I know nothing about this.
Can I have a number of diskless machines booting and running Boinc from a central disk over gigabit network?
To add perspective think pentium D on say MCM1
I know nothing about this.
Can I have a number of diskless machines booting and running Boinc from a central disk over gigabit network?
To add perspective think pentium D on say MCM1
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
Short answer: yes.
There's several ways of doing it, at least two that I can think of would allow you to run diskless machines, the catch there being that it uses the hdd on the head node (main machine) for all it's caching etc and you'll see really high IO on that drive, potentially high enough to slow down the head node if you have enough worker nodes. If you've got buckets of RAM on those worker nodes you could get away with a bit.
It might be easier to build a USB drive for them to boot and run from though, and just copy that image to multiple USB sticks.
[SIGPIC][/SIGPIC]
I notice that I can get Pentium D for pennies without drives and with 1GB. C2D's might be possible too either way these would have to be only in the coldest times or re- located from here.
Adding a drive to each would be prohibitive.... even a stick would possibly cost more than the rig at the price I saw (around $6 each in bulk)
It would be a exercise in minimal setup cost crunching/house heating
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
IF you have plenty of space, by which I mean metric as$ loads, on your main hdd then it's doable. I remember reading about cluster systems that worked exactly like this. I'll see if I still have the links.
[SIGPIC][/SIGPIC]
I think this might be helpful: http://pareto.uab.es/mcreel/PelicanHPC/
[SIGPIC][/SIGPIC]
yes
use ramdisk for slots dir, but load everything else from remote node
or maybe
http://en.wikipedia.org/wiki/Beowulf_cluster
Last edited by Evantaur; 10-24-2014 at 12:57 PM.
I like large posteriors and I cannot prevaricate
I did this years ago with ccboot. The thread should be here somewhere. I had a head unit and 5 slave machines. Worked *great* as long as you don't bootup all 5 nodes simultaneously. The boinc projects also went to one of my famous RAM drives, so there was no problems. Just don't use flaky NICs, cables, etc as a disconnect from the workstation to the PXE server results in an immediate BSOD.
Absolutely it's do-able. Did it with some dual-socket 6 core SB rigs back in 2010 was it...? Had 1 head node and 4 slaves iirc... Had those things hooked up with some infiniband QDR and the IO was still insane. Worked much better once we put HDDs (Might have been SSDs? Don't remember...) in each node. (Dang shame I didn't get to keep those nodes...)
Those Pentium Ds though aren't going to do much crunching - and it'll be a lot of effort. You'd be better off just keeping some of your "retired" rigs for winter heat or joining a GPU project during the winter...
IMHO...
Thanks for the comments guys.
Maybe not with Pentium D's then Otis
My Biggest Fear Is When I die, My Wife Sells All My Stuff For What I Told Her I Paid For It.79 SB threads and 32 IB Threads across 4 rigs 111 threads Crunching!!
Bookmarks