nfsd(1M) nfsd(1M) NAME nfsd, biod, bio3d - NFS daemons SYNOPSIS /usr/etc/nfsd [ -a ] [ -p proto ] [ -c nconns ] [ nservers ] /usr/etc/biod [ nclients ] DESCRIPTION nfsd starts the nfs(4) server daemons that handle client NFS requests. biod starts asynchronous block I/O daemons. This command is used on a NFS client to perform buffer cache read-ahead and write-behind. bio3d processes are the asynchronous kernel processes for NFS version 3. No daemon is necessary to start bio3ds. There is one bio3d associated with each mounted filesystem at mount time. If the NFS traffic to a filesystem increases, up to three more bio3d processes are spawned for that filesystem. When traffic subsides, these extra processes exit. OPTIONS -a Allow the NFS server daemon to service client requests over both the UDP and TCP protocols. This is the default. -p proto Allow the NFS server daemon to service client requests over either the UDP or the TCP protocols. The -a option overrides this selection. -c nconns Set an upper limit on the number of simultaneous TCP connections the NFS server has open. By default this number is 10000. nconns can be equated with the total number of clients mounting files from this server over TCP; it has no effect on clients mounting over UDP. nservers This is the number of NFS server daemons to start. From 6.5.22 the IRIX kernel manages NFS server daemons dynamically according to load; specifying a number will most likely result in reduced performance. See DYNAMIC NFS DAEMONS below. nclients The number of asynchronous block I/O daemons. A good number for nclients is 4. The default is 1. These daemons are started during system initialization from the /etc/init.d/network script if the configuration flag nfs is set on (see network(1M)). DYNAMIC NFS DAEMONS From IRIX 6.5.22, the IRIX kernel NFS service can operate in one of two modes, depending on the options passed to nfsd. The first mode is the pre-6.5.22 behavior, in which the nservers argument to nfsd controls the number of NFS daemons available to handle NFS calls. In this mode, a fixed number of daemons is used regardless of NFS load. A single number chosen by the system administrator must cater for both low-load and high-load conditions. The usual result is a compromise number whereby NFS performance suffers under high-load conditions and system resources are wasted in low-load conditions. When the nservers argument is given to nfsd, IRIX reverts to this old mode, but issues a warning to the system log. In the second mode, the IRIX kernel dynamically manages NFS daemons depending on the load of NFS traffic the machine is serving. Daemons are spawned and killed as load increases and decreases. This removes the need for the system administrator to tune the number of NFS daemons using the nservers argument. Previous versions of this manpage and other SGI documents discussed the tuning of the number of servers with an emphasis on reducing the number of servers to limit resource wastage in low-load or steady-state conditions. This has unfortunately lead to many IRIX systems being under-configured for steady high-load conditions or load peaks, resulting in poor NFS performance at those times. For IRIX systems at 6.5.22 or above, the recommended configuration is to enable the new dynamic mode by removing any nservers argument from /etc/config/nfsd.options. This configuration can be expected to deliver both optimal NFS performance under high-load conditions and minimal resource usage in low-load conditions. This configuration is shipped as the factory default. The new dynamic daemon mode is controlled by dynamic system tunables rather than arguments to nfsd. This enables the system administrator to change aspects of the dynamic daemon behaviour without the interruption in NFS service resulting from killing and restarting the nfsd daemons. See the systune(1m) manpage and the comments in /var/sysgen/mtune/oncrpc. Trusted IRIX (TRIX) LIMITATIONS On TRIX systems, NFS over TCP is unsupported. For IRIX versions 6.5.29 and earlier, it is recommended on systems running TRIX, that in order to ensure that NFS runs over UDP, the "-p UDP" option is invoked when nfsd is started. For IRIX 6.5.30 systems running TRIX, nfsd facilitates NFS over UDP by default, in lieu of the unsupported NFS over TCP. FILES When a file that is opened by a client is unlinked (by the server), a file with a name of the form .nfsXXX (where XXX is a number) is created by the client. When the open file is closed, the .nfsXXX file is removed. If the client crashes before the file can be closed, the .nfsXXX file is not removed. FILES .nfsXXX client machine pointer to an open-but-unlinked file /etc/config/nfsd.options Site configuration file containing options for nfsd. /etc/config/biod.options Site configuration file containing options for biod. SEE ALSO exportfs(1M), mountd(1M), network(1M), systune(1M), exports(4). Page 3