Page 1 of 1
Cluster Computing Using Puppy
Posted: Sun 19 Oct 2008, 18:06
by NickBiker
Hello
I just wondered if anyone had tried to use Puppy as the o/s for cluster/grid computing. This is where you have a number of hardware boxes, all linked by network, that can run code in parallel (see
http://www.open-mpi.org/) - used to tackle NP-hard problems (i.e. lots and lots of number crunching)?
It strikes me that a thin-puppy, with the right software would be great for this?
Any offers?
Nick
Posted: Mon 20 Oct 2008, 11:16
by Aitch
Hi Nick
I've asked the same question before, though don't think there was a conclusive answer
Have a look here:-
http://www.mcsr.olemiss.edu/bookshelf/a ... uster.html
btw your webcam link is broken
Aitch
Posted: Mon 20 Oct 2008, 12:39
by Lobster
There was an attempt
they tried . . .
then we never heard from them again
maybe their head exploded?
These things happen watch the film 'PI' which shows the effects of
too much coding . . .
Clusters
Posted: Mon 20 Oct 2008, 12:56
by NickBiker
Hi
Well I am determined to have a go! I suspect I will use OpenMPI and a cut-down puppy for each of the clusters. I let you know how I get on!
In meantime, anyone know of a nice multi-core processor usage viewer? Dare I say that Windows has a nice task manager where you can see each core and the use - is there similar for Linux?
Thanks
Posted: Mon 20 Oct 2008, 15:17
by alcy
Posted: Mon 20 Oct 2008, 15:35
by Lobster
The matrix has you
http://www.murga-linux.com/puppy/viewto ... 214#137484
Maybe you can collaborate
Good luck
Posted: Mon 20 Oct 2008, 17:04
by Aitch
Nick
It's a while since I looked into this, but what put me off was I was hoping to just use my own gash hardware, but several recommends said 'High speed single core' for the server was more reliable, and
matching P3s for clients
Just what I didn't have!
I would have thought matching network speeds more important, but what do I know?
I have a dualie P3/1.1Ghz dell server/scsi h/ds & several P3 667/500Mhz vanilla boxes/SCSI/ATA mixed
Be very interested, as big_bass/MU have just produced Puppy4.1 with SMP on slackware 12.1 kernel, AFAIK
http://www.murga-linux.com/puppy/viewto ... &start=105
Don't know if it SCSI boots though, as this is new -SitHeelSpeak was working on SCSI boot with the latest kernel
Good luck, can help maybe with research
Aitch
Posted: Mon 20 Oct 2008, 19:29
by John Doe
I've been hoping to test these projects out sometime myself:
openmosix, which only made it to the 2.6x kernel with an alpha version.
LinuxPMI seems to be currently at a 2.6.17 kernel.
Those two are both kernel based.
After having checked out the Open MPI FAQ for
Compiling MPI applications, it looks as though apps would need recompiled to work with this system.
I would prefer a kernel based system myself (in order for 'anything' to just work 'out of the box'), but it's a neat idea to try as an alternative. Perhaps someone could even implement a "Open MPI T2 build system".
Posted: Mon 20 Oct 2008, 21:33
by KF6SNJ
This might help.
http://boinc.berkeley.edu/
I may include this in RoboPup.
Posted: Tue 21 Oct 2008, 19:45
by Aitch
Wolfpup has boinc sorted/confirmed by Jay & others
http://www.murga-linux.com/puppy/viewto ... d6bc8ac8d8
Aitch
Posted: Tue 21 Oct 2008, 20:23
by cb88
http://svn.oscar.openclustergroup.org/trac/oscar/wiki
tried getting some stuff working with oscar.. oscar worked but the front end to the software was poorly programed and we never got it to work
Posted: Wed 22 Oct 2008, 19:27
by shroomy_bee
lol, pi is also about what happens when you show a computer how it executes and runs its code; kinda like if you thought about how you think - and then also experienced the neurons flashing and the neurotransmitters moving through the gates / synaptic junctions........the CPU develops biomaterial, then ants live in your computer. Or something.
Rocks is supposed to be good and user-friendly for clusters. I managed to blag a copy from the official servers but haven't got the boxes to test it; not sure if it would run on multiple virtual machines?
Beware, it is a giant download (multiple disks - the link says 1?! maybe DVD then, not CD) and only seems available on slooooow servers, last time I checked,
http://distrowatch.com/table.php?distri ... ckscluster
see also FreeBSD cluster project:
http://www.mini-itx.com/projects/cluster/?p
Posted: Sun 27 Dec 2009, 18:37
by Aitch
bump
It's been over a year...sadly NickBiker seems to have disappeared shortly after posting interest....?
Many developments have taken place in the business world
NFS/NetApp/Lawsuits by Sun
http://www.theregister.co.uk/2008/10/07 ... validated/
but the CEO is buoyant
http://blogs.netapp.com/dave/2009/11/wh ... nning.html
http://blogs.netapp.com/dave/2009/10/se ... -ever.html
NFSv4.1 has an RFC number
http://blogs.netapp.com/eislers_nfs_blog/
Micro$oft open source [!] NFS client
http://www.engin.umich.edu/newscenter/p ... 111000nvi/
Cloud Computing on free ESXi issue
http://searchservervirtualization.techt ... 40,00.html
Cloud Computing explained
http://blogs.netapp.com/dave/2009/10/iv ... uting.html
Any further interest...?
I came across this apparently dead project, recently, which sparked this post
TCPHA implements an architecture for scalable content-aware request distribution in cluster-base servers. It implements Kernel Layer-7 switching based on TCP Handoff for the Linux operating system. Since the overhead of layer-7 switching in user-space is very high, it is good to implement it inside the kernel in order to avoid the overhead of context switching and memory copying between user-space and kernel-space, furthermore, the responses are sent directly to clients, not passing the dispatcher, which will greatly improve the performance of cluster.
TCPHA, inspired by KTCPVS and IPVS, merges their strongpoints. Otherwise, the installation and configuration are very simple.
It is the initial release, debugged on Linux offcial kernel 2.4.20. There are a lot of work to do, such as good scheduling algorithm, better persistent connction handling, implementing it on Windows (i don't think it is so difficult:)). If you are interested in the development, you are very welcome, hopefully we will make it a useful one in the near future.
TCPHA is released under GNU GPL(General Public License).
http://dragon.linux-vs.org/~dragonfly/htm/tcpha.htm
It's a subproject of LVS, which also seems a little tired [2008 is news???]
http://www.linuxvirtualserver.org/
Not quite AI/Adaptive Fuzzy-Neural Control, but perhaps some of our C coders could have a look?
It maybe of interest to JohnM for the forum server...?
or others....
It seems this is a clear trend in paravirtualized server environments
http://www.vmware.com/company/news/rele ... ation.html
Aitch
Posted: Thu 15 Apr 2010, 20:27
by Aitch
bump
Just to whet some appetites.....I couldn't resist it!
Some render tests result here:
The storming sea scene
file 24 frames in resolution 4000x4000 pixels.
- On a MacPro 2 x DualCore Xenon 2.66 Ghz with 4 GB ram
took 552 minutes (9.2 h)
- Helmer did same 24 frames in 4k format in 64 min.
Some approximate numbers give Helmer a floating point capacity of 186 Gflops
http://helmer.sfe.se/
Aitch
Posted: Thu 15 Apr 2010, 23:07
by DMcCunney
Aitch wrote:Just to whet some appetites.....I couldn't resist it!
Some render tests result here:
The storming sea scene
Sheesh. Next time, may I suggest a simple hyperlink to the image instead of IMG tags?
Doing it with IMG forces the screen width of the entire thread to be wide enough to accommodate the image, and
long lines of text you must side-scroll to read are a pain.
It did quite the opposite of whetting my appetite.
______
Dennis
Posted: Mon 07 Feb 2011, 13:31
by Master_wrong
It strikes me that a thin-puppy, with the right software would be great for this?
Any offers?
i have....
here...
http://murga-linux.com/puppy/viewtopic.php?t=63705