Re: Carbon Efficiency of Semantic Web and Linked Data Queries

I like the way Google is going almost carbon neutral here in Hamina Finland
by way of using cold seawater to cool systems. I hope they will also hook
up the onsite sauna* to use excess HPC heat soon ;)

I am still surprised they continue to run supercomputer clusters in places
like Texas (Frontera), Tennessee (Summit) and Livermore, CA (Sierra)

https://medium.com/arcticstartup-news/saunas-to-use-data-centres-excess-heat-c552e70946b


On Wed, Jun 19, 2019 at 2:17 PM David McDonell <david@iconicloud.com> wrote:

> Thought this might be of relevance to the discussion, re global data
> infrastructures (from my LinkedIn feed):
>
>
> https://www.digitalinformationworld.com/2019/06/the-world-s-most-creative-data-centers-infographic.html
>
> On Tue, Jun 18, 2019 at 6:34 AM Marco Neumann <marco.neumann@gmail.com>
> wrote:
>
>> While we in the Semantic Web / Linked Data community don't seem to fall
>> into the category of worst offenders in energy consumption, (I am just
>> looking at the forecast and data traffic breakdown on the internet[1] and
>> the remarks made by the data-centre expert in Cheltenham[2] that digital
>> mobile camera phone sobriety could reduce data traffic in Europe by 40%
>> immediately) current federated SPARQL queries seem to be less efficient
>> than one would have hoped for 20 years ago.[3] You are probably doing more
>> for your carbon footprint by turning off your monitor completely rather
>> than leaving it in stand-by mode [4] than by optimizing your federated
>> SPARQL queries or going way of Solid Pods. It seems to be still difficult
>> to estimate the number of deployed SPARQL solutions in industry and their
>> footprint in terms of resource allocation. One of the best known projects
>> but still heavily centralized SPARQL services the wikidata WDQS has a
>> rather modest footprint if you go by the numbers published recently [5].
>>
>> Still and since this is my subject interest here the support and
>> implementation for federated SPARQL query solutions is surprisingly
>> underdeveloped [3] . Looking forward to learn more about updates here from
>> QuWeDa 2019 [6]
>>
>> [1]
>> https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/white-paper-c11-741490.html
>> [2] https://www.nature.com/articles/d41586-018-06610-y
>> [3] https://svn.aksw.org/papers/2017/FedEval-summary/public.pdf
>> [4]
>> https://www.energuide.be/en/questions-answers/how-much-power-does-a-computer-use-and-how-much-co2-does-that-represent/54/
>> [5]
>> https://wikitech.wikimedia.org/wiki/Wikidata_query_service/ScalingStrategy
>> [6] https://sites.google.com/site/quweda2019/home
>>
>>
>> On Mon, Jun 17, 2019 at 8:31 PM Zachary Whitley <
>> zachary.whitley@gmail.com> wrote:
>>
>>> I wanted to add some perspective. The principal components of aluminum
>>> refining are electricity and carbon and takes a significant amount of
>>> electricity and produces large amounts of greenhouse gasses. Most of the
>>> electricity consumed is produced by coal. Yes, we should be concerned about
>>> energy consumption for computing but I wouldn't be surprised if you would
>>> save more electricity and produce fewer greenhouse gasses by *expending*
>>> computing resources on making aluminum production and recycling more
>>> efficient.
>>>
>>> [1] https://en.wikipedia.org/wiki/Aluminium_smelting
>>> [2]
>>> http://www.world-aluminium.org/statistics/primary-aluminium-smelting-power-consumption/#histogram
>>>
>>> On Mon, Jun 17, 2019 at 3:09 PM Steffen Staab <staab@uni-koblenz.de>
>>> wrote:
>>>
>>>> I don’t believe that a case can be made for physically decentrallized
>>>> p2p being more energy efficient.
>>>>
>>>> 1. Compute centers can be placed where energy is cheap and cooling
>>>> inexpensive.
>>>> Indeed this has been done a lot.
>>>>
>>>> 2. Cooling reduces energy needs. Generated warmth could even be
>>>> re-used. Not thinkable for a DSL-box.
>>>>
>>>> 3. Modern CPUs use less energy when unused. There is less need to
>>>> re-use unnecessary compute cycles
>>>> in DSL boxes (well, I guess these modern CPUs are only in laptops so
>>>> far - still).
>>>>
>>>> 4. decentralized energy production is good. Globally, however, people
>>>> increasingly live in cities. This is not where most
>>>> energy is or will be produced (though it can become more than today).
>>>>
>>>> For sure, there is a lot of fruitful, middle ground between going for
>>>> DSL boxes vs all using the same centralized compute center.
>>>> I don’t believe in the extremely decentralized scenarios very much.
>>>>
>>>> Steffen
>>>>
>>>>
>>>>
>>>> Am 17.06.2019 um 17:38 schrieb Henry Story <henry.story@bblfish.net>:
>>>>
>>>>
>>>>
>>>> On 17 Jun 2019, at 01:14, Marco Neumann <marco.neumann@gmail.com>
>>>> wrote:
>>>>
>>>> I would agree Henry. I think p2p networks are provably more cost
>>>> efficient than centralized services in particular for small data providers.
>>>> I think there now could be made a case with regards to energy efficiency.
>>>> Taking your example of underused resources I would not be surprised to
>>>> finding big tech already taking advantage of this network infrastructure of
>>>> the underutilized nodes (aka your browser) rather than benefiting the
>>>> individual end-users directly.
>>>>
>>>>
>>>> also good point with regards to using local resources,  similar to
>>>> modern energy networks where most of the budget is not consumed by its
>>>> production but its transportation, storage and infrastructure.
>>>>
>>>> Is there work on p2p search for solid pods underway? I need to look at
>>>> HTTP/2 and solid pods more closely I guess. my pod on solid.community is
>>>> currently not in a good shape and I am not really having the feeling of
>>>> being in control of my own data. Is it more advisable to run my own solid
>>>> pod?
>>>>
>>>> https://neumann.solid.community/public/
>>>>
>>>>
>>>> It depends on how much you want to involve yourself in these early
>>>> stages.
>>>>
>>>> In 1993 I installed Linux on my father’s 40Mhz Laptop to see how well
>>>> it fared,
>>>> but it required quite a lot of knowledge to do that. Now everybody runs
>>>> Linux
>>>> on their phone and calls it Android.
>>>>
>>>> At this point the cloud version would be less work to get going I guess
>>>> :-)
>>>>
>>>> I think of the web when deployed on individual instances as peer to
>>>> peer,
>>>> and with Solid it really is so, since for example you authenticating to
>>>> a server,
>>>> requires the Guard to become a client to fetch data from another server.
>>>> Each node can be in one and the other role at different times - which
>>>> is not
>>>> to say that some nodes like browsers won’t specialize.
>>>>
>>>> P2P file sharing with duplication of content across nodes should really
>>>> be
>>>> named something else, more like distributed content sharing. Adding
>>>> such features
>>>> on Solid pods would be possible, but I think they are trying to
>>>> restrict to keep focus.
>>>> Adding it the right way - with RDF data to link to other copies on
>>>> other pods - would
>>>> be a nice research project. Perhaps the most important place to add
>>>> that for
>>>> Solid servers would be as distributed (encrypted) backups of one's pod
>>>> on friends pods.
>>>>
>>>> Henry
>>>>
>>>>
>>>>
>>>> On Sun, Jun 16, 2019 at 5:25 PM Henry Story <henry.story@bblfish.net>
>>>> wrote:
>>>>
>>>>> My guess is that such studies have not been done, mostly because
>>>>> widespread
>>>>> deployment as would happen if Solid became widespread has not happened
>>>>> yet.
>>>>>
>>>>> But there are some reasons one could be optimistic.
>>>>>
>>>>> 1. everyone has a DSL box at home currently that is on and not doing
>>>>> much
>>>>> a lot of the day, so consuming energy for nothing. Instead with Solid
>>>>> Pods
>>>>> those would be doing something useful, and could use electricity from
>>>>> solar
>>>>> energy produced locally. So you don’t increase local electricity costs
>>>>> that much, you can use locally produced electricity, but you increase
>>>>> some
>>>>> consumption of data.
>>>>>
>>>>> 2. It is likely that most people communicate with local friends, and in
>>>>> most case don’t cross frontiers due to language barriers. This may not
>>>>> be
>>>>> the case for the W3C community, but for the wider populations this is a
>>>>> lot more likely.  So in a way Solid pods communicating with local
>>>>> friends
>>>>> would use less energy, since packets would not need to be sent around
>>>>> the
>>>>> world.
>>>>>
>>>>> 3. There are a lot of optimization strategies that can be made by
>>>>> having
>>>>> widely deployed pods. For example used in p2p networks, by fetching
>>>>> copies
>>>>> of data heavy media in the nearest cache.
>>>>>
>>>>> 4. With the internet of things growing, having the packets stay as far
>>>>> as
>>>>> required in the home rather than go to large service providers, should
>>>>> also improve data costs as well as privacy. That is the role of a
>>>>> local DSL
>>>>> box turned into a data pod is in any case going to grow in importance,
>>>>> so
>>>>> one may as well use this growing infrastructure.
>>>>>
>>>>> Since producing energy locally is more efficient, and communicating
>>>>> locally
>>>>> when that is needed is better, there are reasons to think that some of
>>>>> the advantages of large providers may be offset in other ways. That is
>>>>> without counting the huge improvements in efficiency in communication
>>>>> that come with HTTP2, reactive frameworks, and cpu efficiencies.
>>>>>
>>>>> Henry
>>>>>
>>>>> > On 16 Jun 2019, at 12:41, Marco Neumann <marco.neumann@gmail.com>
>>>>> wrote:
>>>>> >
>>>>> > Has anybody done work on Carbon Efficiency of Semantic Web and
>>>>> Linked Data Queries?
>>>>> >
>>>>> > The very nature of distributed data sets has to come with a
>>>>> substantial computational footprint every time a query is issued to a
>>>>> single node or a cluster of nodes for a federated query. On the other hand
>>>>> decentralization might actually outperform more centralized services in the
>>>>> future.
>>>>> >
>>>>> > I can find a number of papers and articles related to carbon
>>>>> efficiency in general computing and cloud computing environments and data
>>>>> centers but nothing specifically related to the improvement of operational
>>>>> efficiency introduced by Semantic Web and Linked Data infrastructures..
>>>>> >
>>>>> > There is CO2GLE which attempts to estimate the CO2 emissions per
>>>>> second released by web search engines like Google as a reference here:
>>>>> >
>>>>> >
>>>>> https://qz.com/1267709/every-google-search-results-in-co2-emissions-this-real-time-dataviz-shows-how-much/
>>>>> >
>>>>> >
>>>>> > Regards,
>>>>> > Marco
>>>>> >
>>>>> >
>>>>> >
>>>>> > --
>>>>> >
>>>>> >
>>>>> > ---
>>>>> > Marco Neumann
>>>>> > KONA
>>>>> >
>>>>>
>>>>>
>>>>
>>>>
>>
>> --
>>
>>
>> ---
>> Marco Neumann
>> KONA
>>
>> --
> David McDonell Co-founder & CEO ICONICLOUD, Inc. "Illuminating the cloud"
> M: 703-864-1203 EM: david@iconicloud.com URL: http://iconicloud.com
>


-- 


---
Marco Neumann
KONA