There are millions of untapped processors out there

[Ivy Mike]

I was patching in a group of lines for a meeting at my place of employment at a major Las Vegas, NV convention center/hotel and it occurred to me that the switches we use sit idle most of the time yet run 24 hours a day. These switches literally have nothing to do but answer broadcasts from each other informing everyone else that yes they are up and no there is nothing going on.
What if we could have these switches do something useful?

Granted, we aren't talking a lot of computing power here but the switches I most commonly use are HP ProCurve 5400ZL chassis which include a Freescale PowerPC 8540 in the management module and an ARM9 in the Gigabit module and another ARM9 in the 10G module if so equipped (our core switches are so equipped)
While I understand from my days lurking here that the ARM and Power lines are not high priorities due to their low speed and processing ability, I know that we have more than 100 switches in my building that do absolutely nothing for more than 12 hours of the day. The utilization loads on these switches are low during the day as well unless we have a large convention in the building. We also recently installed 22 HP Procurve 3500yl 48 port switches which also contain the same chips as the 5400 chassis units.

Yes the ARMs and Power units are slow but we have several hundred of them sitting around using electricity and generating heat and the only thing they accomplish is telling the rest of the switches that they are up. Is that worth looking into? If a small client could be included in switch firmware, you could have a small army of processors folding away 20+ hours a day.
Ditto for companies like Cisco an Juniper who use much of the same hardware.

[ChristianVirtual]

Nice idea; and you can add millions of iPod Touch getting charged in nighttime being idle, and androids and iPhone ... All ARM CPUs. But as you also said: priority is less on low power CPU's. It's more high power and fast return which enable the researcher doing their part earlier. The ever increasing number of atoms in the WU don't fit into the limited world of ARM system. I assume also that the switches you mentioned not have enough memory to hold and work with the WUs.
In order to fit into this limited system setup the WU would need to get much smaller hence the number increase to get the same level of science done.

[Jesse_V]

There are other things you could do with them though, assuming you can get permission to install something of course. There are many other distributed computing projects out there, and some might work well on ARM. There are also projects that are more bandwidth oriented, such a Tor non-exit relay. As ChristianVirtual pointed out, they may not be appropriate for F@h, but there are certainly other things that they can be used for.

[7im]

FAH on mobile devices [No]
Running FAH on a Raspberry Pi [No]
Folding on an iPad?

[Ivy Mike]

The combination of the PPC and the ARM processors doesn't help? Most of these switches carry a PPC and an ARM with most of ours carrying 3 ARMs (one per blade)
Some of them carry 4 with the core switches carrying 9 (1gb modules plus a 10gb) The management module has 4Mb of Flash, 128Mb compact flash and 256Mb RAM. The ARM units get 144Mb of packet buffer space. This could prevent them from tackling big work units.
I must admit my understanding of this type of distributed computing is very limited having just started playing around with it myself more for data mining purposes. We have nearly 20 years of data on old events and conventions stored on tapes at the moment.
I also hate the amount of waste I see in our technology departments and this seemed like a decent possibility. We also have good sized internet circuits which sit unused for the most part. We rarely come close to capacity even during our largest events.

[Joe_H]

The combination probably would not help. There was a PPC based client for OS X, but development on that ended years ago. The last WU's for that stopped being processed about a year ago. So that code is old and very out of date, even if the code could be ported to the OS used by the servers. Even on a fast PPC processor, I was using an upgrade that used dual 7448's at 1.73 GHz, it took 3 days to complete a WU that could be completed in a few hours at most on more recent processors.

[bruce]

As an imbedded router, it's unlikely that the hardware will include a Floating-point coprocessor since FP operations rarely, if ever, are present in routing operations. The FAH code spends its time predominantly doing FP operations so optimized code wouldn't run and performance of unoptimized code would be abysmal. Realistically, those millions of untapped processors should be working on something else.

[BIG_RED]

I hate to point you to boinc but they have an official boinc for android client. Which runs on arm not sure what os your routers run but you may be able to port it to run. Hopefully soon people here won't think 2008 smart phones then brush arm off. I would prefer to run fah but I can't only an old PC that get way to hot and android devices.

[mdk777]

There are many issues.
How about battery life? Having an android device running at 100% computational potential would burn through the available battery capacity in minutes.
Assuming plugged in use, heat dissipation quickly becomes an issue. (theses devices are designed in cases to dissipate normal usage heat...not 24/7/365)

Finally, the diminishing return issue is really paramount.

Think energy density. There is energy on every wave in the ocean. Why then are we burning coal? Answer, energy density. The capital, the distribution, the return on investment, to harness the energy in waves just doesn't work out.

The same will most likely be the case for smart-phones, tablets, android devices for many years to come. The small computational power is just too widely distributed to make the collection attractive.

Of course this might change, but the low hanging fruit is still GPU...millions and millions of cards sold every year...many new cards over the last 5 years have gone from 1 to 5 TFLOPS /per CARD

This insane computational density and capacity exists now!

It is like a nuclear power plant compared to the wave energy farm

FYI...in case you think I am exaggerating...you can look up the history of supercomputing. The first supercomputer in the world to achieve 5 TFLOPS was around the year 2000. With one sub $1000 graphics card, today you can do the work of the worlds most powerful computer in the year 2000. (yes, not same exact calculations,program languages, limited input out put, recording etc. etc. etc...but on the broad scale of things...accurate)

PS, still only around 30,000 people folding on GPU.

Generating 17,500 TFLOPS

http://fah-web.stanford.edu/cgi-bin/mai ... e=osstats2

Think what a million total GPU could do

[BIG_RED]

Battery life at 80% CPU is 3 hours(10% every 20 minutes) on my galaxy s4 but I keep it plugged in 75% of the time. I keep it below 80% cpu to keep it from over heating. My s4 gets over 3 gflops for c code using rgbenchmm that's faster than the average Mac or Linux using fah. The galaxy s4 is last generation the lg g2 and note 3 should be 50% faster.( snapdragon 600 vs snapdragon 800) Remember smartphones out sale pcs. Tablet are going to out sale pcs next year they will pass them 4q 2013 but the overall year tablets just won't catch up.

I don't care if fah has a client for android they are limited by personal to test the code. I would prefer a fah client but i'm not spending money on something i will use rarely use and don't need just for fah. I will just use boinc until fah catches up to the times I understand it may take a couple of years before that will happen

On a side note didn't Japan just stop using nuclear power and now trying to replace it with wind, solar, and wave power.

[mdk777]

Yes, I am thinking of upgrading my galaxy note to a galaxy note 3 because the multiprocessor capability has doubled in the last 2 years.
Like I said, things might change in the future, but for now the heat, latency, and diluted nature (3gflops =1/1000 of 3 TFLOPS...right?) don't make for an attractive target.

On a side note didn't Japan just stop using nuclear power and now trying to replace it with wind, solar, and wave power.

trying(not succeeding) is the operative word.
time will tell the economic cost of the hysteria...too political to go into detail here.
But a quick check here and 100% replacement with fossil fuel for now.
http://www.eia.gov/todayinenergy/detail.cfm?id=10391

PS...all my opinion regarding android(and energy density), I don't pretend to speak for PG.

[BIG_RED]

Yes, I am thinking of upgrading my galaxy note to a galaxy note 3 because the multiprocessor capability has doubled in the last 2 years.
Like I said, things might change in the future, but for now the heat, latency, and diluted nature (3gflops =1/1000 of 3 TFLOPS...right?) don't make for an attractive target.

And there is 1000 times more smart phones (over a billion activated). With 200 million+ added last q2 with a growth of 43% year over year. Not even talking about the other android powered processors.

PS...all my opinion regarding android(and energy density), I don't pretend to speak for PG.

The last thing I seen from PG is a year ago saying they were watching phones. I just hope they are in contact with boinc to talk how it is working for them and the pitfalls.

As far as "how can we help fah grow" I have only noticed strong growth with new clients( Google toolbar, GPU, ps3) or when fah is in the news for another reason.