Slashdot Mirror
IBM (Offically) Launches Linux Box Clustering
Neumsy writes " IBM has offically announced that it is releasing Linux-Based Server, Software Packages. Yahoo! News has the story . According to IBM, this will expand the use of Linux. It's a good overview article. Not too much in depth, but still nice to see Linux getting out there more and more."
I doubt it, unless its a cluster made by SGI ;)
..
(You could do that with the Onix)
Because most of the stuff is done by GFX card
Before you email me, remember: "There is no god!"
Interestingly enough, just this morning I saw a two-page ad for IBM servers running linux. I haven't found the actual ad online, but it showed the famous "bigfoot" photo, labeled as fake, and then a penguin walking through the server room in the same pose, labeled as real. The other page of the ad was an abbreviated list of the usual Linux myths that we all know and love, with IBM-specific arguments as to why these were no longer true. This is the real fruit of the $1 billion campaign from IBM, and a great answer for your hesitant management.
IBM's main page for this, aimed at upper brass rather than engineering, is at http://www.ibm.com/linux/cio2, and the myths seemed to come from this brochure: http://www.ibm.com/linux/Demystifying_Linux_Brochu re.pdf.
Maybe this isn't entirely on-topic, but I thought it was a great example of some more of that good mindshare. And this time IBM isn't going to have to scrub off any sidewalk paint :)
Your right to not believe: Americans United for Separation of Church and
In the corporate market, most Linux installations are going to come in as X servers and thin clients (see ltsp). This has huge advantages for maintenance and support, as well as hardware upgrade paths.
In this case, wine, or other emulators, are not necessarily the best solution for legacy windows apps. Look at using windows 2000 terminal server to host the windows apps and rdesktop to access them from the clients. Very simple to integrate and maintain. The drawback is the cost of the seats on windows 2000 terminal server - but that cost will certainly be offset by the cost orf managing and debugging an emulator.
Of course, terminal server / emulator is to be thought of as a temporary solution until the legacy app is ported to something which is client independent.
http://access.ncsa.uiuc.edu/Releases/011018.Titan
Top500 org (http://www.top500.org) has the latest rankings out and Platinum is ranked #41 in the world with a 594 rating and Titan is at #34 with a 677 rating. These are not slow systems, but they do require a knowledgeable support staff. Both systems are IBM "out-of-the-box" clusters, running RedHat Linux versions.
I'm just glad to see linux advertised in the mainstream media. Name recognition helps.
More information is available at the main website: http://ncsa.uiuc.edu
The Internet has no garbage collection
Well...
The main factor in determining your q3 performance is not processor speed, but the bandwidth between the processor and video card (and the video card and the actual display buffer).
Even if you could thread quake through a cluster, it would make no difference.
As for threading.. these clusters are not hte same thing as a multirpocessor machine... unless they implement shared memory and such across the network.. which would be slow.
Clustering is such a fascinating area on its own and the article is so shallow that I'm curious why they've published it at all. They could mention the potential benefit of cluster computing as well as examples of some working clusters like Beowulf or Mosix or even the famous fact that there is a cluster among top 500 supercomputers.
The announcement may make it 'official', but IBM has been selling turn-key style clusters for over a year now.
How do I know, and why do I care? 'Cause I work for IBM, and design and build Linux clusters for a living.
We rack purpose-built 1U's (x330's) built for easy large-scale mangement (built-in daisy-chain KVM capabilities, integrated service processor network, cool blinky lights, etc, etc), have internally developed mangement software and system imaging solutions, and ship them to customers at a point where all they need to do is plug them in and log on.
The only real issue is almost no one has a 'standard' cluster application. Almost every one uses a different IPC mechanism, and usually an app is only validated against one very specific software image. So to get the most out of a cluster, customers still need to spend time tweaking.
Maybe someday soon, we'll all be able to 'apt-get install damned_big_cluster', but until then... <shameless>just call IBM.</shameless>
--Matthew
slashdot at sigalrm dot com
Clusters are clusters, but not all clusters are equal.
You have high speed compute clusters like Beowulf.
Then there's the first type of High Availability clusters -- the hot standby/failover configuration, where services and storage on one system are reassigned and restarted by another if the first goes away. Most Unix and Linux implementations haven't got beyond this yet.
Then there's the more grown up version of High Availability clusters, where all the clustered systems have concurrent access to the same storage, cluster wide lock management, and can run multiple cooperating instances of the same application on all systems. Like Oracle Parallel Server. VMS pioneered this ; Tru64 Unix has it now, and Linux is working towards it with GFS (the Global FileSystem).
GFS has got to be one of the most exciting current Linux developments in my book. I've had a taste of this kind of clustering on Tru64 Unix, and believe me, once you've experienced it you don't want to go back.