http://gizmodo.com/5120924/researchers-create-web-skeleton-key-with-200-ps3s

The cost of this cluster could be reduced. Remove DVD-Drive, use MMC instead of hardisks and a different casing structure. Anything else?

I wonder if the Sam440ep could be clustered like this? Amiga OS4 on CELL or multi-core PPC will be breathtakingly delicious.

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@meet.mrnrg

There's no clustering solutions for AmigaOS I'm aware of. Under Linux, sure, Sam might be clustered (I just don't see why)

_________________
OpenWindows Initiative. Port PS3 hardware to bananas. For free. Join today and receive expired $50 cupon from AI!

@meet.mrnrg

Why remove the DVD drive ? theu cost nothing compared to the rest of the lot...

_________________
Theres a time to live and a time to die
When its time to meet the maker
Theres a time to live but isnt it strange
That as soon as you're born you're dying

@meet.mrnrg

Doesn't take my breath away:

1. ps3 offers more cpu power density

2. AOS4 isn't a proper server os. Noone who needs cluster will manage to do anything useful with it: no remote connectivity, jobs scheduling. It's a long way to go beyond multicore capability to become clusterable


Jack

_________________

"the expression, 'atonal music,' is most unfortunate--it is on a par with calling flying 'the art of not falling,' or swimming 'the art of not drowning.'. A. Schoenberg

Well the 440 has certainly been clustered before, for instance in the No. 4 supercomputer in the world.

That said it's a highly customised version of the 440 and there's 212,992 of them...

_________________
Whyzzat?

@jack

1: Well. The PS3 is 7.2 times as large as a bare mini-ITX board. So the CPU power density isn't THAT much better. Better, yes. But not like infinity or something

2: Why would you run a server OS on a computing cluster?

Not to say I don't think the PS3 cluster is a better idea than a SAM cluster, but give me a setup with 2-2.5 times more CPU power in a mini-ITX board (and PCI Express for communication), and I will build myself one seriously disobedient donkey cluster.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@Metalheart

Did you see the picture? Looks like its a waste of space in that design!

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@opi

Quote:

opi wrote: @meet.mrnrg There's no clustering solutions for AmigaOS I'm aware of. Under Linux, sure, Sam might be clustered (I just don't see why)

IIRC, Lightwave had its own custom clustering software and Amiga OS 4 has a dist-cc port which sort of like a clustering solution, but that's it; there are no general purpose clustering solutions. We'd need PVM or MPI to be ported before Amiga OS could be used in a computing cluster.

Hans

_________________
Join the Kea Campus - upgrade your skills; support my work; enjoy the Amiga corner.
https://keasigmadelta.com/ - see more of my work

@olegil

Quote:

2: Why would you run a server OS on a computing cluster?

It's just semantics in sticking "server" to "OS". It means having means to to control an schedule jobs, remote connectivity.
The task is ideal for Unix/Linux which are inherently tailored to be run on servers.

As to density, when you add power supplies and cases, you'll have to create Sam blades to beat PS3.
It's doable hardware-wise but needs some engineering work.
But taking stock Sams inside their own cases+psus will create hardware overhead.

Jack

_________________

"the expression, 'atonal music,' is most unfortunate--it is on a par with calling flying 'the art of not falling,' or swimming 'the art of not drowning.'. A. Schoenberg

@jack

Uhm, use pico-psu's and boot them over nfs or similar. two pieces of plastic per board makes them slide into a rack, 12 per box. I'll design the mechanics for that in an evening.

It's not SAM-specific, though. so keep the OS discussion away from this idea... Am currently looking at the fastest epia system, it's a 1.7GHz C7 with SSE2, onboard GPU and PCIe which I could use to do the clustering. With non-transparent PCIe bridges all the machines would be talking directly into the system memory of the main controller board. Or alternatively just chicken out and use GigE like normal, sensible people do. It ain't as fast, but it does get the job done

All it would need is one monster of a 12V DC PSU, and I can build that, no problem

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

http://distributed.amiga.org/

26-Nov-2008 New beta clients are out for Cell BE (OGR-NG) and nVidia CUDA (RC5-72 only), get them at the usual place. Meanwhile, we are doing very well in OGR-26 and have risen to a 4th overall team rank. Over 13% of the work has been completed already, at current rates OGR-26 will be finished in possibly 4 months' time.
02-Nov-2008 Distributed.net now has statistics up for OGR-26 as well, after one week we are ranked 7th :)
30-Oct-2008 The AmigaOS4/PPC client build 507 has been updated (to 507b) since it crashed on non-AltiVec/G4 machines (compiler used AltiVec instructions in general code without checking).

1) Is any of their ideas relevant to this topic?

2) Can their software help Amiga achieve anything similar to a clustered environment?

3) How can http://boinc.berkeley.edu/download.php help Amiga? Is there anything for the Amiga in this space?

4) How can we encourage more of this activity in-terms of development and donating of CPU or GPU time?

Last edited by meet.mrnrg on 05-Jan-2009 at 10:55 PM.

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@meet.mrnrg

1: sort of. distributed.net is a kind of clustering software, as it distributes workloads between different systems. But not really like a uniform cluster like the cell cluster you linked to in the beginning.

2: not really

Look up beowulf cluster on wikipedia. That should give you a good start

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

I added point 2 & 3.

Ok you mention: http://en.wikipedia.org/wiki/Beowulf_cluster

can AmiCygnix be setup to use this? There is a lot of cross-compiling going on now on the Amiga, can a new branch of code be set-up to make it more native?

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@meet.mrnrg

Quote:

meet.mrnrg wrote: @olegil I added point 2 & 3. Ok you mention: http://en.wikipedia.org/wiki/Beowulf_cluster can AmiCygnix be setup to use this? There is a lot of cross-compiling going on now on the Amiga, can a new branch of code be set-up to make it more native?

OpenMPI would have to be ported. I don't think that AmiCygnix would be much help since it is a parallel processing library/system, not an application with a GUI.

One problem is that OpenMPI's documentation looks incredibly limited, so it means that people will need to figure everything out on their own.

Hans

_________________
Join the Kea Campus - upgrade your skills; support my work; enjoy the Amiga corner.
https://keasigmadelta.com/ - see more of my work

@meet.mrnrg

Well, this is akin to what BBRV claimed Genesi would do with the non-existent 'Pegasos Blades.' (IIRC, based on Efikas or similar, reasonable RU density).

It's not exactly a 'new idea' to make a cluster, but with SAM - why? Performance per dollar spent isn't there with current pricing.

For 'real use,' it had better mount into a real server rack, with a single power supply, and not 'build your own.' Then, you still don't have GigEthernet, although dedicated fiber or 10Gb/sec is used more often for 'performance' computing clusters.

As mentioned, MPI or similar would need to be ported, unless you meant to run Linux on it. With Linux, there are better bang for the $ solutions out there. With OS4, still not high bang for the buck, and you'd also need to port MPI or similar (there are some others out there, that are more 'complete' packages' - but MPI is generally what's used to handle the scheduling and marshalling of computing nodes, or at least was when I was writing some distributed/clustering code.

Then, there's general lack of strong thread performance, which means you'd have to write the app via AOS tasks - once MPI is ported, it's really not just 'throw some job at it, and it's magically parallelized and performs because it's on a cluster.'

There may be some level of compiler or tool help there, but what I ran into is probably similar to what PS3 devs go through - some tasks lend themselves to be split up into highly paralleliseable tasks, and others...not so much. And either requires work, as in, re-design if it's meant to be some specific application (and you still *do* have an application if you're simply trying to crack some latest algorithm, coding genomes, whatever).

distributed.net is sort of a fairly basic, long running, cluster. Each 'node' is a client system, that gets a single batch job to process until done. It then returns the results, and the distributed.net server(s) hands out the next batch, which generally takes 12 or more hours to complete. The app I'd imagine is relatively simple - no consideration as to performance of each 'node' (client system) is used, but instead, a pre-determined amount of operations are given to each and any client, and very little data marshalling is required, in that few jobs/tasks are waiting on the completion of some other node's work...which is *not* always the case.

Other tasks may not be so long running, but again, what are you going to use it for, why would it possibly be done on a SAM vs other cheaper/bang for buck systems, and why would you possibly not run such a cluster on Linux (or BSD..) ?

I don't see any compelling reason there, just another random question from you..

_________________
Are we not done with the same silly arguments and flames yet??!

@wegster

This question will lead into to a more complex SWOT analysis.

For others, this data might be more interesting:
http://stats.distributed.net/team/tmsummary.php?project_id=26&team=200

Quote:

minator Well the 440 has certainly been clustered before, for instance in the No. 4 supercomputer in the world. http://top500.org/system/8968 That said it's a highly customised version of the 440 and there's 212,992 of them...

Any more answers? thoughts? questions?

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@meet.mrnrg

Quote:

meet.mrnrg wrote: @wegster This question will lead into to a more complex SWOT analysis. For others, this data might be more interesting: http://stats.distributed.net/team/tmsummary.php?project_id=26&team=200 Quote: minator Well the 440 has certainly been clustered before, for instance in the No. 4 supercomputer in the world. http://top500.org/system/8968 That said it's a highly customised version of the 440 and there's 212,992 of them... Any more answers? thoughts? questions?

The first link, you're a late-comer. Most of us running OS4 have at least run dnetc for a while ourselves. PPC, moreso with AltiVec (also Cell/SPEs, moreso), has always been decent at crunching numbers.

However, it's still a bang for buck scenario. I'm unsure you read and understood the amount of software work required I mentioned previously, and even assuming you did that, why would you buy a $$$ bunch of SAMs to throw together manually into an ad hoc blade or rackmount 'solution,' when:
1. SAM pricing is > reference board or similar other low power PPC boards
2. You can buy Cell blades from IBM, and run Linux with proper tried clustering software.

Not to mention you're going to also have to accommodate for cooling, and PCI based Gig Ethernet or fiber isn't exactly ideal speeds, as the bus is seriously limiting (I'd put $ that noted cluster certainly had GigE or better, and/or PCIE).

?

I can tell you for a fact on the *very* rare occasion I've ever had a problem with a cluster node, I've had replacement overnight, and it back into operation within 24 hours. Do you really think ACube can offer that, or you'll manage to put it *together* in that amount of time?

There's a saying, 'use the right tool for the job.' I don't believe SAM or OS4, is the right one for this job, as vaguely 'specified' as it is at this point.

Last edited by wegster on 06-Jan-2009 at 01:31 AM.
Last edited by wegster on 06-Jan-2009 at 01:30 AM.

_________________
Are we not done with the same silly arguments and flames yet??!

@wegster

So you won't be working on this thread then.... cause it useless....

One day this space will be much better, but only when more people question and experiment in this space.

Not interested in price! no interested in best solution! A version 0.1 in this space, would be neat.

_________________
Quote:

Easy Pocket Money, Freelancers & Experts Online

MiniMig FPGA, Sam440 Flex 733Mhz PPC, Amiga OS 4.1 Update 2, MorphOS 2.4, Other - AmiKit + Cloanto Amiga Forever 2008 + E-UAE, AmigaSYS

@wegster

Quote:

wegster wrote: As mentioned, MPI or similar would need to be ported, unless you meant to run Linux on it. With Linux, there are better bang for the $ solutions out there. With OS4, still not high bang for the buck, and you'd also need to port MPI or similar (there are some others out there, that are more 'complete' packages' - but MPI is generally what's used to handle the scheduling and marshalling of computing nodes, or at least was when I was writing some distributed/clustering code.

What more complete packages are there? Are any open-source? MPI seems to be prevalent when it comes to writing parallel software, but details are really thin on how it's distributed and how one could monitor a cluster.

Quote:

I don't see any compelling reason there, just another random question from you..

The only reason I could think of is to do it because you can and/or feel like it. I wouldn't mind putting together a computing cluster from components that aren't designed for it, just for fun. In fact, I did have a look at doing something like this a few years ago, but:
- porting OpenMPI would have been too much for me
- Learning MPI after porting it would take even longer
- I simply don't have the cash to put together even a two CPU cluster, whether it's an x86 board or PowerPC

Obviously if a company needed a cluster, their choice would obviously be cheap x86 boards.

Hans

_________________
Join the Kea Campus - upgrade your skills; support my work; enjoy the Amiga corner.
https://keasigmadelta.com/ - see more of my work

@meet.mrnrg

Quote:

meet.mrnrg wrote: @wegster So you won't be working on this thread then.... cause it useless.... One day this space will be much better, but only when more people question and experiment in this space. Not interested in price! no interested in best solution! A version 0.1 in this space, would be neat.

Ok, so then noting such as 'wouldn't it be cool, just for the heck of it' versus any *actual* benefit in doing so..might make the thread read differently.

So many of your (and others) are like 'ooh, wouldn't this make Amiga/OS4 be number one!' or along those lines, when there simply isn't any benefit in the real world for some of these err, 'suggestions.'

Sure, it would be neat, for someone that wanted to spend much more, for much less. But, I'm not sure how good that would be towards 'promoting Amiga' ?

_________________
Are we not done with the same silly arguments and flames yet??!