Wikipedia:Reference desk/Archives/Computing/2010 March 18

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.

March 18

Game development - relative profit now vs then?

Reading about Zork, which was developed by ~4 dudes, made me wonder about the relative profitability of pc/videogames over time. Clearly development costs & team headcounts have skyrocketed in the past 30 years, but gaming is now a 6(8?) billion dollar industry in the US. On relative terms (revenue/cost as a %) is it getting more or less lucrative on average to make PC/videogames? 218.25.32.210 (talk) 05:29, 18 March 2010 (UTC)

It must be more. Normal people who surely scoffed at gamers only 5-10 years ago are now playing World of Warcraft, and paying a monthly fee to do so. ¦ Reisio (talk) 06:02, 18 March 2010 (UTC)

Yeah but how many dozens of MMOs have been outright failures? For a reasonable analysis you've got to include all ventures, not just the phenomenally successful ones. 218.25.32.210 (talk) 06:46, 18 March 2010 (UTC)

Paging SteveBaker --Tagishsimon (talk) 06:03, 18 March 2010 (UTC)

This topic is summarized at video game industry under Economics. The stuff about the current market is unreferenced and most likely based on anecdotal evidence instead of real data. Basically, the video game industry is like any other entertainment industry now. To make a blockbuster hit, it takes tons of people. There are cases where a small indie group get lucky with a big hit - which is nowhere near the size of the big-budget blockbusters, but still profitable. There are many failures and a lot of fear - causing tons and tons of repetition of past hits. -- kainaw™ 15:46, 18 March 2010 (UTC)

That's right. There is no data out there on individual projects' failures or profitability levels, because only the publishers know this data, and they don't release the data. But one visible indication that it's a lot more profitable today is the simple fact that you have publicly traded third-party publishers like ERTS, ATVI, TTWO, and THQI; and salaries for the top people have, I think, outpaced inflation when compared to a programmer's salary from 25 years ago. Comet Tuttle (talk) 16:29, 18 March 2010 (UTC)

It's worth noting that, while small groups (either indies, or small groups within big games companies) rarely get big hits, they do get lots of little hits. Browser flash games, mobile phone app games, etc.. A recent episode of BBC Click discussed the topic and apparently lots of games companies are starting to favour the smaller games. They are lower risk, since you can easily make 100 games and know that at least a few will be successful. If you concentrate on big games then you can only make 2 or 3, and it's entirely possible that they'll all be complete flops. --Tango (talk) 17:02, 18 March 2010 (UTC)

Although that's a great theory that I would love to see them adopt, no major game publishers are doing this currently. Comet Tuttle (talk) 17:22, 18 March 2010 (UTC)

The story said they were starting to, rather than actually doing it. I think it is happening on a small scale at the moment, and not with the biggest companies. --Tango (talk) 17:52, 18 March 2010 (UTC)

I was trying to draw a relation between the rest of the entertainment industry and video game production above. The idea of a small house putting out 100 games and hoping for a few minor hits is not abnormal. There are many straight-to-video movie companies that operate that way. Large companies can do the same. Think of Disney. Yes, they always have 2 or 3 major blockbusters in production, but they also have 20-30 straight-to-video movies like Cinderella VI, Bambi IV, or Ariel's Next Big Adventure... I'm see Square Enix growing to that level. They are already recycling old games into cheaply made (and lower priced) games. If they are successful, others will likely follow suit. -- kainaw™ 00:41, 19 March 2010 (UTC)

(I work in the games business - I'm a graphics programmer). The problem with the videogame business is twofold:

1. Customers' expectations about 'depth' of content has grown. Zork has a few hundred pages of text description of it's world - and a few thousand lines of software to run it - and that was enough back then. Something like GTA IV has every building in a large city, hundreds of music tracks, hundreds of characters, voice acting, hundreds of cars and trucks, etc. The software for a modern game engine could easily reach a half million lines of code. That's why ZORK could be written by four guys - but GTA IV required several hundred. There is a slight reversal in that trend with cellphone games and for things like Guitar Hero and Rock Band - but it's going to be short-lived.
2. Only about one in every 35 games made ever turns a profit. That makes it an incredibly risky business. The reason games like Halo-3 and GTA IV made such an ungodly amount of money was because they were a virtually guaranteed hit. Sequels of games that are popular are generally popular - so you can invest a big chunk of change into them and have high confidence that you'll make money. But a new, novel game is insanely risky.

A typical PC/Xbox-360/PS-3 game requires around 100 people for perhaps three years (although it's wildly variable both in time and staffing). That's mostly split between programmers (earning $80k to $130k maybe), artists & designers (earning probably $40k to $70k) and game testers (who often earn little more than minimum wage). There are also licensing costs for music and outsourcing costs for bulk content creation - and more licensing fees if you are making the game of a movie or TV show or something. It's quite easy to burn through a million dollars a month in the last half of the development of a big ticket game. If it fails - that's a truly massive loss.

The game I was working on for Midway when they folded had 80 artists and maybe 20 programmers with a dozen or so testers - plus managers, designers, audio guys, IT, HR, etc. We'd worked for more than 2 years on our title (a 3rd person game about high-class criminals that set in up-market locations in real-world, present-day, Chicago) - and I have to say that it looked good graphically, it had good audio, we had a famous Hollywood movie director put his personal "feel" into the thing, advising us on everything from dialog to pacing to camera angles. It worked reliably, it ran well on PC, Xbox-360 and PS-3, we had several test missions fully playable in our open-city world. Technically and artistically, it did everything we set out for it to do. But somehow, indefinably - it just wasn't much fun to play and so it was axed. Making something "fun" is tough - nobody really knows how to do that - and nobody really knows how to "fix" an un-fun game because we really don't understand why the game wasn't fun. So you can spend tens to hundreds of millions of dollars and end up with something that doesn't even end up on the store shelves - let alone turns a profit!

This means that small companies can't spend enough to get that game 'depth' without having someone with deep pockets (a "publisher") funding them and taking the risks and most of the profits. A large game company or a major publisher can make enough money off of the 1-in-35 that turn a profit to pay for the 34-in-35 that are either cancelled and never get onto the store shelves, or fail miserably once they get there. But what this means is that when you read about the ungodly amount of money something like GTA IV made, you have to understand that those profits had to pay for a ton of game development that went horribly wrong for whatever reason.

MMO's are even worse because the startup cost for setting one up is quite outrageous - vastly more than a normal game - and the risks are just as high. However, once you have them up and running, the profits are good - even when weighted against the cost of maintaining the servers, keeping the game working and continually adding content. But so many of them fail after just a few months, it's a complete crap-shoot.

Phone games (on iPhone and Android) are popular right now - and we're in that little niche in time when the public will splurge a couple of bucks on a game without it needing to be advertised to death - and the depth of content of the game can be fairly small. That allows small games companies to make them with small groups of developers and relatively little risk. Of course at $1.99 per game, they have to sell an awful lot to make money - but that's OK. Also, you can make a lot of them. A handful of guys can make an iPhone game in a year easily - so even a small company can churn out a hundred games in the time it takes to make one PC/Xbox/PS3 game - and the law of averages says that a few of them will be phenomenally successful and pay for all the rest. But this is a brief window. As these gadgets get more sophisticated - and the prospect of using cloud-computing for cellphone games takes off - the push for deeper content will once again make these games cost a lot more and we'll get all of the risk back.

Games are unlike almost any other product in that close to 100% of the cost is in the development and advertising - both of which cost the same whether you sell 1 copy or a 100 million copies. It costs less than $1 to copy a DVD-ROM and a manual and put it into a box. Most of the copies you sell will go for $20 to $40 which looks like a pretty amazingly good markup. Now WalMart and Microsoft/SONY/Nintendo take their cut (remember - game consoles sell at a loss and the hardware manufacturers take a cut of games sold for them in order to make a profit). If you're a game company who went via a publisher - they'll take a huge slice. But if you spent $1,000,000 a month for several years making it, you're going to look pretty bad if you don't sell a hundred million copies.

SteveBaker (talk) 09:06, 19 March 2010 (UTC)

A year to develop an iPhone game? At the recent Game Developers Conference in San Francisco, the turnaround time for an iPhone-style mobile "casual game" was reputed to be more on the order of three to four weeks. On the one hand, the development of content is fairly shallow, but the market forces (specifically, the tendency to be a brief fad, go viral, and then disappear into obscurity), in conjunction with a very saturated market, forces a constant race to produce content of dubious quality. Three or four indie developers need to churn out new content at a frantic pace in order to hope to sell enough copy to make rent. This GDC 2010 Announces iPhone Games Summit Line-Up overview from Gamasutra links to several in-depth interviews from GDC2010 on the art and business of iPhone game design Nimur (talk) 10:10, 20 March 2010 (UTC)

In response to your last paragraph - doesn't that apply to all software, rather than just games? The other industry that is very similar is pharmacology. Most of their research either goes nowhere at all, makes an ineffective drug or one with too many or too severe side effects, and the few that work have to pay for the rest (which is why drugs are so expensive) and, also, the cost of making a pill is negligible compared to the development costs. Computer processor design and manufacture is quite similar. --Tango (talk) 14:13, 20 March 2010 (UTC)

Dell Optiplex to upgrade

Can anyone recommend a good Optiplex model for upgrading/building my own PC, that wouldn't cost much to buy secondhand? Basically just one with a large form factor case and a power supply that will run a decent processor and graphics card. I'm open to suggestions on other manufacturers, but was thinking of another Optiplex because I know how robust they are. This will be a kids' pc once it's finished! 89.195.206.83 (talk) 09:04, 18 March 2010 (UTC)

Intel Core

what is intel core —Preceding unsigned comment added by Kdcee (talk • contribs) 10:31, 18 March 2010 (UTC)

See Intel Core. -- Finlay McWalter • Talk 11:11, 18 March 2010 (UTC)

Also, this archived question from July, "Difference between core 2 duo and dual core". Note the important distinction between "core" (a generic term for a particular part loosely defined subset of the innards of a computer processor), and "Core™", a brand name for a specific product technology released by Intel. Nimur (talk) 15:57, 18 March 2010 (UTC)

Specific keyboard key will not function

Resolved

When i turned my computer on thiS morning, i found that the 'S' key would not function (i'm uSing the character map to copypaSta the letter). It'S not that it doeSn't function, it'S whenever i preSS it the window'S focuS iS Stolen by Something elSe. The weird part iS that every keyboard i plug in doeS thiS. The laptop built-in keyboard and even the On-Screen Keyboard do thiS. Before i Start a painfully long viruS Scan, haS anyone elSe had thiS problem? (PS: ShortcutS don't do thiS, CTRL-S workS fine)

 Buffered Input Output 12:24, 18 March 2010 (UTC)

I take it you're cutting and pasting the "S" now ? If you have a recent system checkpoint, now would be the time to do a restore. StuRat (talk) 15:30, 18 March 2010 (UTC)

It was a virus. Thanks anyway.  Buffered Input Output 15:40, 18 March 2010 (UTC)

fortran 90

I want to download fortran 90 software (free). please give me some link.Supriyochowdhury (talk) 12:29, 18 March 2010 (UTC)

GCC supports Fortran 95. —Korath (Talk) 13:15, 18 March 2010 (UTC)

GCC also supports Fortran 90. Here's the official gfortran manual, including downloadable installers for most major platforms. If you are using Linux, gfortran may already be installed. Nimur (talk) 14:23, 18 March 2010 (UTC)

[1][2] --Normansmithy (talk) 15:03, 18 March 2010 (UTC)

FrontPage site user registration

I decided to add user registration to my web site. My IP had set it up with a domain name and a prefix and a subsite with frontpage extensions installed. I now realize that the prefix site is considered the top level site where I place things like the user registration which is intended to protect access to the subsite. It also appears that I can only access the top level site using FTP and the subsite using HTTP. However, once I create the user registration I also have to protect the subsite so that only registered users can access it. Currently without a user registration on the top level site anyone can access the subsite if they know what it is named and it appears that only the IP can change the site prefix or the subsite name. Is this a correct understanding and if so how do I protect the subsite from access by non-registered users and how do registered users sign in? 71.100.11.118 (talk) 16:07, 18 March 2010 (UTC)

Ubuntu 9.10 Hibernating without consent

Every time I get the larger update which involves a restart of the computer, I have noticed that my computer tends to go to sleep after a few minutes of inactivity. I did something last time to sort this out, but recently we've had this update again, and it's started again. I have checked Power Manager, and have set everything to basically be on forever (except that 'when battery level is critically low' I have no choice but to put it on 'hibernate'). I have set the screensaver to never come on, but can only set the computer to be regarded as idle after two hours. Still, after a few minutes of inactivity, the screen goes black, and I have to press a button to get the screen back again. Particularly annoying is when I don't do that, the computer switches off. Now, when I leave a computer on, I leave it on for a reason, and don't want it to switch off when it feels like it. Can anyone tell me what the problem is and how to solve it? On a side note, I would ask this on the Ubuntu forums, but it can sometimes take days to get an answer there, and when I do, usually the writer assumes I know everything about Ubuntu, and so, although well-meaning, the answers are too technical and therefore useless to me. Any help would be appreciated. --KägeTorä - (影虎) (TALK) 17:51, 18 March 2010 (UTC)

It's possible this is due to the monitor (which might have a separate low-power mode). Check your graphics configuration to see if a low-power / idle state signal is set up for the monitor (this is often separate from the software-level screensaver). The configuration might also be accessible through your monitor's firmware (using the buttons on the screen, rather than a software interface). Nimur (talk) 09:46, 20 March 2010 (UTC)

Cheers, I will check that. How do I do this? --KägeTorä - (影虎) (TALK) 14:13, 20 March 2010 (UTC)

What happened to Hotmail?

I am on a computer with Windows 7. I have had lots of problems since this software was installed, but never anything like today. I signed in to Hotmail and was doing fine until I did a search for all emails I had sent to myself with the appropriate subject line that I won't bother to explain. I clicked on one of those and then, when it was ready to delete, I did so and clicked on "Back", only to be told the web page had expired. I clicked on "Back" again and ended up in the inbox, which had nothing of value. For some odd reason I clicked on "Back" again and found myself in the other email service I had used earlier. I clicked on forward and the URL turned green (it also had a lock) and the page was completely blank. I typed "Hotmail" and even tried going there from Bing. Same result. I tried other ways of getting into Hotmail (actually, I was looking for Hotmail help) and finally succeeded in getting back to the inbox. But nothing I clicked on would take me anywhere.

It gets worse. I went back to that other email service and found I couldn't get into the inbox. I could see a list of folders but clicking on any of them just got me a bunch of white space where the list of emails should go. Fortunately, Lycos behaved normally. It has similar problems to the other email services I mentioned, but these are the result of the foolish way each one was designed. What's happening to me today is not intended by anyone.

And in one newspaper web site, I was unable to go to pages other than one in the articles with multiple pages. I'm contacting that site now.Vchimpanzee · talk · contributions · 18:03, 18 March 2010 (UTC)

Oh, and the items at the bottom of this page that I can normally click on require copy and paste.Vchimpanzee · talk · contributions · 18:07, 18 March 2010 (UTC)

I don't believe this has to do with Windows 7, but with your web browser. Which program are you using? —Akrabbim^talk 18:15, 18 March 2010 (UTC)

Internet Explorer, probably 8.Vchimpanzee · talk · contributions · 18:45, 18 March 2010 (UTC)

At last. I noticed something helpful. The yellow line that usually is saying a pop-up was blocked has been showing up all day, with a different message. I ignored it because supposedly there's nothing I can do. But I read the message and clicked like it said to and got this result when I matched up the message:

Okay, it won't let me copy and paste. Essentially it says ActiveX needs to be removed from the restricted sites list. There's something about changing security settings. I am at a library and furthermore, they tell me they can't do anything except call IT and that requires a work order. This takes time. They may not do anything today.Vchimpanzee · talk · contributions · 18:53, 18 March 2010 (UTC)

More here.Vchimpanzee · talk · contributions · 21:11, 30 March 2010 (UTC)

do I need to use a switch to segment, or if I have enough LAN plugs I dont need them?

I got a p.o.s. (point of sale) equipment for a small store, it came with a switch, on the back of the switch it says: typical network setup:

internet - Wireless Modem - This switch - stuff (laptop, computer, the pos equipment, etc)

But the wireless modem here already has 5 big yellow lan connectors. Do I need to use one of them for the switch for some (security) reason? Does it "segment" or do anything else with the network? Thanks. --84.153.225.240 (talk) 18:10, 18 March 2010 (UTC)

Network switches exist to give you more ports, so, no, the switch itself isn't necessary if you want to plug stuff directly into the wireless modem's ports. Do you know the model number and manufacturer of the wireless modem? That would let us give more specific advice. Comet Tuttle (talk) 18:22, 18 March 2010 (UTC)

TP-Link brand Model no. TL-SF1005D. By the way the p.o.s. is ingenico brand. So, if there are 5 free LAN connectors on the back of the wireless connection, it does NOTHING to plug one of them into this router switch and then the p.o.s. into the router switch, versus just plugging it into the wireless lan? thanks. So what's this bit about "network segments" in the routet switch article? --84.153.225.240 (talk) 18:44, 18 March 2010 (UTC)

You're getting your terminology a bit mixed up, though you had it right in the original post; you have a DSL modem (which in your case is also a router), a switch, and your other devices which need connectivity (laptop, POS, etc.) There are switches which have special features for network security (often found in managed switches) but the switch you mentioned is not managed. Adding it to the network will do nothing more than give you more ports to connect network cables.

Technical comments: The model TL-SF1005D is your switch that was provided - Comet Tuttle was asking for the wireless modem's information, but it's not really a big deal. Adding the switch would just make the network segment larger, but it's still one segment. Besides, I rarely hear the term "network segment" used outside of large / enterprise networks. Coreycubed (talk) 19:41, 18 March 2010 (UTC)

Sorry, I made a mistake, being in a rush. I've corrected "router" to "switch" as appropriate. It is the switch article whose lead includes the line "A network switch is a computer networking device that connects network segments." and then just sentence later, "The term network switch does not generally encompass unintelligent or passive network devices such as hubs and repeaters." So what does connects segments mean in a technical sense? Also, the part I've just quoted says that a switch has to do something more than a passive "hub", so what would this something be? What is the difference between the switch I've mentioned and the hub? Nothing?

The wireless modem model number is written on it as "Speedport W 503V Typ A". (It is a German modem, in line with my typing from a German IP). This wireless modem is connected to the DSL box thing. So, given that you really know everything now, what, in practical terms, is the difference between connecting the pos into the wireless modem; into the switch I mentioned and connecting that into the wireless modem; or (this is hypothetical, I don't have one), connecting it to a hub and connecting the hub to the wireless modem. What's the difference between any of these three? Thanks. 84.153.225.240 (talk) 20:13, 18 March 2010 (UTC)

Practically, there is no difference. In a technical sense, data needs to be able to travel around the segment, which in this case is your home network. (The wireless network would also be considered part of the network segment.) Let's say you had two computers. If they were connected to two different switches, those switches would in turn need to be connected to each other, or data could not pass between the two computers. Don't get caught up in the terminology and overcomplicate things! :) In lay terms, we'd just say that they needed to be networked. This is probably common sense - there are two devices that need to "talk" to each other. They have to have some way of passing information back and forth. Whether it is through the LAN ports of your Speedport or through the TP-Link switch (or both) does not matter. Your Speedport modem is actually quite the multitasker - it's acting as a modem, translating the DSL signal into a usable Internet connection, and as a router, providing NAT and DHCP addressing, and ALSO as a switch, providing four LAN ports, AND as a wireless access point (think of a wireless AP signal as a giant invisible switch, and connecting to a wireless network as walking up to that invisible switch and plugging your computer into it).

Since you're asking some hypothetical questions, here are a couple of examples of situations where it would matter how you connected it:

One switch is faster. Let's say that the TP-Link switch was capable of gigabit speeds (usually marked 10/100/1000). You'd want to connect all gigabit capable devices to that, and let the modem and switch talk at 100 megabits since that is still likely far more than your DSL connection is providing. That way, your computers can talk to each other faster, allowing them to transfer files among each other at greater speeds.

You have hubs. These are inferior to switches. You'd want to leave the hub out if possible since it is less efficient at exchanging network data than a switch is. The LAN ports in your Speedport modem are undoubtedly better than any hub would be. A hub just broadcasts packets it receives on all the others ports, so if lots of packets are being sent through the hub, you will get collisions. A switch, even an unmanaged one, can direct the flow of packets much better. The article on hubs which you linked goes into more detail, but basically a switch can handle simultaneous packet exchange while a hub cannot.

Since you also mentioned a repeater - wireless repeaters just listen for a wireless signal and rebroadcast it, effectively extending the range of wireless access points. This is why they, along with hubs, are not called switches - becuase they do not actively take a role in getting packets to their destination. Coreycubed (talk) 21:24, 18 March 2010 (UTC)

Copy/paste one partition's contents to the other partition in GParted

Because Windows was having major issues not being the first partition in the drive, I need to move the Windows partition to the front of the drive (trust me, I have explored all other avenues around this), so if I right-click in one partition and say "Copy", then click in the new NTFS partition at the front of the drive, why doesn't it allow me to paste? (The right-click and toolbar buttons are greyed out). Do I need to complete all other operations (resize, format, etc.) before I can do this?

Thanks!

110.175.208.144 (talk) 21:02, 18 March 2010 (UTC)

I don't know about using gparted for this; I've always just used dd - e.g. sudo dd if=/dev/sdc2 of=/dev/sdc1. Obviously you need to be super sure about the if and of - get them wrong and you'll overwrite the wrong partition and ruin your day. -- Finlay McWalter • Talk 21:14, 18 March 2010 (UTC)

Well, failing that, if I manually move the entire partition's content across using a file browser, will it work? Thanks! 110.175.208.144 (talk) 05:18, 19 March 2010 (UTC)

I'd use gparted to blank the NTFS partition at the front of the drive that you were attempting to copy to and then just shift the whole partition to the left. If you want to, you can create the other NTFS partition in the blank area where the Windows partition used to be, or just expand partition #1 to fill the space. Hope that helps. Coreycubed (talk) 21:31, 19 March 2010 (UTC)

Did that, worked like a charm :) Thanks! 110.175.208.144 (talk) 00:58, 20 March 2010 (UTC)

Does an electronic paper word processor exist?

There are ebook readers using electronic paper (e-paper) like the Kindle. Does there exist a device like that where you can edit text? I would really love to have that and be able to easily read outdoors in the sunlight. Too many laptops are hard to read in bright conditions. Is there a technical reason for this, maybe that blinking cursors or altering small portions of the screen are not conducive to e-paper displays? Thank you for your help. --Rajah (talk) 21:43, 18 March 2010 (UTC)

The ebook readers I've seen (a Sony and a Kindle) have displays with a pretty high latency. If memory serves, the Sony screen goes blank, then inverts, then a new screen appears (I don't remember precisely how the Kindle worked in that regard). Electronic paper#Disadvantages suggests that this is the case for the current e-ink technologies in general, and I agree with that article that the one's I've seen would be unworkable (or at least unpleasant) for a normal GUI. I'm not sure there's enough market for general high-ambient-light computing to justify making one with the current, suboptimal technology. -- Finlay McWalter • Talk 21:56, 18 March 2010 (UTC)

The iLiad allows annotation and highlighting of existing documents out of the box using an included Wacom stylus. It's Linux based, and according to the article abiword has been ported to it. The port is not really meant for editing - more for viewing rich text files, but a USB keyboard module has also been ported which allows you to plug in a USB keyboard [3]. So, yes it can be done if you are willing to do some hacking, but the end result probably won't be that smooth or easy to use.131.111.185.69 (talk) 23:09, 18 March 2010 (UTC)

Reading the site I linked to I discover they have even ported an instant messenger app and a spreadsheet program. I love Linux hackers - they are all quite mad (in a "this shouldn't be possible and certainly isn't advisable but we are going to do it anyway" sort of way). 131.111.185.69 (talk) 23:16, 18 March 2010 (UTC)

You can definitely edit text on the Kindle - I do it on mine all the time. The Kindle software offers a way to add notes to your eBooks - and that involves typing in text on the little keyboard and viewing it on the electronic paper display. You can also surf the web and enter text that way. That certainly works - so it must be possible (in principle) to do what you want. The usual problem is that machines like the Kindle are designed for ultra-low battery life - which they achieve buy using the ability of the ePaper to retain the image with the power turned completely off. The Kindle only uses power when you push a button. So you can read a dozen books over a period of weeks on one battery charge because it only takes maybe a thousand button pushes per book. However, if you were doing extensive word processing, the CPU would need to be on all the time and the display would be doing a lot of refreshing - and the battery life would probably be alarmingly short. The Kindle also has a couple of 'easter eggs' - you can play "MineSweeper" and "Gomoku" on it - which, again, requires the display to refresh 'interactively'. However, I don't know of any custom word-processors built around ePaper. SteveBaker (talk) 08:09, 19 March 2010 (UTC)

Why not just upload any documents you want to read to GoogleDocs or similar online wordprocessor? You'd be able to read and/or edit them on a Kindle easily then. --KägeTorä - (影虎) (TALK) 12:50, 22 March 2010 (UTC)

Including gpus, new and used, printers, Playstations, etc, and mainframes and services too what is the best teraflops for the money?

Say I want to do a whole lot of flops in a highly parallel way. Out of all the choices I listed in the subject line, what would give me the most FLOPS for the money. Put another way, if I had $20,000 to spend, what is the most flops I could get out of it, including on the used hardware market, and how. Just about the only thing I wouldnt consider is buying a botnet, or paying some Russian the $20000 to build me a viral botnet, for the calculations, though I suspect this would be the cheapest :) Thanks for your creative answers, and of course please feel free to include anything I didnt think of if it's better flops for the money. :) 80.187.97.187 (talk) 22:33, 18 March 2010 (UTC)

Can you clarify whether this is just fun speculation or whether you're serious about doing this? For example, there is a massive difference in the ease of implementation between a room full of blade servers and 250 Playstation 2s... if you really want to build your own parallel setup, knowing your personal software competency (or the competency of those available to you) will go a long way towards accurately shaping the answer. In short, the real question should be "what's the best speed per dollar parallel computing setup I can build if I know (insert programming language here)?" 218.25.32.210 (talk) 01:05, 19 March 2010 (UTC)

An awful lot depends on your use patterns. If you only need this thing periodically for one or two large calculations - you might be better off buying computer time from one of the large compute farm providers out there than owning your own system. There are all sorts of innovative ways to buy CPU time without owning the hardware (eg [4] or [5]). You can also sell your unused CPU time the same way. So by sticking with PC's and not buying weird hardware like Playstations, you could sell your unused CPU cycles and earn money back from your compute farm when you aren't using it. Many of the service providers sell Linux CPU hours at about half the cost of Windows CPU hours - and Linux computers are always cheaper then Windows ones because you aren't paying for the cost of the operating system - so if you are serious about this, you'll want to be running your application under Linux. If you work in a large organization, you might consider building a Stone soupercomputer - where you set yourself up to take old PC's that have been upgraded in other departments of the company and re-purpose them into a gigantic cluster. These machines are old and clunky - but the hardware is essentially free and your application is (hopefully) highly distributed - so having 100 1GHz machines that cost you nothing is better than having (say) 20 3GHz machines that cost you a thousand dollars a pop. A lot also depends on the nature of your calculations. If you can parallelize the code in such a way that it could run on the GPU instead of the CPU, you can often get a 100-fold speedup on a single PC! SteveBaker (talk) 07:59, 19 March 2010 (UTC)

This is a scientific economical question, such as the question "what is the most human-consumable calories you can buy for $20,000", even on the international commodities market. The answer could be "5000 gallons of vegetable oil". I would be happy with that answer.

Now, in my actual question, I am purely concerned, in a pure "calories" sense, what the MOST number of floating point operations I could get out of $20,000 is. Would the MOST number of floating point operations I could get out of it be by paying the $20,000 to a service, like SteveBaker suggests? Or would it be by buying 150 almost but not top of the line, used NVidia cards and middle of the line cases and power supplies for them, and powering it in Arizona, which (this is just hypothetical) has the cheapest power in the United States that you can start accessing for $20,000, then writing your code to run on those GPU's as a cluster. You see, guys, this is a purely scientific economical question, such as the analogous question about about what the most human-consumable calories you can buy for $20,000 would be.

Does anyone have an answer? 84.153.225.240 (talk) 13:37, 19 March 2010 (UTC)

But that's not a meaningful question. If you buy a $100 programmable pocket calculator and keep it for a million years - you have "bought a teraflop" - but sadly, it's so appallingly slow and can hold so little code that it's virtually useless. But if you're doing something like weather forcasting and you absolutely need to do a teraflop calculation to predict the path of a hurricane and you need the answer within an hour - then you have to buy a roomful of computers at monumental cost. Between those two extremes, there lies solutions like the stone soupercomputer and buying time from Amazon. It's not like buying apples. It's not a matter of how many teraflops - it's tiny details like how often you do this and how soon you need the answer - and how much memory you need - and whether the algorithm is parallelizable or not and whether you need lots of disk space and how much intercommunication between processors is required. You can't treat it like the commodities market. Sorry! SteveBaker (talk) 14:31, 19 March 2010 (UTC)

I recently gave a presentation at University of Nevada on how to develop a "personal supercomputer" for under $1,000 (I'll try to dig out my presentation for release). I basically loaded up a "middle-of-the-road" desktop computer from HP and loaded it with a "middle of the road" GPGPU-capable NVIDIA card. I spec'ed it out around 25 giga-calculations-per-second on the 2D or 3D wave equation, which is something like 300 GFLOPs. In my opinion, this is about the peak of the price-to-performance curve; I have some more metrics, (but many of them are not ready for release yet). But remember that if you are really doing high performance computing, you need to seriously analyze your algorithm to determine where your bottlenecks are. The FLOPS article explains very important caveats when measuring compute performance in "operations per second". For example, can your algorithm be represented in CUDA in the first place? And, if it can, will the data dependency overwhelm the parallel compute capabilities of a graphics processor's cache layout? Also - if you're going to cluster these systems, you will need a parallelizable code, either node-independent algorithms with some type of grid engine or some type of cluster programming API like MPI. (Or you can write pthreads and socket programs manually!) Do you know how to program in those sorts of systems?

Speaking of almost-publication-ready data, I have a friend who is soon to publish an ACM report on bit-error-rates in high-end vs. low-end GPUs from NVIDIA using data from the Folding at Home project. He corroborated my intuition with stastistical evidence; middle-of-the-road GPU systems (like the GTS250) have similar bit error rates to the high-end Tesla marketed cards; so unless your algorithm has a need for the huge GPU memory space, your system is better off buying 4 GTS250 cards rather than one C1060. As always, it depends on your algorithm needs - the C1060 can compute some problems that a smaller memory footprint doesn't allow. In that case, it's not about a price tradeoff - it's about meeting your algorithm needs.

Finally, one last insight regarding pricing - note that Amazon EC2 will rent "FLOPs" by the hour. Interestingly, their price point (~8 cents per hour) is lower than the cost of electricity needed to operate a server in most parts of Europe or the United States. So, keep in mind the power of outsourcing and economy of scale if you really want price-performance.

One last point: an actual published version summarizing some of the above, written by my colleague: Selecting the right hardware for reverse time migration. Note that this paper mentions a specific algorithm, because the computational needs vary so widely depending on application. Nimur (talk) 14:20, 19 March 2010 (UTC)

The low cost is because they have all of these bajillions of computers that are needed for peak-hours operations that sit idle for much of the rest of the time. They can't easily power them down - and they still take up airconditioned space - so the logical thing to do is to sell the (otherwise-wasted) CPU cycles and recoup some of that expenditure. SteveBaker (talk) 14:24, 19 March 2010 (UTC)

Some good points have already been made but the other problem is your question ignores the complexity of the market. SB already pointed out you how you can buy old computers for next to nothing and use these. But in addition, while $20,000 isn't a great amount relatively speaking, it's still enough that you'll be dumb to pay retail. (Even the retail price varies from vendor to vendor anyway.) In other words, the price will depend on what you negotiate with your supplier. This gets complicated too since at a guess you can probably (particularly if you had a larger amount to play with) negotiate a better deal (cf retail) for the PCs then you can with PS3s for example (given that modern consoles tend to be somewhat sold on the idea they'll make the money back with the games) unless perhaps you can convince Sony it'll be got advertising. Of course you could even go the home built route, but the CPUs, GPUs and assembled PCs yourself. This will generally be cheaper compared to the prebuilt solution but you're going to need someone to assemble them so it'll end up costing more unless you have slaves or are willing to do it yourself without for some reason counting the time it takes you to do it as a cost. Also, as others have stated, in the real world, what you can actually do with the assembled network will vary depending on what you've chosen and this matters. People have already given examples but something which occurs to me, you'll probably save some cost if you only give each computer the lowest amount of RAM possible. Say 256mb RAM (and get the cheapest one you can find at that, obviously no ECC and don't care if it's basically brandless because no manufacturer dared but their names on it). However a quad core with 256mb of RAM is gonna be fairly limiting. You can also probably skimp out on the PSU and motherboard, get the crappiest ones you can. Cheaper but when it may die 5 months down the road and maybe even cause numerous problems from the outset. Also you mentioned middle of the line cases. Why bother with a case? Just leave it out in the open. Saves on cost even if it complicates cooling, requires much more room, increases dust etc problems and it's at far greater risk from someone spilling something on it. These won't really gain much, and few people in the real world will consider them but if you want the best price/FLOP option, why not? Another example, the above examples reference Nvidia cards. While I don't pay that much attention to the GPGPU world, it's my understanding with the launch of the new ATI 5x00 series, these may be a better bet in price/performance terms (at least until Nvidia launches their new series). However they obviously lack CUDA support, AMD/ATI stream never really took off and OpenCL is still in its infancy and AMD/ATI are somewhat behind in supporting it anyway so it doesn't receive so much attention in the GPGPU world. A final example which occured to me, if you know the right people, perhaps you can buy stolen computers and stuff. Probably cheapish particularly if you really know the right people. Of course once you get arrested for receiving stolen goods your network will be dismantled. The precise cost is not something I expect anyone here can estimate and likely as I've already emphasised likely depends a lot on who you know and how well you know them. Even more so then your botnet idea. Heck if you aren't counting your own time as money, why not just set up a botnet yourself? P.S. While I believe you intend the US, geographical location will make a difference to a number of things as well. Nil Einne (talk) 16:00, 19 March 2010 (UTC)

I'm going to address the other replies more thoroughly, but first a quick retort to your "if you aren't counting your own time as money..." Well, if I'm not counting my own time as money, then obviously the most FLOPS per dollar (infinite, in fact) is doing the calculations in my head, though admittedly it would take a while :) 82.113.106.89 (talk) 18:44, 19 March 2010 (UTC)

In the same way that many sysadmin and budget people forget to count the cost of electricity, you're conveniently neglecting the cost of feeding and housing yourself while you do floating point math in your head. For perspective, you will die if you don't eat - so you need to pay for food, unless you are a hunter-gatherer or something. (In that case, how will you find time for floating point calculation? Your effective compute rate will be significantly diminished!) This isn't a minor detail. In the same way as you conveniently "forgot" to account for food and essentials in your human-computer concept, your budget estimate for a supercomputer is proportionately incorrect if you only measure part of the computer system cost.

Our great total cost of ownership article has a whole section on the equivalent costs for a computer. Interestingly, in today's marketplace, you will spend more money on electricity for your server, than the initial cost to purchase that server. If you want to ignore the cost of electric bills in your considerations, then you should anticipate a fun budget overrun - by a factor of 2 or more! Electric bills are the single greatest expense for a computer cluster. And don't forget air conditioners - that's another 30% of your total expense - or, again, equal to the cost of your server hardware! [6] Nimur (talk) 09:55, 20 March 2010 (UTC)

In addition, even if you argue zero cost for you, you may get a division by zero for the FLOPS per dollar which you may argue is infinite FLOPS per dollar, you're clearly not going to get infinite FLOPS since we hopefully all agree you're not likely to have an infinite lifespan anymore then the rest of us even if you automagically sort out the costs of housing, medical expenses food etc. In fact even SB's hypothetical calculator doesn't really work when we consider such real world facts since it's not going to last 1 million years. But hey at least if we talk a solar powered calculator we shouldn't (well baring catastrophic events, the sun should still be shining on earth a million years from now, strong enough to power your calculator) have to worry about the energy side of things albeit it limits the time you can use the calculator :-P Nil Einne (talk) 20:54, 21 March 2010 (UTC)

Connection reset by peer.. meaning?

What does 'connection reset by peer' even mean? If random people on the internet can just reset my connection any time they want, there's a security concern right there. Hopefully it actually has a deeper meaning than that. Preferably I shouldn't have to ask this question, but the fact of the matter is that the rest of the error codes actually make a lick of sense without a lengthy explanation. ArchabacteriaNematoda (talk) 22:46, 18 March 2010 (UTC)

"peer" doesn't mean "any random node on the Internet", it means "the host at the other end of this connection". 98.226.122.10 (talk) 23:44, 18 March 2010 (UTC)

I - a different poster - never suspected this answer. You'd think if it meant "connection reset by host" it would say "connection reset by host". I guess I have high expectations. 80.187.107.143 (talk) 23:47, 18 March 2010 (UTC)

It means exactly what it says. As 98 pointed out, not just any host on the Internet can reset your connection—it has to be your peer, the particular host you're connected to. -- Coneslayer (talk) 11:21, 19 March 2010 (UTC)

Anyone who read "connection reset by host" would understand it's the server you're connecting with. 84.153.225.240 (talk) 13:31, 19 March 2010 (UTC)

But if your computer is the server, the peer would be the client, and you'd still get the same message if the other end dropped the connection. "Peer" can refer to either end. In fact, communication that doesn't fit the client-server model is sometimes called "peer to peer". 66.127.52.47 (talk) 23:06, 19 March 2010 (UTC)

A TCP connection (which is the kind that gets the "Connection reset by peer" message) is an association between 2 hosts (a host on the Internet is basically anything that has an IP address assigned to it). From the point of view of one of those hosts, the other one is "the peer". 98.226.122.10 (talk) 23:54, 18 March 2010 (UTC)

And to expand on that a little more, the "reset" refers to the RST bit in the TCP header. "Connection reset by peer" means that a packet was received in which the RST bit was 1, indicating that the peer (i.e. the other host that you were connected to) believes the connection is no longer valid. For some random third party to cause this to happen, they would have to be able to forge the IP address and TCP sequence number, which used to be surprisingly easy to do, but defenses against that are fairly good now. Your ISP, however, can still do it to you, and don't think they won't. (I believe Comcast has used forged resets to slow down bittorrent users...) 98.226.122.10 (talk) 00:00, 19 March 2010 (UTC)

"used to"? Did Comcast stop attacking torrent use with fake resets? -- kainaw™ 00:35, 19 March 2010 (UTC)

Yes. The FCC ordered them in August 2008 to stop before the end of the year, and they did.--Chmod 777 (talk) 01:10, 19 March 2010 (UTC)

See Comcast#Network_neutrality. -- Coneslayer (talk) 16:29, 19 March 2010 (UTC)

The term can be traced back to the OSI Model (although our article doesn't use the term - instead see here). Each layer in the model must (and can only) communicate with its peer on the other side of a connection. --LarryMac | Talk 00:02, 19 March 2010 (UTC)

Our article should (sorry, SHOULD) use the term, because it is included in the official X.200 recommendation specification. An enthusiastic editor MAY edit the appropriate sections in our article, which SHOULD comply with the X.200 draft, to reflect that an OSI network model element MUST communicate with an N-PEER element and MAY NOT communicate with a node or element from a separate N-level. Nimur (talk) 10:01, 20 March 2010 (UTC)