> If they want new equipment go for the least expensive > processor available in > the client - but that is likely to be > 600 Duron.
If you're doing that, you probably have more than 32MB of memory in each one. In that case, since the whole of the LTSP filesystem fits into a 16MB block of cramfs, you can omit the NFS service completely for most clients. This makes those terminals completely autonomous of the LTSP boot environment when in a static stable environment. Your boot server starts your test clients (say the 10 machines closest to your office) with normal NFS usage. Machines known to have 32MB or less memory use NFS too. The remainder receive both the kernel and the cramfs using TFTP, with no subsequent load on the boot server. Thus, an app server can act as a backup boot server. > If you can > load balance them you can also build in redundancy and > probably use less > expensive components so more medium spec servers end up > cheaper than one mega > all singing and dancing one. I suggest telling the clients to use XDMCP broadcast to find their login server. By dynamically changing the contents of the Xaccess files on the application servers, you can move clients around as availability and loading patterns change throughout the day. Each of the 490 machines contacts the boot server once on power up, then never again. Every time a user wants to log in, the client machine will find a desktop server that is willing to accept its IP address. _____________________________________________________________________ Ltsp-discuss mailing list. To un-subscribe, or change prefs, goto: https://lists.sourceforge.net/lists/listinfo/ltsp-discuss For additional LTSP help, try #ltsp channel on irc.openprojects.net