I did have a quick look at the paper. As I understand it, certain batches of drives purchased at the same time tend to fail after roughly the same time. That leads to an overall non-Poisson distribution of failure events. However, if you have a population of 1 million drives, each of which has a 5% chance of failing at some point, and you take a few random sample of 100 pieces each, you will still get on the average 5 faults per sample, with a standard deviation of sqrt(5), according to Poisson statistics.
The question is whether taking samples according to operating temperature at 1 C intervals is sufficiently random. If each batch of drives (with a non-Poisson failure pattern) falls in exactly one temperature bin, then you are right. However, I find it highly unlikely that their operating temperatures are so close to each other. Their operating temperature is likely to depend on whether they are sitting at the bottom or at the top of a server rack. I would expect that to lead to at least a 5 C temperature spread, which means that each temperature bin represents a mixture of many different vintages, which will add a lot of randomization to the sampling.
...even though everyone assumes a Poisson distribution that's not what disk drives actually do... what both papers say is that the vintage effect is very pronounced,
That does not matter, since I did not look at the vintage data but rather at the correlation with temperature. The vintage effect would only be an issue in my statistical consideration if the temperature range 20 to 21 C were mostly of one specific vintage, while the ranges 19-20 C and 21-22 C where of a completely different vintage. That is not likely. And even then, it would only make the estimation of the standard deviation harder from the data alone, but the Google researchers did provide the error bars themselves and error bars are commonly 68% (1 stdev) or 95% (2 stdev) confidence intervals. Within one temperature interval, failures are still a Poisson process, unless you think the hard drive manufacturers deliberately put a fail-early fault in exactly every 20th drive they sell.
The paper claims "more than 100 thousand drives". But the nice thing is that you can derive the actual number from the error bars, for example those in figure 4. The data should be governed by Poisson statistics, which means that the standard deviation in the counts is equal to the square root of the count. However, their error bars seem to be about a factor 2 larger than the standard deviation, because normally around 68% of the data points should lie within one standard deviation from the "smooth curve". Let's assume the error bars are 95% confidence intervals, i.e. 2 standard deviations.
Look at the data for 20 to 21 C. It tells you that it represents a fraction 0.0135 of their total drive population, with an average failure rate of 7 +- 0.5 %. Following the reasoning above, this 7% should represent 784+-28 drives. Since these represent 7% of 1.35% of the total number of drives, we can derive that the total number of drives is 784/0.07/0.0135 = 830,000 drives. Trying the same thing for 30 to 31 C gives 826,000 drives, which seems fairly consistent.
So can we assume that Google has deployed 830,000 hard disk drives since 2001? How many servers do they have now?
I use Opera, and I regularly see the data counter in the status bar go up to several 100s of kB. I never bothered to use a packet analyzer, but I suspect there is quite a bit of CSS, javascript, and advertisements that you didn't count.
I explicitly add index,follow,archive (and will add snippet) to my pages just to make sure things get archived, indexed, etc.
Those are the defaults for the search engines, so adding those is completely redundant. Whether or not a page is indexed or not depends on other factors such as duplicate content, the number of incoming links from both the same and other websites, and the total number of pages on the site (A 100,000-page site with just one incoming link won't be fully indexed no matter what the meta tags say).
Who wants to find a short text describing what you're searching for, only to find out that I have to pay or go through some procedure to actually read the stuff?
I kinda' like some_audio.ogg.vorbis and some_video.ogg.theora.
Blahblah.vorbis.ogg would be more logical, since the music "Blahblah" is contained in Vorbis encoding, which is contained in an Ogg file format. Like you name a tar archive foobar.tar.gz rather than foobar.gz.tar.
The first person who figures out how to make a SQL server that clusters, automatically replicates, and blah blah blah
I think Wikipedia is on several mysql servers simultaneously (not to mention well over a hundred http and squid servers). But I don't know the technical details.
I'm partially mistaken. Www.google.(com|nl|fr|de|etc) all resolve to the same ip addresses (www.l.google.com) from here, but www.google.cn resolves to a different one. I was assuming that.cn would follow the same pattern as the other ones. Google has dozens of data centers over the world; apparently they store the Chinese stuff in specific data centers.
The redirection occurs at the HTTP level, not at the domain-resolver level. It turns out that whether or not you are redirected depends on the language settings in your web browser. If I put zh-tw and zh-cn on top of my list of acceptable languages, I am not redirected from google.cn to google.com. So that's why some people are and others aren't redirected.
I'm in Netherlands and google.cn redirects. I'm not sure what algorithm Google uses for determining when to redirect and when not to.
In any case the servers are the same anyway, the tld will just present a localized version. If your google redirects like mine, you can always manually insert query parameters into the URL to get the Chinese version:
parameter hl: Google interface language, e.g. en for English, zh-TW for taiwan-style chinese (traditional), zh-CN for PRC-style Chinese (simplified).
parameter lr: Language of the websites that appear in the search results. E.g. lang_en,
lang_zh-TW or lang_zh-CN.
Then you add that as a layer in photoshop, subtract it from the real image, and the non-random noise disappears.
I doubt that that will work. Once in the computer, the pixel values are not proportional to the absolute brightness, see gamma correction on wikipedia. You would need to do the substraction on linearly encoded data (12 or more bits rather than 8). Maybe photoshop can indeed do this, provided you find the right settings, but GIMP as far as I know doesn't.
However, a search for "click here" still results in a long list of download pages for various plug-ins (acrobat, shockwave, quicktime, and so on) from all those webmasters that try to googlebomb Adobe and Apple with the "click here" phrase. Coincidentally, I searched for that phrase just a few days ago and it didn't look very different.
the energy stored in a flywheel is proportional to the moment of inertia of the wheel and square of the angular velocity (or rpm). Thus to increase the capcity it is better to jack up the speed than size.
But the moment of inertia is roughly mass * diameter^2. If you increase the mass of a flywheel by increasing its diameter, it turns out that the stored energy scales with the square of the mass ass well.
Moreover, the centrifugal forces scale with angular velocity squared and with mass to the 1.5th power (for the case where you increase mass by increasing diameter). So the tensile strength is less critical in the latter case.
...just verify the forged transaction, rather than the one the customer thought he was entering.
That's of course still an issue; it's the weakest link in the chain that counts. Still, with time-limited cryptographic challenge/response verification, it requires much more effort from the attacker. With user/password or user/password/one-time-key login schemes, the weakest link is even weaker. My Dutch bank actually tells me on the login screen: "Please verify that the URL starts with "https://bankieren.rabobank.nl/". ABN Amro mentions "please check the padlock icon to verify that you are connected to ABN Amro." The same for ING bank. (We have one major bank that uses a less secure login protocol.) Compromising the entire web browser is probably harder than installing a keylogger or building a phishing site.
I was curious about the security protocol for Nordea bank and although
links on the Nordea site are currently broken (an attempt to cover
up?), I could find them on Google.
Login
demo - a fixed PIN code plus a one-time key from a scratch card.
So the scammer just needs the fixed PIN code, plus a few of the
one-time codes.
I used to have a bank account in Sweden with a different bank that
uses a cryptographic challenge/response key generator, both for
logging in and confirming a transaction. The website supplies you with
a code number that you enter, as well as a PIN code. The device uses
the code together with a secret key and the time from an internal
clock and lets you send back the data.
Banks here in the Netherlands use similar systems, often with a
generic card reader that uses a chip that is built into the bank
cards. Others send a confirmation code by SMS to a mobile phone number
that is registered to your account.
I think cryptographic systems are inherently much more secure than
predefined one-time keys. The cryptographic keys are only valid for 30
seconds and, more importantly, only for a specific
transaction. Keylogging wouldn't help the scammer; instead he would
have to take over the entire browser in order to actually display
your transaction information together with his
transaction challenge code.
I can certainly, and with ease, measure things to well under 1/256 of an inch.
Okay, apparently, all the dual-scale calipers (not micrometers) here are with a 1/256 or 1/512 inch vernier scale, which is a huge pain to read compared to something that is in a 1/1000 inch vernier scale.
Oh, and screw thread measurements make a *lot* more sense in the Imperial system than with the metric millimeter pitch measurements.
Hmm, common metric screws are very simple: M2, M3, M4, M5, M6, M8, M10, and so on. It refers to the outer diameter of the screw and that's something you can easily measure with a caliper. The pitch of the thread is not relevant because every M4 screw has the same pitch. This is unlike the imperial screw system, where there are several pitches with similar diameters. I often have to work with American (laser) equipment and it is a huge pain: you see a hole for a screw but you don't know whether your screw has the correct pitch.
And imperial hex keys come in so many very close sizes that it is nearly impossible to judge by eye whether a hex key will fit a particular hex socket.
And how about reading a caliper? It is unimaginable for me that there are people who can read a caliper with 1/256 inch accuracy as quickly as I read it with 0.1 mm accuracy.
Here is the wikipedia on Union of Concerned Scientists. They are basically ideological twins of Greenpeace - hard-line peace activists
The way you write it suggests that there is a general consensus about this, but this is what the article actually says:
Critics have called the UCS an "unlabeled left-wing activist" group.[10] UCS received an "Ideological Spectrum Rating" of "1" (Radical Left) from the Capital Research Center.[11] Activistcash.com states that the UCS "embraces an environmental agenda" and "politicizes science" itself.[12]
Critics of the Capital Research Center and Activistcash.com claim that these two groups have their own biases because they are run by conservatives [13]. Additionally, Capital Research Center does not explain the criteria used to determine an ideological rating, and Activistcash.com does not cite references in its article on the UCS.
IIRC, a typical iron-core transformer is around 85% efficient at 50/60 Hz line current.
The primary winding of any transformer carries a current in the absence of a secondary load. Ideally, it would behave as a purely inductive load. See for example the wart under my desk: it says primary 230 V, 140 mA (=32 VA), secondary 15 V, 1 A (=15 VA). Unfortunately, the primary winding for low-power transformers is made of very thin wire with a significant ohmic resistance, which makes it dissipate energy while running idle. Even though a transformer might be 85% efficient at full load (which seems low for a typical wall wart), most of the losses are constant and independent of the load. Hence at low load, the efficiency approaches 0%.
I've long wanted to install a single power supply with a low voltage bus to feed everything.
That's a bad idea, unless the equipment either is designed to be powered from a common supply, or is not interconnected in any way. Otherwise you will get ground loops in the best case (causing noise and hard-to-reproduce errors in digital transmissions), and in the worst case a shortcut that fries both the equipment and the power supply. The latter is because the ground levels of the different devices do not necessarily correspond to the minus pole of their power supplies. That's most obvious with things like audio equipment, where the ground level is halfway between the minus and the plus.
On keywords without competition, the advertiser pays $0.05 per click. I.e., you have to work hard to really cause financial harm - and probably you will trip a clickfraud filter if you do so...
I would start with reading the wattage close to the handle.
That number is the maximum power consumption. There's a thermostat inside that switches the heating element depending on the inside temperature, moisture content of the load, and the drying program. Using that number will just give you an upper limit.
60-120kHz. The flicker should not be noticeable by people. However, some people still claim to sense the flicker.
It is theoretically possible that electronic ballasts generate flicker. After all, the AC line voltage is rectified and smoothed with a capacitor before going through a circuit that rapidly switches on and off the current through the tube. Without smoothing, the voltage would vary between 0 and 230*sqrt(2)=325 V (for 230 VAC); with the smoothing capacitor, it would be vary between something like 230 and 325 V. It is well possible that this variation shows up as 100 (120) Hz in the light intensity, although the amplitude of the flicker should be much lower than with a magnetic ballast.
Slashdot Mirror
I did have a quick look at the paper. As I understand it, certain batches of drives purchased at the same time tend to fail after roughly the same time. That leads to an overall non-Poisson distribution of failure events. However, if you have a population of 1 million drives, each of which has a 5% chance of failing at some point, and you take a few random sample of 100 pieces each, you will still get on the average 5 faults per sample, with a standard deviation of sqrt(5), according to Poisson statistics.
The question is whether taking samples according to operating temperature at 1 C intervals is sufficiently random. If each batch of drives (with a non-Poisson failure pattern) falls in exactly one temperature bin, then you are right. However, I find it highly unlikely that their operating temperatures are so close to each other. Their operating temperature is likely to depend on whether they are sitting at the bottom or at the top of a server rack. I would expect that to lead to at least a 5 C temperature spread, which means that each temperature bin represents a mixture of many different vintages, which will add a lot of randomization to the sampling.
That does not matter, since I did not look at the vintage data but rather at the correlation with temperature. The vintage effect would only be an issue in my statistical consideration if the temperature range 20 to 21 C were mostly of one specific vintage, while the ranges 19-20 C and 21-22 C where of a completely different vintage. That is not likely. And even then, it would only make the estimation of the standard deviation harder from the data alone, but the Google researchers did provide the error bars themselves and error bars are commonly 68% (1 stdev) or 95% (2 stdev) confidence intervals. Within one temperature interval, failures are still a Poisson process, unless you think the hard drive manufacturers deliberately put a fail-early fault in exactly every 20th drive they sell.
The paper claims "more than 100 thousand drives". But the nice thing is that you can derive the actual number from the error bars, for example those in figure 4. The data should be governed by Poisson statistics, which means that the standard deviation in the counts is equal to the square root of the count. However, their error bars seem to be about a factor 2 larger than the standard deviation, because normally around 68% of the data points should lie within one standard deviation from the "smooth curve". Let's assume the error bars are 95% confidence intervals, i.e. 2 standard deviations.
Look at the data for 20 to 21 C. It tells you that it represents a fraction 0.0135 of their total drive population, with an average failure rate of 7 +- 0.5 %. Following the reasoning above, this 7% should represent 784+-28 drives. Since these represent 7% of 1.35% of the total number of drives, we can derive that the total number of drives is 784/0.07/0.0135 = 830,000 drives. Trying the same thing for 30 to 31 C gives 826,000 drives, which seems fairly consistent.
So can we assume that Google has deployed 830,000 hard disk drives since 2001? How many servers do they have now?
I use Opera, and I regularly see the data counter in the status bar go up to several 100s of kB. I never bothered to use a packet analyzer, but I suspect there is quite a bit of CSS, javascript, and advertisements that you didn't count.
Those are the defaults for the search engines, so adding those is completely redundant. Whether or not a page is indexed or not depends on other factors such as duplicate content, the number of incoming links from both the same and other websites, and the total number of pages on the site (A 100,000-page site with just one incoming link won't be fully indexed no matter what the meta tags say).
Try reporting these cases as a Google spam report for
- Page does not match Google's description
- Cloaked page
- Doorway pages
And "Page requires paid subscription" in the comment field. If enough people do this, these sites will be delisted.Blahblah.vorbis.ogg would be more logical, since the music "Blahblah" is contained in Vorbis encoding, which is contained in an Ogg file format. Like you name a tar archive foobar.tar.gz rather than foobar.gz.tar.
I think Wikipedia is on several mysql servers simultaneously (not to mention well over a hundred http and squid servers). But I don't know the technical details.
I'm partially mistaken. Www.google.(com|nl|fr|de|etc) all resolve to the same ip addresses (www.l.google.com) from here, but www.google.cn resolves to a different one. I was assuming that .cn would follow the same pattern as the other ones. Google has dozens of data centers over the world; apparently they store the Chinese stuff in specific data centers.
The redirection occurs at the HTTP level, not at the domain-resolver level. It turns out that whether or not you are redirected depends on the language settings in your web browser. If I put zh-tw and zh-cn on top of my list of acceptable languages, I am not redirected from google.cn to google.com. So that's why some people are and others aren't redirected.
In any case the servers are the same anyway, the tld will just present a localized version. If your google redirects like mine, you can always manually insert query parameters into the URL to get the Chinese version:
Example: http://www.google.com/search?q=Tiananmen&hl=en&lr= lang_zh-CN
Whatever censoring they were doing is more likely to be related to the ip address of the search requester rather than interface language.
I doubt that that will work. Once in the computer, the pixel values are not proportional to the absolute brightness, see gamma correction on wikipedia. You would need to do the substraction on linearly encoded data (12 or more bits rather than 8). Maybe photoshop can indeed do this, provided you find the right settings, but GIMP as far as I know doesn't.
However, a search for "click here" still results in a long list of download pages for various plug-ins (acrobat, shockwave, quicktime, and so on) from all those webmasters that try to googlebomb Adobe and Apple with the "click here" phrase. Coincidentally, I searched for that phrase just a few days ago and it didn't look very different.
SEB in Sweden, Rabo in The Netherlands.
But the moment of inertia is roughly mass * diameter^2. If you increase the mass of a flywheel by increasing its diameter, it turns out that the stored energy scales with the square of the mass ass well.
Moreover, the centrifugal forces scale with angular velocity squared and with mass to the 1.5th power (for the case where you increase mass by increasing diameter). So the tensile strength is less critical in the latter case.
That's of course still an issue; it's the weakest link in the chain that counts. Still, with time-limited cryptographic challenge/response verification, it requires much more effort from the attacker. With user/password or user/password/one-time-key login schemes, the weakest link is even weaker. My Dutch bank actually tells me on the login screen: "Please verify that the URL starts with "https://bankieren.rabobank.nl/". ABN Amro mentions "please check the padlock icon to verify that you are connected to ABN Amro." The same for ING bank. (We have one major bank that uses a less secure login protocol.) Compromising the entire web browser is probably harder than installing a keylogger or building a phishing site.
I was curious about the security protocol for Nordea bank and although links on the Nordea site are currently broken (an attempt to cover up?), I could find them on Google.
So the scammer just needs the fixed PIN code, plus a few of the one-time codes.
I used to have a bank account in Sweden with a different bank that uses a cryptographic challenge/response key generator, both for logging in and confirming a transaction. The website supplies you with a code number that you enter, as well as a PIN code. The device uses the code together with a secret key and the time from an internal clock and lets you send back the data.
Banks here in the Netherlands use similar systems, often with a generic card reader that uses a chip that is built into the bank cards. Others send a confirmation code by SMS to a mobile phone number that is registered to your account.
I think cryptographic systems are inherently much more secure than predefined one-time keys. The cryptographic keys are only valid for 30 seconds and, more importantly, only for a specific transaction. Keylogging wouldn't help the scammer; instead he would have to take over the entire browser in order to actually display your transaction information together with his transaction challenge code.
Okay, apparently, all the dual-scale calipers (not micrometers) here are with a 1/256 or 1/512 inch vernier scale, which is a huge pain to read compared to something that is in a 1/1000 inch vernier scale.
Hmm, common metric screws are very simple: M2, M3, M4, M5, M6, M8, M10, and so on. It refers to the outer diameter of the screw and that's something you can easily measure with a caliper. The pitch of the thread is not relevant because every M4 screw has the same pitch. This is unlike the imperial screw system, where there are several pitches with similar diameters. I often have to work with American (laser) equipment and it is a huge pain: you see a hole for a screw but you don't know whether your screw has the correct pitch.
And imperial hex keys come in so many very close sizes that it is nearly impossible to judge by eye whether a hex key will fit a particular hex socket.
And how about reading a caliper? It is unimaginable for me that there are people who can read a caliper with 1/256 inch accuracy as quickly as I read it with 0.1 mm accuracy.
If anything, ergs are "centimetric". An erg is 1e-7 joule. Obligatory wikipedia reference: http://en.wikipedia.org/wiki/Erg.
The way you write it suggests that there is a general consensus about this, but this is what the article actually says:
The primary winding of any transformer carries a current in the absence of a secondary load. Ideally, it would behave as a purely inductive load. See for example the wart under my desk: it says primary 230 V, 140 mA (=32 VA), secondary 15 V, 1 A (=15 VA). Unfortunately, the primary winding for low-power transformers is made of very thin wire with a significant ohmic resistance, which makes it dissipate energy while running idle. Even though a transformer might be 85% efficient at full load (which seems low for a typical wall wart), most of the losses are constant and independent of the load. Hence at low load, the efficiency approaches 0%.
That's a bad idea, unless the equipment either is designed to be powered from a common supply, or is not interconnected in any way. Otherwise you will get ground loops in the best case (causing noise and hard-to-reproduce errors in digital transmissions), and in the worst case a shortcut that fries both the equipment and the power supply. The latter is because the ground levels of the different devices do not necessarily correspond to the minus pole of their power supplies. That's most obvious with things like audio equipment, where the ground level is halfway between the minus and the plus.
On keywords without competition, the advertiser pays $0.05 per click. I.e., you have to work hard to really cause financial harm - and probably you will trip a clickfraud filter if you do so...
That number is the maximum power consumption. There's a thermostat inside that switches the heating element depending on the inside temperature, moisture content of the load, and the drying program. Using that number will just give you an upper limit.
It is theoretically possible that electronic ballasts generate flicker. After all, the AC line voltage is rectified and smoothed with a capacitor before going through a circuit that rapidly switches on and off the current through the tube. Without smoothing, the voltage would vary between 0 and 230*sqrt(2)=325 V (for 230 VAC); with the smoothing capacitor, it would be vary between something like 230 and 325 V. It is well possible that this variation shows up as 100 (120) Hz in the light intensity, although the amplitude of the flicker should be much lower than with a magnetic ballast.