🚀 go-pugleaf

Michael Bruewer writes
> 
> We have a pool of four NeXTs. One netinfo-server, one  
clone server and two  
> clients. All user accounts are network accounts. All home  
directories are nfs  
> mounted on all machines.
> Now the problem is, that if one machine is overloaded and  
the nfs server is not  
> responding, all machines are hanging (little disk icon  
turning) and nobody can  
> work. (I'm not sure if it should read "nearly all" and  
"nearly nobody"). Even  
> if I have a remote shell connection to a machine outside  
the pool, and the nfs  
> server of the NeXT, where my home directory lives,  
doesn't respond, the rsh  
> connection hangs.
> Is this behavior correct or is there something wrong with  
our installation?

Unfortunatly, this is "normal".  NFS is stateless.  So when  
a server is unavailable, it just waits for the server to  
become available again.  The workspace under 2.x ran as a  
single process, so that if it was stating a remote file, it  
would hang there.

Your rsh to another machine would hang probably because csh  
was trying to read .cshrc in your home directory.  If the  
server was unavailable, it waits, and waits and waits...

Some ways around:
	1.  If the nfs server is also a print server, give  
it LOTS of memory.  Ours run 32 MB.
	2.  Disable running of Mathematica on the nfs  
server.
	3.  Try increasing nbuf.  (See discussion a few  
weeks back)
	4.  Increase the number of nfsd's running on the  
server.  I've settled on running 4 per disk served.
	5.  If there is a router between the server and  
client, change rsize and wsize to 7k.  (They default to 8K)   
This is especially true if it's a cheap PC based router.   
Repeaters don't seem to have this problem.
	6.  Turn on the switch for large file systems in  
Preferences.  This stops the prefetch in the workspace so  
that when an NFS server goes down, the workspace doesn't  
hang until you access something on that server.