0 installation via the same situation works well. The tcpdump did not show data transforming: tcpdump src iss-p2 or dst iss-p2 tcpdump: listening on ep0 12:50:58.756149 arp who-has ir10gw.lbl.gov tell iss-p2.lbl.gov 12:50:58.756680 arp reply ir10gw.lbl.gov is-at 0:0:c:49:13:b 12:50:58.765132 ns2-old.lbl.gov.domain > iss-p2.lbl.gov.1024: 3326* 1/5/5 (237) 12:50:58.766958 arp who-has pesto.lbl.gov tell iss-p2.lbl.gov 12:50:58.767112 arp reply pesto.lbl.gov is-at 0:20:af:eb:84:56 12:50:58.767724 iss-p2.lbl.gov.628 > pesto.lbl.gov.sunrpc: udp 56 12:50:58.768207 pesto.lbl.gov.sunrpc > iss-p2.lbl.gov.628: udp 28 12:50:58.769588 iss-p2.lbl.gov.629 > pesto.lbl.gov.sunrpc: udp 56 12:50:58.769974 pesto.lbl.gov.sunrpc > iss-p2.lbl.gov.629: udp 28 ----------------------------------------------------------- pesto # uname -a FreeBSD pesto.lbl.gov 2.1.0-RELEASE FreeBSD 2.1.0-RELEASE #0: Fri Feb 9 15:55:01 PST 1996 root@pesto.lbl.gov:/usr/src/sys/compile/PRODUCT i386 ----------------------------------------------------------- Finally, I proved my guessing (RPC problem) by trying and making the errors: ========================= ************ ======================== DEBUG: Network initialized successfully. mount_nfs: bad MNT RPC; RPC: Timed out DEBUG: Mounted NFS device pesto:/usr/local/src/test44.bsd onto /dist DEBUG: Request for floppies/root.flp from NFS hanging ========================= ************ ======================== Jordan tald me that 2.2 is using a new RPC. I believe it works, but the one in the installation must be have a bug that makes such painful. Otherwise, how 2.1.0-RELEASE work fine, but 2.2 not? I am not insist on getting the installation code for debugging, also, I believe that our ethernet may be acting a little strange. The 2.1.0-RELEASE, however, can work such envronment, and 2.2 cannot, then it is 2.2 installation problem. It may go away after the RCP code improved in the future, but we don't know the certain time. So, if you cannot reproduce the problem, which implys a hidden bug in RPC, and would like me to do debug on this issue, please let me know. I will take sometime to do so. Thank you, -Jin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199604102133.OAA21067>