Computing desk
< October 7	<< Sep \| October \| Nov >>	October 9 >

Welcome to the Wikipedia Computing Reference Desk Archives
The page you are currently viewing is an archive page. While you can leave answers for any questions shown below, please ask new questions on one of the current reference desk pages.

October 8

Wikipedians who edit Wikidata

Hi, I'm curious to know how many Wikipedia editors also edit Wikidata and vice versa?

I actually asked this question already on the general Reference Desk and was told to ask here. Wavelength perhaps you could give your partial answer here? thanks--Plarishome (talk) 06:14, 8 October 2015 (UTC)[reply]

Plarishome, some editors are categorized in Category:Wikipedians who contribute to Wikidata, but I do not know the total number.
—Wavelength (talk) 16:20, 8 October 2015 (UTC)[reply]

Indirctly You are contributing to WD by linking articles to its pendant in other languages, left side frame in each article. The languages – in the WP also called "projects" – are handled by WD. --Hans Haase (有问题吗) 11:27, 11 October 2015 (UTC)[reply]

How common is following the moon for data centers?

How many data centers, cloud computing platforms, or the like, follow the moon in their activities? That is, how many companies distribute their work-load to the night-shift (if they have computing capabilities around the globe, obviously)? Is it feasible at all? It looks like a good idea to save on the electricity bill, but technically things could get tough. Keeping integrity, low latency, security, and predicting load (among other problems) do not seem as a minor task.--YX-1000A (talk) 14:04, 8 October 2015 (UTC)[reply]

The moon does not rise at night: the roughly 29-day cycle means that the moon is up during the day about half the time. Here are Sun and Moon rise-time tables from The United States Naval Observatory.

In fact, at the time of this writing, you would not see the moon from Earth at all during most nights until the very early morning, just about at sunrise. Nimur (talk) 14:36, 8 October 2015 (UTC)[reply]

You are completely right that moon != night. However, "following the moon" is the expression of choice for this type of work-flow, even if it's a misnomer. It's the complement of "follow-the-sun" which linguistically makes more sense, but I wonder too whether this scheme gets implemented. The technical problems here are also steep.

Anyway, do data centers transfer at night their processing of information to a computing facility on the antipodes? --YX-1000A (talk) 14:45, 8 October 2015 (UTC)[reply]

Thanks for explaining. I don't know much about this, but a quick search does seem to indicate that "following the moon" is done by data centers to save energy costs [1] [2] [3]. I think it will be hard to find hard numbers in terms of how many services do this, how they manage the hand-offs, etc, as some of that is private knowledge that could give the company a competitive advantage, and may be guarded as a trade secret. SemanticMantis (talk) 15:54, 8 October 2015 (UTC)[reply]

The links indicate that the concept exists. You might be right that it's difficult to find operational details about this though.YX-1000A (talk) 15:59, 8 October 2015 (UTC)[reply]

I guess you're aware that it's also mentioned in the article you linked to with a source Follow-the-sun#Follow-the-moon.

Note however, transferring to the antipodes probably won't be the most sensible solution. While it may help with seasonal effects on on daylight, lots of the earths landmass doesn't have antipodes, and even in cases where it does, the connection costs etc will often vary significantly. It would probably make more sense to worry most about longitude, without worrying so much about latitude. In fact, if you did consider latitude, considering seasonal issues, I imagine it's possible the price would actually by higher or lower for most of the day in one place during part of the year, while different in the other.

Also, I think the most sensible solution is more complicated than following the moon (where moon actually means nighttime). For example, if you consider a place like NZ, I'm not sure that prices during daylight at 1pm would on average be cheaper than (generally) nightime at 8pm. Unfortunately I couldn't find a clear spot price graph showing average price variance over time of day after extensive searching. This is of course assuming you are exposted to spot prices, if you're just a small data centre and aren't, then your rates would generally be 7am-11pm and 11pm-7am. Except for one lines company, who's peak, off-peak and night prices would seem to support my view that 8pm may very well be more expensive on average than 1pm [4] (these are the line chargers, but the unit prices are unsurprisingly similar [5]).

Then we probably get to one of the most significant issues namely that when you're comparing widely differing locations, the electricity price may vary enough that even considering spot prices, it's nearly always cheaper or more expensive. (You also have to consider cooling issues, hence why some data centres locate in places like Iceland, where you can rely mostly on natural cooling during most of the time. Then there are people doing unusual things like designing computers to be installed as heaters [6] [7].)

I would note that if you look at the study indirectly linked in our article, it doesn't seem to discuss night or moon at all. It does mention time of day variance, and stuff like 6am, but it doesn't actually seem to be suggesting following the moon makes sense. For starters it seems to be looking at data centres in the US only. Second, it seems to be looking at far more sophisticated systems either based on real time prices or recent historic prices. These would seem to make far more sense than naïvely following the moon, without considering whether it's actually cheaper to do so. (It also seems to highlight another issue namely that for some types of data centres, your expected load is also going to vary including depending in the time of day.)

My guess is that most of those who've actually implemented this, would have tried to "follow the price/cost", rather than "follow the moon".

Nil Einne (talk) 16:56, 8 October 2015 (UTC)[reply]

Yes, I checked those sources. However some of them are 7-8 years old. That's eternity for the Internet.

And by antipodes or moon, I mean it figuratively. You all are being too literal, even for the computing desk.

Maybe the whole thing is feasible for cryptocurrencies (like bitcoin) miners, but also in this case, they would be following the cents, as you point to. Maybe someone is just buying the night time of some servers for mining purposes.YX-1000A (talk) 17:37, 8 October 2015 (UTC)[reply]

No one can know what you mean unless you explain it, which you didn't until now. We can only go by standard definitions of the terms, and standard definitions of the terms for antipodes and follow the moon, are as I used them. Both distinctions are important here, the later particularly so, and one entirely missed by our article, nor did you give any comment suggesting you understood it until now. It's not clear that many, if any data centres actually follow the moon as defined by our article, or even the direct source it's using. The smartest strategy is not only not simply following nighttime, but may involve coming back to a location during daytime while the location you are at hits night time. Following the moon is an oversimplification of what the indirect source was actually suggesting. Tne antipode thing is important not just because of the issue of latitudes, but because as illustrated by the source I mention, in some cases you may have a scenario where even in terms of longitude the places involved aren't close to being antipodes.

Incidentally, by the same token, I don't understand what you mean "some of them are 7-8 years old". I only referred to one source that directly related to following the moon, namely the one present in our article, and it was published in 2009, not 7-8 years ago. (If you're going to say by 7-8 years you didn't mean it literally, you were including 6 years even though there was only one source and it was 6 years, well I don't know what to say....) Some of the data may be about that old, but in any case the source it self is focused on whether the approach may save money and how people should go about it, rather than whether anyone does it.

Not only did none of the other sources relate to follow the moon, or the internet, they were even more recent than 2009.

Nil Einne (talk) 01:55, 11 October 2015 (UTC)[reply]

It makes sense to crunch huge amounts of data, be it for mining bitcoins, for finding the next biggest prime, or for finding that message from space. However, keep in mind that this is not the same task that's being passed to other servers. Each server's admin get to choose how much resources he want to invest into the crunching, for whatever reasons. In principle you could have your server with 0% idle time if you wanted, independently of the kw price. However the admin will have to ask himself whether a task makes sense.Yppieyei (talk) 19:00, 8 October 2015 (UTC)[reply]

I think to answer the question, not common at all. The power it takes to run the servers is just a fraction of what it costs to run a data center, regardless of how much the servers are "working". Air conditioning, security, maintenance, etc, all have to be running before you even start installing servers.. If you take the above example, you have 2 data centers mining bitcoins, one is running during the night using cheap power, the other is sitting there idle during the day? But it will still cost you a considerable amount to maintain during the day, unless you literally shut everything down and turn the power off completely and tell everyone to leave, which for a high volume data center with redundant power and batteries, etc is not trivial. Vespine (talk)

For the vast majority of data centers, load is user-driven. There are a few that use thermal storage to lower the day-time energy use (using a reservoir of coolant, or producing ice at night). Ssscienccce (talk) 03:55, 9 October 2015 (UTC)[reply]