from the junior-staff-gets-the-raspberries dept.
Cheap Supercomputers: LANL has 750-node Raspberry Pi Development Clusters
One of the more esoteric announcements to come out of SuperComputing 17, an annual conference on high-performance computing, is that one of the largest US scientific institutions is investing in Raspberry Pi-based clusters to aid in development work. The Los Alamos National Laboratory's [LANL] High Performance Computing Division now has access to 750-node Raspberry Pi clusters as part of the first step towards a development program to assist in programming much larger machines.
The platform at LANL leverages a modular cluster design from BitScope Designs, with five rack-mount Bitscope Cluster Modules, each with 150 Raspberry Pi boards with integrated network switches. With each of the 750 chips packing four cores, it offers a 3000-core highly parallelizable platform that emulates an ARM-based supercomputer, allowing researchers to test development code without requiring a power-hungry machine at significant cost to the taxpayer. The full 750-node cluster, running 2-3 W per processor, runs at 1000W idle, 3000W at typical and 4000W at peak (with the switches) and is substantially cheaper, if also computationally a lot slower. After development using the Pi clusters, frameworks can then be ported to the larger scale supercomputers available at LANL, such as Trinity and Crossroads.
(Score: 4, Insightful) by fadrian on Wednesday November 15 2017, @02:56PM (4 children)
That being said, the notion isn't particularly innovative - people have been developing code on machines slower than ones that will ultimately run the code for cost savings since organizations could afford to have more than one computer.
That is all.
(Score: 2, Interesting) by khallow on Wednesday November 15 2017, @03:24PM (3 children)
It appears that each node has four Pis [bitscope.com] on it. One possibility that stands out right now is using the nodes for high I/O work, perhaps for certain sorts of real time signal and image processing.
(Score: 2) by frojack on Wednesday November 15 2017, @07:04PM (2 children)
Of course this being a developmental platform, the real intent is to prototype software.
But the spin off from BitScope is likely to find a hobbyist market, as well as programmer training.
But not at those prices. At $500-$800 they would get snapped up in large numbers, even if they only had 4 Pi's on board.
Building these in quantity is going to bid the price of Pi's up a lot, initially, but then the price should drop substantially below the current price.
The Pi guys, especially Eben Upton, CEO have said they manage Pi production very closely so as not to get stuck with unsaleable inventory since there seems to be a new pico-board computer released every month by various vendors.
Doing this they have managed to keep the Pi price of older models unusually high compared to the newer models.
No, you are mistaken. I've always had this sig.
(Score: 3, Informative) by khallow on Wednesday November 15 2017, @08:28PM
(Score: 3, Informative) by t-3 on Wednesday November 15 2017, @09:42PM
Hardkernel (odroid), recently announced a similar offering - http://www.hardkernel.com/main/products/prdt_info.php?g_code=G150152508314 [hardkernel.com] 8G ram, 32 cores, 220$.
(Score: 1, Funny) by Anonymous Coward on Wednesday November 15 2017, @03:30PM (9 children)
Times of Israel [timesofisrael.com]
(Score: 4, Touché) by khallow on Wednesday November 15 2017, @03:41PM (1 child)
(Score: -1, Redundant) by Anonymous Coward on Wednesday November 15 2017, @04:47PM
Whoosh!
(Score: 2) by c0lo on Wednesday November 15 2017, @03:58PM
A hat tip to hot grits.
(if you are young, unafraid and don't-know-what's-about, head to Urban dictionary [urbandictionary.com])
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 2, Funny) by infodragon on Wednesday November 15 2017, @06:34PM (3 children)
When I read Natalie Portman and Slashdot refugees I started imagining Beowulf has something to do with this cluster. That twitch has started up again....
Don't settle for shampoo, demand real poo!
(Score: 0) by Anonymous Coward on Wednesday November 15 2017, @07:31PM (1 child)
How is it not a Beowulf cluster?
(Score: 0) by Anonymous Coward on Wednesday November 15 2017, @09:19PM
I know. Image that . . .
(Score: 2) by Geotti on Wednesday November 15 2017, @10:53PM
As long as we steer clear of them sharks on the intertubes, we'll be fine!
(Score: 0) by Anonymous Coward on Thursday November 16 2017, @04:08AM
Imagine a beowullf cluster of..... what?!!!
(Score: 2) by Bot on Thursday November 16 2017, @01:26PM
The announcement is date 7 nov.
Which means the award has been given likely the 6 nov. Not the remember remember fifth of november, by one friggin day.
Is this the way to say thanks for the V for vendetta depicting of high ranking prelates? Very disappointing.
Account abandoned.
(Score: 5, Interesting) by leftover on Wednesday November 15 2017, @05:05PM (2 children)
Just wondering why they used self-standing RPis rather than RPi Compute Modules which are made for this. They are powering 750 sets of I/O gear they will never use. Takes up more space too, although that might spread the heat dissipation.
Bent, folded, spindled, and mutilated.
(Score: 5, Interesting) by richtopia on Wednesday November 15 2017, @05:20PM (1 child)
I suspect it is for the on board network port. Also at this scale size probably isn't an issue, as you need some air just for cooling. And the accessories that aren't used can be powered down: most low-power distributions for the Raspberry Pi have built in routines for turning off unused hardware.
I am a little surprised they did go with the Raspberry Pi instead of other hardware, there are many other options that are either faster or cheaper, and almost always with improved networking. I typically recommend Raspberry Pis to new users because of the community support, but if you are building a cluster with 150 SBCs you probably can spend a few hours dealing with a less popular board.
(Score: 0) by Anonymous Coward on Wednesday November 15 2017, @08:58PM
I am a little surprised they did go with the Raspberry Pi
...as opposed to an x86-compatible architecture?
Perhaps they heard about this:
In ARM-related news, on Monday the 13th, Red Hat made available an ARM port of Red Hat Enterprise Linux.
RHEL now ready for power-efficient server-grade chips [theregister.co.uk]
-- OriginalOwner_ [soylentnews.org]
(Score: 2) by crafoo on Wednesday November 15 2017, @05:42PM (5 children)
Why not any number of more capable and cheaper alternatives? Is it because anything with RPi immediately gets tech news coverage for some reason?
(Score: 3, Informative) by DannyB on Wednesday November 15 2017, @06:24PM (2 children)
Why?
Because RasPi's don't have a Management Engine. Can you even trust that binary you compiled on box with Intel inside?
But more likely . . . because a 750 x RasPi box is way cheaper than a real supercomputer. Yet it lets you test all of the same programming concepts. Even the same binaries if you use Java. Same source if you use Python or Lisp. And just a recompile away for most anything else that works on Linux.
So if you've got developers building a massively parallel computation program, how do they test it? On a real (eg expensive) supercomputer? Or on a 750 node RasPi fuster cluck? The RasPi seems a good way to prove that it is ready to try on the more expensive hardware.
When will the TLAs get management engines into ARM processors?
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 2, Informative) by Anonymous Coward on Wednesday November 15 2017, @07:39PM
>Because RasPi's don't have a Management Engine.
That we know of.
>Can you even trust that binary you compiled on box with Intel inside?
Broadcom is so open... [ycombinator.com]
(Score: 2, Informative) by Anonymous Coward on Thursday November 16 2017, @01:25AM
No blob, no boot. Reverse engineering in progress but far from complete https://github.com/christinaa/rpi-open-firmware/blob/master/README.md [github.com]
Oh, BTW, many modern ARM implement TrustZone. https://en.wikipedia.org/wiki/ARM_architecture#TrustZone [wikipedia.org]
The only informative part is the one about using cheap slow parts as a prototype, like testing first ideas of a ship hull at reduced scale in a big pool
(Score: 2) by VLM on Thursday November 16 2017, @02:17PM
Familiarity and ubiquity. Not mentioned is you're supposed to start troubleshooting on your own individual board to get the code running at all, then try working code on the 3000 core monster.
And its not just troubleshooting your project's code, but all the support stuff and libraries and compilers you're using.
Everybody's got a pi...
(Score: 2) by VLM on Thursday November 16 2017, @02:26PM
Also rapid development and portability and marketing (but not marketing like you'd think)
I was motivated enough to visit their website... WRT marketing, I've seen bigger more power hungry things shipped to trade shows. Its only 6U rack high, but looks very deep or weirdly photoshopped. I could see anyone in the supercomputer ecosystem selling "something" using hardware like this for demos at trade shows and sales events. Look at my cool WTF, and I can demonstrate it on this 144 node system that easily fits on a single handtruck and plugs into a wall.
I played around with stacked clustered desktops around the turn of the century and learned a lot, but its a very LARGE hobby. Eight large desktops (not this modern HTPC stuff) stacks up into a tippy stacked tower and its only eight machines.
(Score: 2) by stormwyrm on Thursday November 16 2017, @01:26AM (1 child)
You'd need some really fast interconnects for most serious supercomputing applications. The Raspberry Pi 3 has a built-in 10/100 Mbps Ethernet, which is pretty damn slow for as far as that goes. There's 802.11n wireless on the Pi 3 too, but I seriously doubt that you can really get the nominal maximum of 150 Mbps reliably. Unless they have a custom USB 2.0 protocol that could do 480 Mbps, but it doesn't look like it. In contrast, Infiniband, widely used for supercomputing interconnections, does at least 2 Gbps and can go into the 100+ Gbps range. I'd have tried using the ODROID C2 instead, which at least has an honest to goodness gigabit Ethernet and isn't that much more expensive.
Numquam ponenda est pluralitas sine necessitate.
(Score: 1) by kai_h on Thursday November 16 2017, @08:50AM
This cluster is not designed to give you the performance of a supercomputer, but rather as a test bed to run code on to see how well it parallelises on a cluster. It's like a cluster with it's L Plates on.
Processing speed & interconnect speed are secondary considerations over allowing researchers to see how their code handles a large number of nodes.