Thanks for the ideas. We are a statistics research department, so folks are very keen on where they run their apps since any given simulation may have different RAM and processing requirements. So really, the only thing I want to virtualize is the X session itself. All they do in the X session is run Thunderbird, Firefox, and sometimes they'll run R or SAS for simple compute jobs (not long-running simulations). If they have a large matrix to invert, they'll fire up an R session via SSH on another box. My predecessor set up the thin client environment so that each thin client that gets booted has 1 core/cpu to run off of. So, if he had a dual core server, he would assign 2 people to that machine. The problem is that these machines are ancient and sucking a ton of power. They need to go. How many cores/processors should there be for 1 person's thin-client session?
This is why I was thinking of having just a few machines with at least 8 cores each, and not have the user be concerned with which they're logging into. They used to run their sessions off of other machines if they had placed something in /scratch and needed to get back to it. I'm correcting this by implementing centralized storage as well. If clustering and/or virtualization is not the solution, I could continue to use a pseudo load balancer using DNS. Right now our servers look like this: desk00 desk01 desk02 . . . desk22 The user can choose which to run their X session off of. Alternatively, to help balance this out and not assign students to "desk" servers, I have another dns name called best-statdesk, and if they use that it just gives them the least used machine at that moment. Not completely accurate but it helps. I could continue to do this with just 3 highly powered machines and centralizing their storage so they won't be compelled to log into server1 vs. server2 etc. Hopefully this lays out a little more clearly what we've been up to. If anyone has a better way to deal with this I'm all ears. The goal is to reduce 22 back-end machines down to only a few for about 80-100 thin-clients. On 12/13/11 6:32 PM, Peter Scheie wrote: > One approach worth considering would be to dedicate specific > applications to their own isolated server on the backend. This wouldn't > reduce the number of server machines you have, but it would allow you to > devote hardware resources to those apps that need it, and not waste > hardware on those that don't. That way a gluttonous app wouldn't > negatively affect the other apps. You'd have to do some customization > to the menus to make the menu selections call the respective server, but > that should be a one-time thing, one of the benefits of LTSP. > > Peter > > Jeff Siddall wrote: >> Statistical load spreading should be trivial -- just a matter of sharing >> home directories and then spreading your clients around your various >> servers. >> >> True load balancing? I have no idea if/how that would work. Hopefully >> someone else can answer that. >> >> Jeff >> >> ------------------------------------------------------------------------------ >> Systems Optimization Self Assessment >> Improve efficiency and utilization of IT resources. Drive out cost and >> improve service delivery. Take 5 minutes to use this Systems Optimization >> Self Assessment. http://www.accelacomm.com/jaw/sdnl/114/51450054/ >> _____________________________________________________________________ >> 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.freenode.net >> >> > > ------------------------------------------------------------------------------ > Cloud Computing - Latest Buzzword or a Glimpse of the Future? > This paper surveys cloud computing today: What are the benefits? > Why are businesses embracing it? What are its payoffs and pitfalls? > http://www.accelacomm.com/jaw/sdnl/114/51425149/ > _____________________________________________________________________ > 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.freenode.net ------------------------------------------------------------------------------ 10 Tips for Better Server Consolidation Server virtualization is being driven by many needs. But none more important than the need to reduce IT complexity while improving strategic productivity. Learn More! http://www.accelacomm.com/jaw/sdnl/114/51507609/ _____________________________________________________________________ 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.freenode.net
