This is File-Servers-and-Network-Nodes in view mode; [Up]
Date: Sun 04-Nov-1991 17:00:01 From: goldman@atlantis.ees.anl.gov (Daniel A. Goldman) Subject: File Servers and Network Nodes What are you all doing to serve 105MB clients (actually any 2.x standard release) off of file servers to access the extended release? Do you mount everything from the server, use NFSextend, something else? Any opinions on the RIGHT way to do this and/or what NeXT will/should do?
Date: Sun 05-Nov-1991 01:42:07 From: eps@futon.SFSU.EDU (Eric P. Scott) Subject: Re: File Servers and Network Nodes In article <1991Nov4.170001.16610@mcs.anl.gov> goldman@atlantis.ees.anl.gov (Daniel A. Goldman) writes: >What are you all doing to serve 105MB clients (actually >any 2.x standard release) off of file servers to access the >extended release? Do you mount everything from the server, >use NFSextend, something else? Any opinions on the RIGHT >way to do this and/or what NeXT will/should do? Wipe the 105, rename it swapdisk, edit /clients/*/etc/rc.swap to raise hiwat= from 31457280 to 78643200, and pray that nobody tries to do any real work in Mathematica, because even if you could turn 100% of the 105 into swap, it still wouldn't be anywhere close to enough. As far as I'm concerned, there is no viable configuration for a 105MB system other than making it into a swap disk and NetBooting from a well-endowed server, unless you are happy using the machine as little more than a glorified terminal. Also, if you only have 8MB RAM in the clients, immediately upgrade to at least 20MB (24MB for color slabs), or they WILL thrash. While some people may tolerate this behavior in standalone systems, a fairly small number of "diskless" 8MB NeXTs can easily bring an Ethernet to its knees (not to mention that the performance will be a fraction of what a crippled-but-diskful configuration would exhibit). This is particularly important if you share the wire with other (non-NeXT) users. The NeXT software is incredibly good about caching stuff *if* it has enough free memory available. "The System"--even without any user applications--needs more than 8MB. NeXTs aren't intrinsically slow machines, but you can make them so by starving them. It's a lot easier to justify a small additional hardware expense than lost human productivity. Machines are cheap, people are not. -=EPS=-
Date: Sun 05-Nov-1991 15:26:19 From: steve@math.tamu.edu (Steve Johnson) Subject: Re: File Servers and Network Nodes In article <1991Nov4.170001.16610@mcs.anl.gov>, goldman@atlantis.ees.anl.gov (Daniel A. Goldman) writes: |> What are you all doing to serve 105MB clients (actually |> any 2.x standard release) off of file servers to access the |> extended release? Do you mount everything from the server, |> use NFSextend, something else? Any opinions on the RIGHT |> way to do this and/or what NeXT will/should do? We mount the following from our server ('040 cube) to 5 clients, each with a 105: /usr/local (ro), /LocalApps (ro), /LocalLibrary (rw - for MakeTeXPK), /NextDeveloper (ro), /NextLibrary (ro). /usr/spool/mail (rw), & home directories are mounted from a Mips & a Sun. On the clients, clean out most of /NextLibrary & /NextDeveloper; keep some stuff there (fonts, etc) in case of problems with the server. Replicate the remainder of the extended release in /NextApps, /usr/{lib,bin,include}, /lib, /bin, (others ?). Your local disk will be very full (90%), leaving around 20 MB for swap. This works for us, but it's ugly. OS upgrades will be a nightmare, but I'll probably use rdist, or similar. If we can get more RAM for our server, I may just make the 105's into swap disks. However, I don't have a great deal of confidence in the Cube as a fileserver (not the best disk performance and inability to do sync scsi, along with the increased ethernet load). - Steve Johnson, Systems Manager, Dept of Mathematics, Texas A&M University
These are the contents of the former NiCE NeXT User Group NeXTSTEP/OpenStep software archive, currently hosted by Marcel Waldvogel and Netfuture.ch.