Date: Sun, 3 Mar 1996 13:01:41 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: monboso@masternet.it (Amedeo Beck Peccoz) Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: xfs not working properly Message-ID: <199603032001.NAA04482@phaeton.artisoft.com> In-Reply-To: <199603031640.RAA04762@masternet.it> from "Amedeo Beck Peccoz" at Mar 3, 96 05:42:47 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> I'm the sysadmin of our 7 PCs Ethernet LAN and I decided to switch from > Novell Netware to FreeBSD 2.1, so I put xfs 1.91 as a TCP client on the PCs > side, but unforunately I had to observe the following: > > 1) the FreeBSD server is much slower as now I can only get an average speed > (Norton SYSINFO) of 87 K/s in readings and ONLY 10K/s in writings!!! while I > used to run at 250K/s with the Novell one. Is this a problem in xfs or in > FreeBSD? Note that at present some applications (e.g. Excel) are really > unusable!! :-( I've also noted that nfsstat on the server reports an > execessively high value in the "Server Cache Stats: Misses" value. Is this > correct? How can I configure nfs caches? I haven't found any help in the mans! > Here is a tipical server cache line from nfsstat after 2 day the servers has > being working: > Server Cache Stats: > Inprog Idem Non-idem Misses > 1 30 0 611022 Where is this information coming from? In general, the NFS on the server does not, strictly speaking, have a cache. The FreeBSD VM system disassociates the vnode from the underlying inode when the last refrence goes away (since NFS is stateless, this is each time. This seriously impacts the NFS numbers; unfortunately, other than significantly changing the buffer cache (to use a device/offset rather than a vnode/offset block index and thus not invalidate cache pages for which there are not active vnodes), there's very little that can be easily done. This is not enough to account for your very low numbers, however, since NFS-to-NFS doesn't have numbers this bad in a pure BSD environment. I suspect it is the cache on the PC, not on the server itself, that is blowing out. Make sure you have configured it to be large enough. Make sure the XFS cache is keying off the invariant portion of the cookie; most modern NFS's vary the cookie as part of making valid cookies hard to guess. If the cache is keying off the whole thing, it will get an inordinate number of misses. My personal advice would be to go to SMB clients instead of XFS and run Samba (a LAN Manager compatible server available at no cost) on FreeBSD. > 2) I have to set to rwxrwxrwx (777) the rights of all the files in the > exported dirs to let the clients see them, and for the same reason a line > exists in my /etc/crontab file that executes a "chmod -R 777 /export" every > 10 minutes, otherwise the newly created files are not accessibles by other > users. Is this a problem in the FreeBSD rpc.pcnfsd daemon or maybe in the > nfsd daemon or is there something wrong with xfs? I made some experiments > and found out that the clients can properly see a directory if its group is > the login group of the user, but if the group of the dir is another group > the user belongs to, he (or she) cannot enter the dir if I set its rights to > rwxrwx--- (770) Group permissions are generally sent across the wire in the NFS case, since the machine at the other end is assumed to be a trusted host. For PCNFS (ie: xfs), there is an authentication daemon that does the cookie generation on name lookup, and the xfs client is expected to authenticate to that. You should look at the default umask and the GID for the client program; in general, the need to open things up wide like this is a reflection of not having set up the client to authenticate correctly to the server. Hope this helps. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199603032001.NAA04482>