[p2p-research] from an energy standpoint, distributed networks are just as > wasteful as centralized cloud computing

Tue Jul 14 04:24:19 CEST 2009

Below is an interesting, and disturbing, calculation by Andreas Schiffer,
that was shared over the IDC mailing list recently, showing how wasteful p2p
computing can be ...

It intuitively makes sense that centralization may be less wasteful than
peer repetition, but what about the use of spare cycles, it's potential
combination with distributed energy infrastructures ..

Michel

----- Forwarded Message ----
> From: Andreas Schiffler <aschiffler at ferzkopp.net>
> To: Michael Bauwens <michelsub2003 at yahoo.com>
> Cc: "idc at mailman.thing.net" <idc at mailman.thing.net>
> Sent: Friday, July 3, 2009 12:31:39 PM
> Subject: Re: [iDC] MySpace staff cuts
>
>
> I fully agree with your statement, however my point was less about
community
> control and capital interests, but more about the environmental impact
that all
> this networked tech-stuff has.
>
> I think that from an energy standpoint, distributed networks are just as
> wasteful as centralized cloud computing, if not worse. Each peer node is
using
> consumer hardware - a PC, router + cable modem in always-on mode. These
are
> components which seldom have the fine-grained power control and load
balancing
> modern server farms can achieve. I totally agree that a distributed peer
> networks - in absence of a centralized server infrastructure - would be
better.
> But that setup is an utopia in my view; the centralized infrastructure
will
> always be there as well. It's like with spam - it does not stop to flood
our
> inboxes, because enough people (who?) will still click on the links.
>
> Let's estimate for example the instantaneous power load for the eDonkey
network.
> My trusty mlDonkey client will immediately connect to a few servers, which
show
> about 850K users are connected to the p2p network right now.
> Server                          Users       MaxUsers  peerates.net
>     94783        250000  www.usenext.to          611621     1000000
> Sharing Kingdom 3      159871     350000
>
> These users will support my request for - say - the download of "Fr - Les
> Pirates De La Silicon Valley - Histoire Des Fondateurs De Microsoft Et
Apple -
> De Martyn Burke, 1999 - (Film Documentaire p2P Peer To Informatique Pc)"
which
> is supposed to be available in the network (illegally?).
>
> Each endpoint will probably consume on average 100W of power for the
download
> which takes probably 8 hours. That's a nice 680kWh of energy during this
time -
> for the endpoints alone. Of course during these 8 hours the network will
also
> carry hundreds of thousands of other downloads (mostly porn), so the
individual
> will only see his/her own cost ... for my rig that is about 1kWh or about
> 0.20cents in actual costs all the while the system was used to write this
email
> and doing a couple of other online tasks + I watch the movie for another
> 0.03cents in power once the download completes.
> Cost: 23cents, Time: 11h, Own: 1 file on the HD, Legal: no
> User Power Consumption: 1kWh, Fractional Power Consumption for delivery:
> 680kWh/80K downloads = 0.01kWh
>
> Now let's compare that to a Netflix-scenario and stream "Hang 'Em High
(1968)"
> (the Pirates movie is not available as stream). They probably run a big
> datacenter and the 2 racks of blades that support my query and and feed
the
> actual movie stream run probably about 10KW each, thus sucking 60kWh out
of the
> grid in 3h. Again at least a thousand other users are supported during the
same
> time by the providers hardware. So I watch for 2-3 hours and turn of the
PC
> afterwards. I also pay 0.33 cents in membership fees + 0.03cents in power
while
> watching the movie.
> Cost: 36cents, Time 3h, Own: nothing, Legal: yes,
> User Power Consumption: 0.13 kWh, Fractional Power Consumption for
delivery:
> 60KWh/1K streams = 0.06kWh
>
> Of course a proponent of peer networks might now say, that this is a wrong
> calculation, since the computer in the Netflix-scenario is not on for 3
hours,
> but for much longer (i.e. for work and other tasks) and thus can do the
> peer-stuff in the background for free. Totally true - I have my box on all
the
> time and never turn it off. For convenience the cable modem and wifi
router are
> on all the time as well. I am sure many on this list will have similar
setups
> and PCs that are on at least 50% of the day.
>
> With the assumption of an always-on endpoint, the equation changes again:
> User Power Consumption: 2.4kWh, Fractional Power Consumption for delivery:
> negligible compare to user power consumption
>
> So for most people the discussion forks here around issues of ownership,
media
> cost, speed of access ... because energy is cheap (compared to AC) and
> unavoidable (for convenience).
>
> But the problem that I originally raised about the network creating a
> significant power load, shows up again when we consider all the services
one
> could have used during the 24h on-time of the PC: UPS, eBay, thinkgeek,
google,
> yahoo, slashdot, email server, hotmail, skype, 10 websites, etc. etc.. Now
the
> Fractional Power Consumption goes up again rapidly ... because all the
servers
> that support a "possible delivery" of content are on as well, regardless
if I
> download content or not. In fact the more access to services I can have
and the
> better the services get the worse this fractional power consumption gets
even
> with data center optimizations in place. While actual numbers are hard to
come
> by, they are probably not very good. One could read 1-2% of total global
power
> consumption a few years back. Tendency increasing, since every 1000 iPhone
which
> are sold need an extra AppStore server.
>
> So the bottomline for me remains: Unless endpoints are getting more
> carbon-footprint neutral, distributed networks do not change the
ecological
> equation of the Internet whatsoever.
>
> --Andreas
>
>
> Michael Bauwens wrote:
> > Dear Andreas,
> >
> > Thanks for the calculation and reminding us that indeed, the internet
and its
> services rests on a costly social and private infrastructure.
> >
> > But I would also suggest that one of the reasons that platforms like
Google
> and YouTube costs a lot of money is because of the choice of highly
centralized
> and controlled infrastructures, which requires a high treshold of capital
> outlays and therefore serves as a competitive advantage.
> >
> > Perhaps if some of the companies went bankrupt in the future (as many
web 2.0
> services have already done), the hacker community would discover that they
could
> be rebuilt as truly distributed networks, thereby distributing the costs
as
> well. In this way, like the Wikipedia and Linux production, the private
> interests would become substantially more periferal and the relative power
of
> the community relatively stronger,
> >
> > Michel
> >
> >
> >
> > ----- Original Message ----
> >
> >> From: Andreas Schiffler
> >> To: "idc at mailman.thing.net"
> >> Sent: Thursday, June 25, 2009 11:21:42 AM
> >> Subject: Re: [iDC] MySpace staff cuts
> >>
> >>
> >>
> >>
> >> If the Internet should or shouldn't be about making money is really
besides
> the point. To "run the net" costs money - boatloads of money - hence one
cannot
> disjoin the internet from "things that have to do with money" including
making
> some from nothing but electrons flying around in wires and semiconductors.
> >>
> >> I think it is helpful to provide some context to where these costs are
coming
> from ... just calling out bandwidth isn't doing this justice. Let me try
to
> develop a chain that needs to be put into place for something to happen.
> Assuming we want to engineer the average June 2009 page view after
> time-traveling from the stone age, we'd need to build and operate this:
> >>
> >> Client Device/Connection - Last Mile Provider - Cloud
Providers/Telecoms -
> Hosting Facility - Server Device - Operating System/Software -
> Programmer/Designer - Content
> >>
> >> Next let me try to estimate some cost numbers for each item, say for
the US,
> for one year, for fancy commercial stuff.
> >> [Disclaimer: These are just crude order of magnitude estimates without
much
> research backing them up. Maybe we can crowd-source better numbers!]
> >>
> >> Client Device/Connection: 150M online devices at $100/year for power,
maybe
> 30M new devices/year at $500 a device = 30B
> >> Last Mile Provider: 10M new hookups/year at $1K a pop (pays for
provider
> infrastructure and personnel) = 1B
> >> Cloud Providers/Telecoms: 10 providers, 10B each/year in revenue, 25%
is
> internet (pays for provider infrastructure and personnel) = 25B
> >> Hosting Facilities: 10 big/100 small facilities, 10M setup+10/100M in
> power/cooling costs = 2B
> >> Server Device: 1M servers at $1000 each = 1B
> >> OS/Software: 1M servers at $2K each = 2B
> >> Programmer/Designer: 300K people at 100K/year each = 30B
> >> Content: priceless ... not really, 10B stock images/sounds/blogs at $1
each,
> 100M books/videos at $20 each = 12B
> >>
> >> Now let's split this into two groups (a) what crowd sourcing/open
source can
> affect in terms of price and (b) what needs to be payed to "the system"
anyways:
> >> (a) open/crowd-sourceable: Last Mile Provider, OS/Software
(Linux/apache),
> Programmer/Designer (phpBB/mozilla), Content (wikipedia/flickr) = 50B
> >> (b) corporate system: Client Device/Connection (PC/Power Utility),
Cloud
> Provider/Telecoms (ATT/Akamai), Server Device (HP/IBM) = 56B
> >>
> >> So even if open/crowd-sourcing is 100% effective, the corporations who
run
> "the net", make "the gadget" and provide "the juice" still need to come up
with
> the other half of the bill. Thus I conclude it will never be a profitless
> enterprise. The current preoccupation with advertising and user generated
> content is in my view partially a fad (yes, many people don't care about
many of
> the services) and partially an economic reality (it pays, as Google
demonstrates
> so well).
> >>
> >> To cut something out of (b) many, many people would need invest into
devices
> that use less power and last much longer, build community mesh networks
and
> mini-clouds, and consume less server-based streams.
> >>
> >> And another note on the original discussion about mySpace layoffs - one
of
> the biggest chunks of the cost in the chain is the personnel as one can
easily
> see in the above estimate. So if a company wants to loose less money
(YouTube)
> or make more profit (mySpace), they just fire most of the people who made
the
> software but keep the servers running. This will go without anybody
noticing
> anything (this works for at least a year - been there, done that).
> >>
> >> --Andreas
> >>
> >
> >
> >
> >
> >

-- 
Working at http://en.wikipedia.org/wiki/Dhurakij_Pundit_University -
http://www.dpu.ac.th/dpuic/info/Research.html -
http://www.asianforesightinstitute.org/index.php/eng/The-AFI

Volunteering at the P2P Foundation:
http://p2pfoundation.net  - http://blog.p2pfoundation.net -
http://p2pfoundation.ning.com

Monitor updates at http://del.icio.us/mbauwens

The work of the P2P Foundation is supported by SHIFTN,
http://www.shiftn.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listcultures.org/pipermail/p2presearch_listcultures.org/attachments/20090714/5566232b/attachment-0001.html>