Michael Marschall said: > As someone has already stated, if you use the same hardware > configuration and features you do not need to worry about mac > addresses. In some cases if you use different hardware and the same > features you again will *often* not have to worry about mac addresses. > > Michael >
Okay thanks! This really rocks my boat :) I'll try and pull of large networks with this, if it works I'll post hardware/network requirements to this list. dP > On Tue, 18 Jun 2002, dP wrote: > >> > No offense but this sounds like you want someone to tell you how to >> > setup a solution that you have been asked to do. >> > >> > Michael >> > >> >> No not really, I know how I want the solution to be (although I've >> already changed my mind a few times thx to this list), I just need to >> know if it is possible using current tools (linux) and cheap hardware >> (jammins). >> >> And so far the only bottleneck seems to be an issue with manual >> per-workstation configuration (macadds) and network traffic. >> >> dP >> >> >> > On Mon, 17 Jun 2002, dP wrote: >> > >> >> I have been reading up on diskless networks, and some of the ways >> >> this works seems ludicrous to me. >> >> >> >> First of all, if you have large networks for schools, you want the >> >> systems to be exactly equal. Furthermore, you (well, I) would want >> >> a user-account for each student, managed by their teachers (I'll >> >> use LDAP or something), so when the user login through xdm, it >> >> automaticly mounts the home directory via nfs (this does not have >> >> to be fast). A few nice benefits: >> >> >> >> The entire root of the ws' would be alike, no changes are wanted so >> >> it's a simply nfs ro export (home directory mounted seperately from >> >> nfs server on X login). This means that you need only one standard >> >> nfs export for all workstations, which also makes updating/changes >> >> very easy (I'd use devfs and mounting /var to local ram and linking >> >> /tmp to /var/tmp as suggested). >> >> >> >> And when I say alike, I mean it, stuff like ~/GNUstep (for >> >> windowmaker, dunno what it's called in icewm, or whatever) would >> >> simply be symlinks to directories on the ro root. >> >> >> >> This makes it possible to put the whole root system in a ramdisk on >> >> the server (just 1GB max) - so what's the big deal about fast >> >> harddiscs??? On boot the server would simply extract a tarball to >> >> the ramdisk. And think about this: 1GB ram is cheaper than many >> >> super-fast-hd's. >> >> >> >> Please note that all hardware configurations will be similar (I'm >> >> thinking: http://www.disklessworkstations.com - seems fast enough >> >> for a client, right? though the quantity pricing isn't very big (75 >> >> max? I'm thinking what's the catch with these systems). >> >> >> >> And now to the irritating part, why oh why do you need to type in >> >> the mac- address in a configuration file. Isn't the whole point of >> >> DHCP to assign it a random IP automaticly? Since all workstations >> >> would have no bootup independent stuff, I cannot see the usage of >> >> static IP and hostname assignment. Is it possible to implement a >> >> new configuration scheme where the configuration file would simply >> >> contain an IP-range, and then it would automaticly assign a new ip >> >> and hostname (maybe based on the mac-address)? I just think that >> >> hardcoding mac-adds is lame, and it would remove the... automatic >> >> feel of it all. >> >> >> >> I'd very much like to know. >> >> >> >> I would be willing to donate money for someone to implement this >> >> (if it's not already there, God I'd sound lame then, but it didn't >> >> look like it in the docs) and build a woody package of it ;) >> >> However, I will not be payed until completion, so it would be out >> >> of my own pocket which have many holes (otherwise I'd not say yes >> >> to a project like this, but u know how it is). >> >> >> >> Anyways, my real question is this (for the server): >> >> What kind of hardware architecture would you use if there was say, >> >> 500 to 1000 workstations (and is it even possible with that many?). >> >> What about RAM, how many GB is required? I read somewhere that more >> >> than one cpu wouldn't mean a heck of a lot and probably make things >> >> worse, I happen to agree with that (this is something that needs to >> >> be stable for at least 10 years, by using ram-systems I hope this >> >> can happen). But as to the architecture, uh, I only know i386, all >> >> those fancy words (like alpha) have just been dreams until now... >> >> Can anybody recommend anything (doesn't need to be cheap, but not >> >> mainframe-priced either :) links to shops and useful information >> >> would also be nice. >> >> >> >> As to the network, 1Gbit from server to switches, 100Mbit from >> >> switches to workstations, this would be okay lowcost. And I'm >> >> thinking compressing the X data, albeit this would introduce >> >> greater load on the server. >> >> >> >> As to the software, imagine that all clients would run 1280x1024x16 >> >> (1024x764 would be acceptable, but...), and hmm, I don't know, say >> >> a custom icewm with openoffice1 and mozilla1.1. I know that these >> >> two apps are cpu- intensive, but I cannot for the life of me think >> >> about something else that would do for education and old windows >> >> lusers (no not opera, and thats final and just me). Note that >> >> mozilla would also be having plugins for java and flash... >> >> >> >> I'm very nervous about this, as I have never build such... grand >> >> networks, I've build my own linux systems from scratch (not LFS) >> >> sure, and I've optimized everything for everything on our >> >> webhosting servers, but this is very upscale for me, however, being >> >> a Linux freak, I have to do this myself, it's simply to great a >> >> temptation (imagine that you were hired to build linux networks >> >> like this, and not only build, replace windoze machines ;) >> >> >> >> Anyways, the diskless scheme seems superior to me, I just hope that >> >> the servers can handle that many workstations... well, they _have_ >> >> to. >> >> >> >> dP >> >> >> >> >> >> >> >> _______________________________________________________________ >> >> >> >> Sponsored by: >> >> ThinkGeek at http://www.ThinkGeek.com/ >> >> _____________________________________________________________________ >> >> 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 >> >> >> >> >> ---------------------------------------------------------------------------->> > Bringing you mounds of caffeinated joy >> >>> http://thinkgeek.com/sf <<< >> >> _____________________________________________________________________ >> 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 >> > > > ----------------------------------------------------------------------------> > Bringing you mounds of caffeinated joy > >>> http://thinkgeek.com/sf <<< > > _____________________________________________________________________ > 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 ---------------------------------------------------------------------------- Bringing you mounds of caffeinated joy >>> http://thinkgeek.com/sf <<< _____________________________________________________________________ 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