Slashdot Mirror
Apple Releases Xgrid Technology Preview 2
dark_lotus writes "Apple has announced the availability of Xgrid Technology Preview 2. This version improves on Xgrid's breakthrough ease-of-use by adding the most requested features, including an 'xgrid' command-line utility, support for MPI jobs, and a comprehensive Xgrid User's Guide, as well as numerous bug fixes. Groovy!"
I hope apple does to grid computing what they did to local subnet computing. Rendezvous is an awesome technology for finding people nearby, or doing any simple/quick home networking.
- tristan
Most likely, the other distributed computing entities will analyze XGrid and make their products better by incorporating new Apple technologies. Just like every other industry has done when Apple comes out with something new.
Truthfully, the applications are different. SETI and the like are analyzing predetermined/presegmented bits of data, while XGrid is targeted as more of a local (intranet), real-time distributed computing application. Agree/disagree?
***
When you can use xgrid and enable any type of grid enabled program that might use Rendevous and some simple plug in of some kind to use however many Macs there are on a given network to crunch away at something is the day Apple will start making serious inroads again. Imagine a Photoshop Mac pool at a Ad agency using Xgrid, or the same thing at the movie making place with Final Cut Pro 5 or whatever.
I really do like the directions that Apple is going in these days. Stock market does not seem to mind either. :)
Still, it should be interesting to see how this could affect the rendering crowd. Imagine being able to use a program like Maya then when everyone goes home at night, use all their workstations to help process a render job. That could save a lot of businesses a lot of time and increase their profits. I know because we have a 100 CPU render-farm we rent out to local businesses so they can get a jump on their next business.
"The problem with socialism is eventually you run out of other people's money" - Thatcher.
I highly doubt that is there goal. I doubt there are that many consumer applications for it.
However, I can see it as being very useful for educational institutions (both higher ed. and K-12 in the U.S., as well as their international equivalents) as well as small media and software developers, the sorts who could make some usage of distributed apps but not have the funding for a full-time sysadmin to run the thing. And of course, that's a selling point for large businesses as well, since lower admin costs = firing IT staff = salary increase for the CEO and higher stock prices for the do-nothing class.
fuck you.
I think you will see new uses that haven't been thought of. I can't give an example because I haven't thought of them yet. ;-)
It would also make current tasks even better. For example:
You would be able to fork a new process transparently to under used machines. The OS would know which machines were under-utilized (the iMac someone's kids use when at home, for example) and if it was maxed out, it would send the process to that other machine. All of it transparent to the user.
xgrid is great for *applications* usage *now*.
Apple will incorporate it into the OS itself and it will become even more useful when it makes usage of the CPU cycles (disk transfer perhaps too by sending disk bound processes to another machine) available to anyone on the network who needs them without user interaction.
Other uses:
* iMovie and iDVD encoding farmed out to other machines on the network.
* The Mac equivalent of MythTV (or TiVo or ReplayTV) uses other machines to encode if one is busy.
XGrid is an extremely interesting project, but it's not designed to take on a dedicated, custom-designed cluster like VT's Big Mac.
Some calculations can be split into pieces that don't require much "talk" with other pieces. For example, Apple's Mandelbrot demo--you don't need to know what's running on other processors.
OTOH, many problems require quite a bit of cross-talk with other processors. For example, most of the quantum chemistry calculations I run require calculating big integrals. These are run across multi-proc boxes or clusters, but the speedup depends a *lot* on the latency of the network. So XGrid won't really help here--most of the ad-hoc networks serviced by XGrid would have something like 100MBs Ethernet, which is slow.
I'm willing to put up $$ to use supercomputing centers like VT's Big Mac because they're *designed* to handle hard-core parallel number-crunching. Right now, I'm running jobs on a 24-proc POWER3 cluster with 4GB RAM per processor. (Yes, the extra RAM really helps too since I don't hit the hard drive much.)
I think XGrid will see a lot of use for academic or corporate environments to allow adhoc clustering. As an example, I can run some calcs on an XGrid "cluster" at night on all of the desktop Macs in a lab or across an office. These won't be anywhere near as fast as a well-designed cluster. But it will give me access to "untapped" CPU cycles.
Here are some interesting articles which I've seen today:
Xgrid: High Performance Computing for the Rest of Us Apple's paper Xgrid example: Parallel graphics rendering in POVray Here's an example which slashdotters should appreciate-- Tim Buchheim
Yes the problem with standard TCP/IP is it isn't bootstrapping enough for a home user to plug two or three computers together and see them all on a network without configuring a lot of stuff. Typical client server tcp/ip apps require you to know the address or name of each server, or at least to run an SLP server.
I don't know why Apple haven't released an OS X version of the OpenMosix project, which works wonders on Linux. It moves processes automatically between nodes while they are running, automatically re-routing disk and network access, and copying memory data across. Needs tricky work on the kernel, but combined with Rendezvous technology, it could be a killer. Your heavy tasks would be automatically routed around the workgroup, as and when is appropriate, even if they are only half-way complete.
Some years ago, I asserted that DHCP basically had no good reason to exist. Any time a machine was going to be on my network, I wanted to know about it and explicitly handle its placement myself, rather than just having things reconfigure themselves willy-nilly.
Predictably, I've now changed my mind about that for many environments. If I were running a network of a thousand workstations, I'd much rather deal with the small chance of one of them doing something inappropriate than configure them all manually.
I have a guess that you may undergo a similar change in thinking about the appropriateness of Rendezvous and/or Xgrid. When it's an unusual task that only gets handled in small and exceptional circumstances, it seems best to handle it explicitly. When it just becomes part of what normal computers do all the time, it seems unthinkable to handle it manually.