MandrakeClustering Shows Off At ISC2003
joestar writes "Just released today at ISC2003, Germany, is "MandrakeClustering", a high-performance computing Linux distribution/solution, which sounds interesting, at least in the PR: Pentium support with optimizations made with the Intel compiler, 64-bit Opteron support (with in this case, up to 16 GB of RAM for each cluster's node!), parallelized URPMI (Mandrake's apt-get) and other dedicated tools. This product is based on a one-year research project "CLIC" involving MandrakeSoft and partners. A good snapshot of the product running a 3D real-time demo is available here. The interesting point now: MandrakeClustering's goal is to provide a system which is easy to deploy, easy to administer and use. Well... Mum would certainly love to play Quake with this toy."
But what are people doing with these things in the wild? Is everyone running rendering farms?
Do file/mail servers really need this kind of horsepower (assuming you aren't Google or Yahoo, of course)?
I have been pwned because my
Is it as fast as the new Macs?
Caffeine Good
It's easier to manage an openmosix cluster than this mandrake cluster. Plus you can run diskless nodes using ClusterKnoppix!.
If so, then I think it's a pretty good investment. :)
:)
So much for the "bailing out a failing business, let the market sort it out" mindset. If Mandrake hadn't appealed to the public, they would not have had the $ to come up with this.
Consistent? /bin/sh is /bin/sh wether in NetBSD, OS X or GNU.
I'd consider xmms second only to win amp, dunno what your problem is...
easy OS install? How often do you have to install your os?
clustering research that improves multiplayer gaming well into the next decade? In my mind, that's priceless...
Wow, first /. posting of something Mandrake related in some time. Is it just me, or have a lot of announcements not even mentioned Mandrake recently?
Even though I don't use Mandrake I do wish them the best, especially with a number of Linux distro's competing for the desktop. Hopefully they will make some money from this venture.
StarTux
Uh... thats entirely a 'personal preference' sort of thing.
Heres some stereotypical information about a few distros, though:
RedHat: Corporate Environment
Mandrake: Easy install, easy to use, but has some hard to resolve dependency issues
SuSE: If you are german
Debian: Do it yourselfer
Gentoo: huge 'coolness' factor, a get-it-while-its-still-hot sort of thing.
Slackware: The traditional disto for hard-core geeks
Unless I read the article wrong, these clusters will not be operating on the same level as Windows and OSX. This is designed for research and heavy-duty number crunching, something that XP and OSX aren't. I agree, talented programmers are spending too much time on worthless projects, but this isn't one of them.
The $2,960 is to get support with it.
An OS that surrenders under pressure?
because we all know that render times kick ass when processors have a 12mb/s wifi link between them. What were multi-processor manufacturers thinking building the processors into one unit when beowulfing is available!
(Beowulf clusters have long-term power but slow reaction times. Firewire networking (maybe scsi?) would help with this, but the node->node bandwidth is still an issue.)
Why would Mandrake use Intel's compiler for the Opteron?
You can't judge a book by the way it wears its hair.
It looks like I'm the Apple troll today, so I have to point out that there is a specially designed Xserve U1 rackmount for clustering apps. I think that part of this has to do with the fact that Apple is still used disproportionately often in academia, and part of it has to do with the Apple-Pixar connection. But it turns out that OS X (the server edition, anyway) is a clustering OS. See here for more details.
/bin/sh is /bin/sh wether in NetBSD, OS X or GNU.
/bin/sh is vanilla bourne shell under NetBSD and OSX, but under GNU and Linux it's all too often a symlink to bash. Although bash has very good bourne compatibility, and squeaks past the POSIX shell requirements, it's managed to cause numerous headaches for me when I specify /bin/sh only to get /bin/bash instead. /bin/sh should always be a plain boring bourne shell, and not a link to bash, ksh or another "big" shell. /bin/sh is not going to be used by the enduser, it's going to be used by scripts.
Not really.
A Government Is a Body of People, Usually Notably Ungoverned
I've been using Mandrake for a while, and I would hate to see this company go - they have many great tools, and they put every single line of code they wrote under the gpl (unlike YAST for instance, even though I love SuSe too). They probably need as much PR as they can get, and this was a good idea.
Its time for some features, like their excellent urpm* tools to get more attention (I wonder why it received such scare coverage, for it is the only package management tool that is on par with apt-get among rpm-based distributions - maybe with the exception of apt-rpm). Another great tool, excellent in large deployments is draksync.
Check out these sites:
urpmi mini-howto and easy-urpmi
It ain't a cluster until it has a clustered file system. Until it has that it is nothing more than another high availability solution.
Got Code?
It's a shame there isn't a "-1: Religious War Initiation" mod.
"God, root, what is difference?" - Pitr, userfriendly
But slightly related. :-)
This clustering thing gave me the idea -- pardon me if not original:
Mandrake could sell CPU time! Like those SETI, cancer drug molecule manipulation, code breaking/ prime number finding etc.
We, Mandrake users, could (on a voluntary basis) donate hours of unused computer time to give them what to sell.
Is there a market for this?
Is this possible/viable?
At least, they could use such processing for their distro making related tasks.
Additionally:
They could install automatically (again, on a voluntary basis) bit-torrent servers, so as to reduce their bandwidth bills.
Of course, the usage of such CPU time/peer-2-peer serving would be subject to an agreement. Think here about a list with checkboxes to let the user choose which kinds of content would her machine serve.
This time, I gave my 2 cents.
I am the main programmer of the 3d demo shown on the screenshots. This demo combines a distributed interactive fluid simulation with graphical intensive rendering using pixel shaders. It uses Net Juggler to manage the distribution on the cluster. You canse some picture and a video of the fluid simulation in our gallery . The shader for the rabbit's fur is explained on this page of my website (please be gentle with the server...).
Here it comes... I'm going to vent. How is that ./ readers (clearly amongst the most well-educated of the 'net surfing masses) always fall for the bigger == better claim? First of all, Quake, or any computer game that I can think of, does not have multi-node support. You have to rewrite the code to support the PVM (Parallel Virtual Machine) message passing library, for example.
And it's not just with clusters we seem to have this problem. Let's look at Apple computers and the G5. A 64-bit machine will simply allow you to add/subtract/multiply/etc really big integers faster. How often do think you use numbers that large? Encrypting or decrypting emails is the only thing that comes to mind for even an "above average" user. Unless you run software that supports multiple processors such as web and database servers or want to watch a DVD while you play Quake, a multiprocessor machine is not for you. The relative performance increases are negligible. It's simply a marketing ploy that allows Apple to make a high profit on the low volume of units they ship. They throw an additional $300 uP in their computers and charge an extra $1000!
But in Apple's defense, their displays rock. Any luck getting one of those running on a PC? I'm done ranting. Have a nice day.
Imagine a beowulf cluster of MandrakeClusters!