Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : All About Sun Microsystems -- Ignore unavailable to you. Want to Upgrade?

To: dmf who wrote (21727)	10/25/1999 1:36:00 AM
From: JC Jaros	Read Replies (3) \| Respond to of 64865

What does Sun think/do about this? Thanks, dmf My understanding is that 'supercomputing' isn't particularly an area of business interest for Sun. After all, they left that division of Cray to SGI and walked off themselves with more meaningful goods. Beowolf is cool. Cooler than that would be an UltraSparc cluster. The coming generation Ultra/Solaris scales exponentially better than Linux, particularly Linux on Pentium. These beowolf installations are for lower budget special (number crunching) projects requiring a specific type of processing horsepower. It's my understanding that these (Beowolf) are not typically solutions that compete with Sun. -JCJ

To: dmf who wrote (21727)	10/25/1999 2:07:00 AM
From: QwikSand	Read Replies (4) \| Respond to of 64865

What does Sun think/do about this? (Note: I wrote this overly long reply while JC was writing his short one that says basically the same thing, but what the hey, I wrote it so I'll post it anyway. That way I do't have to ask JC to analyze my feelings of worthlessness.) I think they try to hire the graduate students before Microsoft gets to them. On my home computer I have the "Intelligent Life@HOME" screen saver from UC Berkeley. Whenever my NT machine is idle, this program runs as the screen-saver and does a little more scientific number crunching on a piece of data that some radio-telescope collected, looking for patterns that hint at intelligent life in outer space. Depending on how much I use my machine, it takes a day or two to finish crunching one chunk. Then it sends the results back to Berkeley over the net, gets another chunk, and starts on that. I've done about 80 chunks so far. I got this program after I read that the distributed computer that Berkeley had cobbled together out of idle cycles in the world's spare bedrooms was the fifth fastest supercomputer in existence, or some such thing. Faster than anything Sun builds, by some metric or other. Should Sun be worried about the competition? Not IMHO. Same with geowulf. It's a lab project that can parallelize a certain specific kind of number crunching, where you have to respond 'fast' relative to some computing alternative but not in terms of human response time. You have to respond in hours instead of days, or days instead of months. You notice that the article said that it was "the future of scientific computing". This cluster isn't a server or a transaction processor. It crunches a lot of numbers. I do believe that a big refinement of this architecture is the future of all kinds of computing, out somewhere 10-15 years from now. Since bandwidth changes everything, it must change what operating systems are, what storage is, etc. As bandwidth goes to infinity, the concept of data having a location vanishes so you can no longer tell the difference between an SMP and a cluster and a network. But though Linux-lovers everywhere will protest the contrary, these lashed-together PC's aren't yet ready to challenge industrial-strength servers that have to process lots of transactions quickly while keeping huge databases consistent. This is because interprocessor communication is something they can only do infrequently (in computer terms). If they did it frequently the overhead would swamp the device. The modifications to the O/S are not that major (so the Beowulf people say), and the speed of the interconnects that tie the pieces together is whatever the builders want it to be (i.e., slow). I haven't looked at the details of how this machine is used of course, but I'll bet you assign each node a chunk of data, it tells you when it's crunched, then you assign it some more, just like my screen saver. Transaction and server-type processing, on the other hand, depends on starting and stopping large numbers of interconnected small activities efficiently, exactly what these machines are not set up for, and what a good SMP is set up for. All computer companies are working on clustering. What Sun needs to do is to cluster its SMP's with fast interconnects and sophisticated operating software so that a business can start with one or two powerful boxes and incrementally add varying size chunks of efficient compute power and redundancy, whether it's by adding processors to a box, more disks, or more boxes. I think that's exactly what they're doing. And as time moves forward and clustering evolves into the future of all computing, Sun's products will evolve with it. But it's going to take a while. I'm sure James will put down his spatula long enough to argue that the real way to do it is to make all applications distributed, etc. Maybe, but...we're a long ways from that too (and this post is too long too). So IMHO, I don't see much Sun should take away from this article except "Recruit at Princeton"...and they probably already knew that. Regards, --QwikSand