Slashdot Mirror

It may be true for some hardware configurations, but I've been running 2.4.x on several different slackware boxes, including one old laptop (Toshiba P-120 "Satellite Pro"), with no problems whatsoever (except for the glitches in loopback mounting, which I hear should be fixed in this release) for months now. Just download the source and go - why wait for an official .tgz?

I just wish Mosix would get updated for 2.4.x support - I've been itching to set it up at home again.

Use MOSIX. It is a single Linux patch ( to 2.2.18 or 2.2.17 ) which
allows clustering of normal applications.
Beowulf apps have to be specially written to take advantage of the cluster, with MOSIX,
any process or thread can be scheduled on another cluster member.

I don't think .NET has been released yet, though. As for the "open source hype", well, I am using open source technology exclusively on our systems at work and it has been an extremely successful venture. To give you some idea, we have up to 5,000 mail accounts running on exim, 3,000 shell accounts, run an industrial strength DNS system, industrial strength, internally developed network management systems running on Zope/Python, and a staunch news server all running off an 8-node MOSIX cluster.

What I would really like to see would be something that I could install on these windows workstations so that the linux render servers could use their spare cycles when they aren't being used. Something like mosix, but with Windows clients would be amazingly useful for us. Does anybody know if such a program exsists.

I like this article because it has hard facts and doesn't seem to have the standard media hype tone about how Linux will revolutionze the desktop, topple MS, take over the world, and declare world peace. (MS, on the other hand, would declare world FUD.)

Linux is a mighty fine OS, but there are others out there that people will still want to use. So may all *nix*es unite and take over the world *together*! Can we say... Beowulf? Or... MOSIX?

(just kidding...)

The SETI@Home idea is also not worth doing because SETI@Home is so perfectly distributable that barely any communication between nodes is needed.

I don't know how bothered you are about being able to use the magic "beowulf" word, but can I suggest instead that you look at mosix clusters instead? Mosix allows processes to migrate accross a network transparently. So you don't have to write your applications specially for mosix, you just have to apply the kernel patch to each machine in your cluster.

Has anyone actually used Mosix?

What's the difference between a beowolf cluster and a Mosix one?

To me Mosix looks better suited to those of us who have more workstations than they can use at once, as software doesn't have to be written specially to use it.

How about using the machines as that?

...you could always use Mosix. Its a clustering implementation that doesn't require the special optimization of code or a recompile like PVM and other tools do.

It's really quite easy to send processes to other nodes of the cluster... and I'm thinking that if the program this guy is using performs well on an SMP system, it might-possibly-i-reckon-could work on a MOSIX cluster.

Any ideas people?

Mike

Isn't this exactly what Mosix is all about? Dumb cpu time clustering for threaded applications?
Some day, I'll try it so I know, I promise :-)

What about Mosix?

I'm eagerly wondering about these, as well. I've got a collection of 'scrounged' computers at home on my network, a copy of Mosix, and the ability to write download scripts for packets...I'm just itching to find out how big my home cluster will have to get to finance a broadband connection for me... :-)

Been there, done that.
It's called mosix , it has some pretty neat herustics to figure out if it would be better to swap this to another machine, or leave it on the current machine. In fact, if you run it at a low priority, most of the people on the network probably wouldn't notice much of a speed drop. However, it's not ported to NT, so you'll just have to stick with linux (I know that'll just ruin your whole day)

Mosix allready exists and is an cluster extension for Linux.

--

check out mosix here

• If so, how would it handle process sharing?
• Are you going to lay dynamic process scheduling over some sort of heterogenous message passing system?
• Or similarly, but more simply, are you building a preemptive process distribution without dynamic scheduling? (Gathering data from network when finished.)

there already is such a thing, although it's not as closely integrated as a beowolf. it's a process-migration kernel enhancement that allows several machines (whether workstations or clusters) to share cpu loads.

check it out at mosix.org. it is especially useful for compiling large projects (i.e. xfree86).

jon

And sure they did. In fact, more than 10 years before Beowulf, e.g. look at MOSIX

While the Beowulf patches come out of NASA, there's a whole bunch of stuff out there which isn't written in the US at all. For example, the best session clustering technology for Linux is MOSIX which is put out by Hebrew University in Isreal. To my knowledge this isn't export restricted at all, and is released as a set of patches against the main kernel tree. Anyone with basic System Administration skills could set up a Mosix cluster pretty quickly.

If you're less interested in interactive clustering and need computational load balancing instead, there's a whole slew of batch queuing packages available from GNU QUEUE to the many derivatives of NQS out there. Here's the a href="http://www.cmpharm.ucsf.edu/~srp/batch/syste ms.html">Yahoo Batch Queuing Page" for a short list of many popular packages.

I don't think the US government could stop any nation from purchasing commodity hardware manufactured from around the world, installing a basic Linux or BSD distribution, and setting up a batch queue or other type of basic cluster. Never mind that a sufficiently serious government could just up and write their own... in my department at BBN (Speech and Natural Language Processing) we use an internally written batch manager which is surprisingly simple... all written in C.

I think I agree... :)

There is indeed an art - if not a Zen - and a science to knowing when to optimize stuff, and I think it boils down to 'use the hottest algorithm around' (especially if you start in C), *followed* by 'hack the ASM by hand'.

Something I finally got round to doing just this past weekend, that I'd been meaning to do for a while, is it compare the speed of execution of loops that simply count between 0 and a big number, but using different comparison methods, and different compiler optimization levels in gcc, with Perl for the hell of it as well.

The full source I used follows:

#include
#include
#include

#define INNER 5000000
#define OUTER 200

void up(void), down(void), xor(void);

int
main (void)
{
int c;
time_t t0, t1, t2, t3;

t0 = time (NULL);

for (c = 0; c Mosix cluster, which was "interesting", to say the least... ;)

As for perl, well that behaved the way I expected. The whole thing was way slower than C, but it did at least do an xor when I asked for it - that was almost invariably faster than counting up with "".

Well, those be the results... compile the source yourselves with all the various gcc options and compare the assembler, folks :)

We conducted a study of an ALPHA port of MOSIX and could do it.
A. Barak

We had a small cluster of Alpha PC164's and a much larger of Pentiums. We "upgraded" from a Beowulf-style system to a Mosix system, and performance shot straight through the roof. Unfortunately, Mosix doesn't run on Alphas (yet), so we bit the bullet on 'em and have a bunch of glorified paperweights.

I'd love to get back on the Linux train with the Alphas, but Beowulf seemed to have too many if's and require far more effort than it returned benefits. Has this changed, and how?

MOSIX supports several cluster configurations, e.g workstations join/leave the cluster when you don't/use it.