I think the reason that if you underclock a processor by 25%, it puts off a disproportionally smaller amount of heat is because you can lower the processor voltage. The power consumed is linearly proportional to the frequency and proportional to the square of the voltage, I believe. If the heat being emitted drops by more than 25%, it is probably because your motherboard is automatically adjusting the voltage for the lower clock speed. Then again, Silicon is, of course, a semiconductor, and so the P= V^2 / R equivalence almost certainly requires adjustment if you really want to know the real answer. I'm sure it is closer to being proportional to V^2 than V, however. (In fact, it wouldn't surprise me if the exponent was > 2.)
Exactly. The problem is, however, that the two on the ends are always going to be out of phase with each other in such a way that there will be a torque around an axis perpendicular to both the crankshaft and the axis of the motion of the center piston. (I'm not exactly sure where this axis is with respect to everything else in the engine, just that it will be perpendicular to those other two axes.)
I'm 99.9% positive that is what they do. It would almost certainly be cheaper to just take a quad core die and intentionally disable one of the cores than to make different wafers for 3 core processors. (Unless the yield on the quad core dies was really low such that the probability of even having 3/4 dies work was low.)
As I understand it, back in the 386 days, Intel tried to sue AMD with regard to the instruction set architecture of the 386. (AMD produced a 386 that was in some ways superior to the one that Intel made, and they both used the same socket and were thereby almost completely interchangeable). Intel was eventually forced somehow to back down due to the US government's requirement that there be at least two sources for hardware sourced for national security (military) purposes. I am not sure if the arm twisting was achieved by the threat of not being awarded a contract to produce something for the government or by some other means, but I have heard that this was a substantial part of the reason that AMD didn't end up in serious trouble at that time. This requirement still applies in many cases, although I am not sure about the details of when it applies and when it does not.
I suspect that AMD won't be going away any more than the major US automakers have gone away when they were completely bankrupt in the past. AMD, like Chrysler in the 80s, for instance, is "too big to fail" because AMD represents the only real competition to Intel for processors in the ubiquitous computer market.
There are other candidates that could do decently well to provide a lot of the functionality that the majority of people want, but not that use the same instruction set, so all the software would have to be massively overhauled. I'm also not sure if any of the other vendors would have the fab capacity to meet demand. Of the other candidates, I'm speaking mainly about the IBM/whomever Cell processor / Power6 / Altivec derived chips or whatever they are calling the now exclusively big-iron oriented successor to the G5. Another possibility might be something built on top of the latest stuff from ARM, but a solution of that sort wouldn't solve a fab capacity problem if such a problem exists.
AMD has, in my opinion, taken prudent action by buying ATI. When you look at the Phenom architecture, there are some aspects of it that make it seem substantially better than Intel's Core architecture. The fact that it has an L3 cache that is shared by all cores makes it so that multiple threads attempting to perform shared-memory type parallel processing can avoid having to access RAM to share data. (For instance, a Core 2 Quad has two dies each of which has an L2 cache that is shared by the 2 cores on that die. If a user is running 4 processes, and they are attempting to share memory, the RAM access is likely to be the bottleneck. Intel's next generation chips are supposed to have an L3 that is shared among all cores like the AMD model.) I'm not sure why the Phenom isn't performing as well as a cursory look at the architecture makes me think it should, but then again, I don't know of any benchmarks that focus on the kinds of things that the Phenom architecture should be good at. (I'd like to see a benchmark that includes a lot of database activity or something that uses GSL in a multithreaded app to do some huge matrix calculations or something.) Perhaps the lackluster performance has to do with the much smaller cache sizes that the AMD chips use. I somewhat doubt that explains it all, however. I was originally going to post that the usually smaller set-associativity factor of the AMD caches makes them faster than the Intel ones, but I realized that I have no idea if any of this is true of the Phenom/Core 2 comparison. Does anyone know any details about how the Phenom caches work compared to the Core 2 architecture ones?
I've been using PostgreSQL pretty heavily for the last year or so, and since the version of Gentoo I was originally using had 8.0.12 as the default version, and I am now using 8.2.6, I can say that it has improved tremendously just through those two minor releases. I have also not found the documentation to be significantly lacking. It is well organized and covers everything you need for even a fairly heavily used system.
I haven't tried doing any of the more esoteric things with it that the documentation suggests can be done; one thing that I have not tried is installing any of the alternative languages that can be used to write embedded functions. (The list of available languages is pretty impressive, including some that do not initially seem to be semantically even remotely similar to PL/PgSQL.)
I've currently got a database running whose data takes up about 25 GB on disk, and queries that involve using a good portion of this data seem to have scaled well versus queries that use only a tiny fraction. In moving from one machine to another, I've used the utilities that come with it to dump and restore the database, (and I use them regularly to do backups), and they work well.
The only thing that ever makes me covet MySQL is the ability to use different storage types for tables. PostgreSQL only has one type of table storage, but any performance issues I have had have been curable in a way other than changing table types.
#postgresql on freenode is quite useful; fairly frequently, there are representatives of the core development group hanging around, and they have always been willing to answer questions and explain how stuff works.
One bit of performance advice: if you can get 3 drives, one for the data, one for the indexes, and one for the write-ahead logs, performance will improve drastically. Also, the default settings for the tweakable operational variables are not really set well for speed or efficiency. I'm not sure what kind of system they would be good for, but if you read the documentation, you can glean reasonable settings fairly easily and they make a huge difference. Overall, I have found it to be very reliable. I have never had any problems with this system that were related to the system itself; everything that I have had to deal with has been because I initially didn't know what I was doing and had created some fairly awful database designs.
I wonder why nobody has yet released a serious game that runs from a bootable DVD on a regular PC. Yes, they'd have to include a plethora of hardware drivers, but they'd have unfettered access to the hardware on the machine -- they can set up the interrupts to do as little as possible to keep latencies down and optimize everything for the performance of their game. Then they wouldn't have to worry about support issues that arise as a result of the host OS being compromised by malware and such. Game companies need to know a lot about the hardware in order to make things perform well anyway, so while they would have to put a lot of effort into making the game run natively on the hardware instead of on top of an OS, it wouldn't be impossible. People could save data on flash drives with little additional financial commitment. Or, if they so desire, they could save data on a normal hard drive. I wonder if the reason they haven't done this is because they are afraid of having to support all the hardware possibilities or if they aren't doing it because it makes one of their "partners" unhappy somehow.
To me, it seems like there are some real advantages for the gamer -- games with better performance and potentially better reliability, and they wouldn't have to worry about their regular computer use functions interfering with their ability to play games. This would be basically the same model as console gaming, except with superior hardware and improved ability to control "user experience" for the game developer.
This is essentially what Microsoft is trying to do with the XBox line, except the game developers would be holding all the cards instead of the company making the console.
I'm not necessarily advocating this model of game development; it is just interesting to consider the pros and cons of such a model becoming widely used. Anyone wish to discuss the potential ramifications of this idea taking off as a trend?
My personal opinion on this topic is somewhere in between "college is pointless" and "college is worthwhile." I tend to lean more toward "college is worthwhile" because having the experience of grinding through the formal frameworks of your field imparts some perspective and intuition fodder. By "intuition fodder", I imply that ideas all come from somewhere; having a large quantity of examples of problems solved in various ways in one's head allows connections to be made between a problem at hand and a similar problem that one has seen before. While the solution to such a problem may still not be obvious, (the problems are usually different enough that a solution cannot merely be cross-applied), a formal background provides conceptual origins from which solutions might arise. It is easy to underestimate the influence on one's abilities that such a background provides. After years of post-college experience, it isn't often easy to pinpoint where one's inspiration comes from, but in my case, when I am successful in figuring out how I came up with something, it can usually be traced back to some part of a theoretical framework. This is the reason they make you learn calculus when you are in a CS program, even though most software engineers probably don't use it directly.
Math in general is full of logical elements that can be applied to a very wide variety of things. Understanding differentiation and integration will change the way you think about a *lot* of things in a subtle manner. Any process you witness where something happens some amount per unit of time you will have more accurate intuition about. It can give you a better feel for things like "thresholds" and "break-even" points and the nature of the "end game" scenario that will apply to something that is happening, be it economic change, the eventual position of the burrito you are sliding out of your styrofoam take home box, why people frequently leave the last 1/2 inch of beer in the bottom of a glass, etc... All this despite the fact that most people rarely calculate integrals after they're done taking calculus. This sort of enhancement of basic understanding is the real reason for education. The problems you solve in school are there to give you the experiences that will enhance this understanding, not because you need to remember the answers afterward.
All this being said, I'm not sure that there is a huge advantage to going to a place like MIT over any of the numerous state schools that have 4 year programs in something. I agree with the many other posters who have mentioned that the largest difference at the undergrad level is the peers with whom you will associate. I have gone to a few different schools of widely varying prestige, (from community colleges to top 10 universities), and I can say with confidence that you will find brilliant, inspiring people at all of them, and also people who just want the piece of paper with as little fuss as they can get away with.
My advice: figure out how to find the kind of people who will inspire you to greatness, because after college, you'll have to do this anyway or you will end up sitting on the sofa watching some dumb TV show wondering how life became so lame. Then, continuously challenge yourself in school -- take the hard classes. You'll learn more and they won't actually be harder because they'll be more interesting. It is more fun to spend 10 hours trying to figure something out than it is to spend 10 hours grinding out some assignment whose intellectual content you absorbed in the first 3 minutes of studying. Those "GPA boosters" always waste a whole bunch of time because they have to make them hard somehow, but since they figure the people taking them don't have the aptitude for the real stuff, they substitute a bunch of foolishness for the stuff that will actually teach you something. (Those "physics for poets" type classes are the major culprits here -- avoid those like the plague and always take the class for people majoring in the subject if at all poss
For me, there are two main reasons that I use Linux as much as I can. The first, and most important one, is that I trust it. Having used Linux since before kernel version 1.0, I can say that I have never (knowingly) had a problem due to the fact that what I wanted to accomplish ran contrary to what the provider wanted me to be able to do with the software.
This kind of business-motivated software behavior thing happens in Windows all the time. I remember way back in the day when I was trying to run something under DR DOS on a friend's machine and, upon execution, a message was returned that said something like, "this requires MS-DOS 5.0 or greater." After breaking out a hex editor, (probably part of Norton Utilities), I discovered that this executable contained a text string that was something like "MS-DOS v.5.0 Copyright, Microsoft Corporation". I wondered if it would work on DR DOS to some degree anyway, so I modified command.com, (I believe using the same hex editor) to produce the version string that the program was looking for instead of the DR DOS version information. The program, (I think it might have been Desqview), worked fine after that. Completely fine. As in: no apparent incompatibility whatsoever. This made me realize that there is more to commercial software than just making something work. I have been suspicious of Microsoft software ever since, and I have realized that running commercial software effectively gets you involved in the software business wether you like it or not.
Running commercial software means that your success is subject to the motivations of entities that have absolutely nothing directly to do with your problem at hand. While this doesn't amount to much of a problem under casual circumstances, if you are deciding what to use for a project that is going to take up most of your time for the next couple of years, I honestly can't imagine why anyone would choose a platform that is even remotely likely to have any unnecessary obstacles to success. And before someone mentions all the ways in which these obstacles can be avoided / worked around, let me point out that they are still unnecessary and can be crossed completely off the list of potential problems when you choose to simply not get involved in something that has nothing to do with your intentions. Having to account for the unforeseen possibility of something being "nearly impossible" on a platform is much easier, (and usually more deterministic), when computer science determines what is possible instead of the vendor's business.
While Linux has had many problems over the years, I have been repeatedly impressed by how quickly show-stopping problems get fixed in the open-source community. I can't really say that I've had any serious problems with the core functional parts of a Linux system, (aside from hardware drivers); everything that really *needs* to work right does so reliably.
Hardware drivers, however, still remain as the biggest negative about open-source operating systems for me. I see this as being the most significant obstacle to open-source ubiquity that remains. (The "fit and finish" type stuff is being addressed, and while it isn't up to the level, say, of OS X, it has improved greatly, and is now quite usable. I'd prefer a Gnome or KDE based UI over the WinXP UI any day.) The biggest reason that I want to see Linux more widely used is so that hardware manufacturers get positively involved in the driver issue. Once they realize that 95% of what is required to make a usable product is the effort involved in implementation, (for instance: debugging, interoperation details, etc.., not the simple knowledge of what to implement), I expect that more will start cooperating with the open source people who are trying to write drivers for their hardware. We've seen some of this already, but obviously there are still those holding out in the belief that keeping their hardware specs private will preserve the exclusivity of whatever they have accomplished in the hardware realm. Perhaps this is a side effect of typical management belief that this data somehow has a large amount of inherent value, like the constituents of their "team".
When one reads TFA, one discovers that the article is talking about a difference in enrollment around 1% from 2006 to 2007. They also only count programs at institutions that also have PhD programs. (While most of the good schools do have PhD programs, there are some fine institutions that do not.) I'd tag this as !news; it doesn't say anything significant.
Keep in mind that most of the missiles that would be fired at an aircraft were designed a long time ago. All but the newest generation do not have CCD sensors in them. The heat seeking missiles need to detect wavelengths in the mid to far IR wavelength range (4um to ~10um) or so, and none of the normal semiconductor stuff can see those wavelengths. Until relatively recently, the detectors were all cryogenically cooled single pixel or quadrant detectors. The seeker heads on those consist of a rotating head (inside a dome transparent to the wavelengths in question -- I can't remember off the top of my head, but I think they're sapphire usually,) on to which the detector is mounted on a bracket that can be swept over an angle between straight forward and some amount off axis. When the missile is in flight and the bracket is pointing straight forward, the missile is dead on, as the signal fades, it turns the bracket off to the side a bit to sweep a cone out in front of the missile. At some rotational angle, it gets the maximum signal, and so it knows to turn in that direction. The head spins at a few thousand RPMs. Those seeker heads are marvels of engineering if you ask me. And yes, many of the missiles do not store their cryogenic material internally -- for those that are fired from a plane, the LN2 or whatever is usually piped into the missile when it is armed, and it can only stay armed for so long before the cold stuff is gone and the missile will be unable to seek. Most of the ways of confusing such a missile involve identifying the missile type, and therefore the seeker head, and pulsing a laser at it out of phase with the signal from the hot parts of the target so that the missile spirals off in the wrong direction. So no, you can't defeat it with a pinhole.:)
I seriously doubt it. I don't know who designed the BAE system, but I used to work for one of the companies competing with BAE for the contract to make the laser for this system. I know (personally) the people that designed the competing laser, and probably all of the people that know how to put one of the lasers together. I seriously doubt they are pawns of the Israeli government. There are a few systems like this that exist in various forms, (or have existed.) I'm not sure which one will be flying on that AA jet, but I'm pretty sure that what is flying is BAE's version of the same thing I worked on. It might even be the laser that I worked on bolted to a bunch of BAE hardware. I know that the entity overseeing the progress of the project is the USAF. Every time they showed up to scrutinize the project, they were all wearing USAF uniforms. Actually, the way I found out about the company originally, (which led me to applying for a job there,) was through a friend of mine who was in the USAF and was associated somehow with the project. He had visited the company for one of those meetings, and told me, rather humorously, not to let them know how I found out about the company because he kept dosing off during the meetings and didn't want me to look like the friend of the dude who kept falling asleep. (Those meetings *are* pretty long and arduous.)
I had an enlightening conversation with a Comcast CSR a couple months ago about this. I posted notes about it shortly thereafter. I'm sure they are thinking that they can get away with saying that they don't block anything because they aren't technically blocking certain types of connections or connections to certain ports. They are sending RSTs to new and existing connections inbound to the subscriber's IP after a certain number of inbound connections per unit time have been detected. (I think it is measured per unit time -- there may be some sort of weighting scheme that favors some ports or detected connection contents or something else, but for me, the relevant metric involved how many SSH connections I could make to my machine at home from school before new ones failed and all the established ones stopped responding.) There is absolutely nothing beneficial to anyone's user experience about blocking my very low bandwidth ssh sessions.
Particularly since they don't just send RSTs to the new connections, but all the existing ones as well, unencrypted P2P apps are essentially worthless if you are subject to this kind of TCP meddling. I'm not a P2P expert, but every P2P network I have ever used would blow the inbound connection quota in a few seconds. You'd be lucky to be able to do anything except leach.
See previous comment and pseudotranscript of conversation with Comcast guy here: http://yro.slashdot.org/comments.pl?sid=287993&cid=20477309
There is more to it than that, methinks. Part of the problem with the relationship between America and cars is that a huge percentage of the population of America simply can't live without a car, or at least some way to regularly get a ride. When the nation decided to build the critical transportation infrastructure around highways, the atmosphere in America was one of rather naive optimism, (people didn't think about overcrowding or limited fossil fuels or people who couldn't drive for one reason or another,) and it became normal for companies to build suburban developments in such a way that cars were the only effective mode of transportation. Now, we are stuck with that kind of system -- riding a bike anywhere except in the most bike-friendly areas involves substantial risk to life and limb because one can rarely avoid riding through places designed for cars.
Combine that with the fact that during that time, as the parent stated, people who rode bikes were assumed to be either kids, broke, or a serious bicycle racer, and we have a situation where nobody wants to fund improvements to non-car thoroughfares. I don't think this is going to get any better until driving cars becomes out of financial reach for most people.
It is fairly obvious to me at this point that advertisers are going to have to totally change how they do things. First of all, most of the people I know *cannot stand* advertisements that can't easily be ignored. The last thing I want is more distractions. Getting anything done is hard enough as it is. Reading websites that have obnoxious flashing ads on them arouses in me nothing but a desire to wage war on the asshats that are trying to prevent me from doing what I intended to do, which almost certainly had nothing to do with the advertiser or whatever they are touting.
I don't think people will be willing to pay more for TV -- the good content occupies a very small percentage of channel hours. Between PVRs and growing intolerance of distraction, I don't see any future for the traditional advertising businesses. Then again, I am also strange in that I just don't care about a lot of popular culture. My wife, on the other hand, seems to just totally ignore commercials that aren't entertaining. When I have casually complained about TV and all the commercials, she has mentioned that she has seen a growing trend in commercials that attempt to amuse viewers instead of pathetically trying to ram things into people's heads.
I have to admit, some of the Geico commercials are pretty humorous the first couple of times you see them. I suppose in the case of the "Mr. Jiggy Fly" Geico commercial, among other Geico commercials, the ad must have worked in my case -- I did end up changing car insurance companies, which I probably wouldn't have done had the commercials been just another patronizing attempt to get me to believe that product X is far superior than all other options and will probably improve my sex life.
Just an FYI to anyone in the general Seattle area who has Comcast as their cable provider, (or possibly anywhere else, but I don't have information about anywhere else): there is an unadvertized cable package called something like "the cable TV sucks ultra mega basic package" which costs, I believe, $1.99/month if you already have cable internet service. For years, I got basic cable because I wanted the cable internet service, and the price of the internet service goes up by about the same amount as the cost of basic cable if you don't subscribe to any TV packages. What I found, through asking a bunch of questions to a Comcast rep, is that there is a very cheap spartan cable package that allows you to still pay $44.99 or 49.99 or whatever it is for cable internet while only costing $1.99/month for the TV package. This seems to be by far the cheapest way to get internet access from Comcast in this area. (I haven't been particularly fond of Comcast's internet service since they started injecting RST-flagged packets into TCP sessions whenever you have exceeded your incoming connection quota for some unit of time, but it is my best realistic option at the moment.)
I agree with the sentiment expressed in many of the posts about this article, (that Andy Grove has been ignoring a lot of the very real developments that have occurred in the last few decades), but I do think he is right in that there are many aspects of the American biomedical research culture that are holding back advances significantly. One post specifically mentioned the fact that generally it is not considered OK to die during a study of a possible cure for something that is fatal anyway. This makes no sense to me. If I was in a situation where I had a few months to live, and this was more or less a certainty, I would be pissed if legal ramifications prevented me from participating in what might be a very enlightening study even if it had little chance of curing me and risked killing me. I would simply rather die a little bit sooner having given the world valuable data than die a little bit later without causing any additional benefit. As I understand it, this kind of thing happens frequently due to researchers' justifiable reluctance to risk getting sued.
I also believe, as a consumer of a few different prescription pharmaceuticals, that drug companies are deliberately not releasing potentially useful formulations because they are afraid that doing so would erode their existing drug businesses. (Think about "extended release" formulations of just about any drug currently on the market -- if there isn't one yet, it is because the patent on the original hasn't expired yet. Once the patent expires, expect that the "extended release" version will be coming out shortly, having already been tested and approved, accompanied by a massive advertising campaign that tries to tell everyone, doctors included, why the old version sucks. The fact that our government encourages this behavior is appalling. It makes voting for anything a bittersweet experience.
For me, fiction wasn't particularly inspirational. I was mostly intrigued by the automation power of computers. Since computers automate that which, at the lowest level, is mathematics, I was naturally inclined to attempt to learn mathematical techniques for tackling problems because I could then devise a machine that would tackle those problems for me with great speed and proficiency. So essentially, it was the computers themselves and their capabilities that inspired my interest in maths. Science was also fundamentally attractive to me because it presented a way to discover truth with little foreknowledge. And since it involved a methodology, it could also be automated to a substantial degree. Of course, the "wow" factor of things like fighter planes, nuclear warheads, solar cells: ("look! it does work for free!"), and postage stamp sized ICs with bajillions of internal components also contributed to my interest in science. I eventually got a BS in physics and I am currently working on the core BS requirements for a CS degree in pursuance of CS graduate school, so one could say that my interest in these subjecs is significant.
To answer your question more specifically, even though I can't stand to watch it now, Star Trek:TNG offered a look at a possible future society that was attractive to me. I suppose the general benevolence of the characters and the "mission" combined with the reverence the characters showed toward those who were knowledgable and proficient made me think that a future that transcends the usual social, economic, and political BS that our society is riddled with might be possible. I think I had a more indulgent imagination back then.
I have a cable modem connection from Comcast. Being in the habit of working from places other than home and trying to sync files and whatnot with my machines at home, I was very irritated when they started doing this. I noticed it immediately. After I had made a few SSH connections, (around 5 or so,) all of my open SSH connections would cease responding, and I couldn't successfully connect again for a substantial period of time -- maybe 30 minutes or so. After doing some research, I discovered that Comcast was using some sort of system from Sandvine. (http://www.sandvine.com/) I don't really have proof that Sandvine is providing whatever they are using to do this, but the behavior seems consistent with the type of capability that Sandvine claims to offer.
I was able to work around this by keeping an SSH session open with a tunnel to the machine in question, (thereby not triggering the Sandvine gear by only having initiated one connection over the timeframe.) One day, I was irritated enough by the situation to attempt to communicate my displeasure to Comcast and see if they would do anything about it. Being a regular customer trying to maintain the level of service I previously had from a large company whose greed steadily increases, you can guess what happened. I saved the chat session. I'm not going to post it, because they probably have it stored somewhere as well, and will sue me into oblivion if I post it, but I will provide the relevant details.
I asked the guy why they started disconnecting my TCP sessions every time I try to make more than a very small number of connections in a decently-sized period of time. He replied something like: " I know of no recent changes to the connection protocol that would do what you're describing. "
Thinking this was a slightly strange answer, I forwarded him links to all my research about Comcast's use of Sandvine gear that is advertised as having the capabilities that I suspected they were using, and he didn't say anything. I told him that this was causing me substantial difficulties and asked if they would disable this "feature" from my account. He said he couldn't do that, and that if I wanted a "remote session" that I should upgrade to the "Comcast Workplace" service that was almost twice the price of my current service. I then very clearly asked him something like: "so, basically, if I want to continue having Comcast as my internet provider, I have to pay twice as much as I am now, (already about $50/month) or I have to deal with Comcast messing with my TCP sessions?" He said something like, "Pretty much, yeah."
So there you have it. It isn't a myth -- they've reduced the value of your service without telling you, instead just letting you find out by having to debug your stuff, and then lie to you about having done it until you shove it in their face that you know what is going on.
I'd jump ship immediately, but Verizon has decided that they aren't going to support DSL at my address, so I'm either going to have to pay a lot more, live with this crap, or move. I haven't fully decided what I'm going to do yet. Needless to say, I'm extremely pissed off.
To add to the RIAA bashing, I would like to point out that most recordings are compressed, (dynamic range reduced,) when a CD master is made so that the album will still sound "good" on a crappy car stereo in heavy traffic with the windows down. To make it worse, radio stations compress the recording even more for the same reasons. (Possibly also because standard FM doesn't have the dynamic range anyway? Haven't done the math on that possibility.) Compressing the dynamics of music makes it sound a lot more like amorphous noise to me. I hate the fact that I can't even buy a non-castrated recording of anything via any form of retail. It would be nice if independant, self-distributing artists would make available uncompressed mixes of their work, so that those of us who care about nuance and detail can actually hear it via our expensive headphones.
I actually spent a considerable amount of my employer's money researching this particular question as it applied to other types of devices. (TECs, instead of CPUs, with big heat sinks on them.)
After doing a lot of tests with different thermal interface materials, (different types of thermal grease, Arctic Silver 5 included,) we concluded that Arctic Silver 5 really was the best, but not by much. The best performances of Arctic Silver 5 were less than 10% better than the best performances of regular thermal grease (Thermagon 2500). The manner of application, however, did make a large difference. We tested the razor blade method, (using it like a spatula to apply a thin, even layer,) versus the blob in the middle, (and sometimes 4 small blobs near the corners,) methods. The blob methods won every time. Whomever mentioned that the air bubbles are difficult to get rid of is correct -- if one tries this experiment with two microscope slides so that one can see the grease pattern, it quickly becomes obvious that a patient attempt at causing a central blob to spread throughout the interface moving in one direction only, (outward,) produces the best results. It is also helpful to try the experiment with the slides to see exactly how long one has to apply consistent pressure and very small movements to get the material to spread over the whole interface evenly. (Hint: it's way longer than you might think -- best results were obtained by placing a weight on the heat sink and leaving it there overnight. I suppose the spring on a CPU heat sink would provide the same results if it is stiff enough.) With regard to the optimal amount required for the interface, the manufacturer's instructions were an excellent guide. The manufacturer's spec sheets usually indicate the mass of thermal interface material required for some area of interface with a specific interface thickness. Note that the interface thickness that one is able to achieve is limited ultimately by the flatness of the surfaces being mated by the thermal interface material. Patience in making the thermal material cover and then flow out the sides of the interface is also important.
So, if you really want optimal performance, buy a heat sink with a CPU-contact side flatness that is very good. (Or find a way to polish it yourself using a granite table or some other extremely flat surface.) Use Arctic Silver 5, and *do* do the bit where the instructions tell you to rub the stuff on the heat sink and massage it around, wipe the excess off, then apply your blob to the CPU. My tests suggested that the pre-rubbing bit was not just extra work. I think this is probably because aluminium and copper both gain surface oxide layers very readily, so one is usually not contacting metal to CPU with thermal grease in between. (One is usually contacting very thin metal oxide layer to the heat source, and the oxide layer has much higher thermal impedance than the metal itself.) Arctic Silver 5 has nitride particles in it, whose presence in the mixture I can only imagine to be useful as a very very fine abrasive that is actually hard enough to remove aluminum oxide from heat sinks. And use the blob method of thermal material deposition. Practice with microscope slides first so you can get a feel for what is happening in the interface where you can't normally see it.
Slashdot Mirror
I think the reason that if you underclock a processor by 25%, it puts off a disproportionally smaller amount of heat is because you can lower the processor voltage. The power consumed is linearly proportional to the frequency and proportional to the square of the voltage, I believe. If the heat being emitted drops by more than 25%, it is probably because your motherboard is automatically adjusting the voltage for the lower clock speed. Then again, Silicon is, of course, a semiconductor, and so the P= V^2 / R equivalence almost certainly requires adjustment if you really want to know the real answer. I'm sure it is closer to being proportional to V^2 than V, however. (In fact, it wouldn't surprise me if the exponent was > 2.)
Exactly. The problem is, however, that the two on the ends are always going to be out of phase with each other in such a way that there will be a torque around an axis perpendicular to both the crankshaft and the axis of the motion of the center piston. (I'm not exactly sure where this axis is with respect to everything else in the engine, just that it will be perpendicular to those other two axes.)
I'm 99.9% positive that is what they do. It would almost certainly be cheaper to just take a quad core die and intentionally disable one of the cores than to make different wafers for 3 core processors. (Unless the yield on the quad core dies was really low such that the probability of even having 3/4 dies work was low.)
As I understand it, back in the 386 days, Intel tried to sue AMD with regard to the instruction set architecture of the 386. (AMD produced a 386 that was in some ways superior to the one that Intel made, and they both used the same socket and were thereby almost completely interchangeable). Intel was eventually forced somehow to back down due to the US government's requirement that there be at least two sources for hardware sourced for national security (military) purposes. I am not sure if the arm twisting was achieved by the threat of not being awarded a contract to produce something for the government or by some other means, but I have heard that this was a substantial part of the reason that AMD didn't end up in serious trouble at that time. This requirement still applies in many cases, although I am not sure about the details of when it applies and when it does not.
I suspect that AMD won't be going away any more than the major US automakers have gone away when they were completely bankrupt in the past. AMD, like Chrysler in the 80s, for instance, is "too big to fail" because AMD represents the only real competition to Intel for processors in the ubiquitous computer market.
There are other candidates that could do decently well to provide a lot of the functionality that the majority of people want, but not that use the same instruction set, so all the software would have to be massively overhauled. I'm also not sure if any of the other vendors would have the fab capacity to meet demand. Of the other candidates, I'm speaking mainly about the IBM/whomever Cell processor / Power6 / Altivec derived chips or whatever they are calling the now exclusively big-iron oriented successor to the G5. Another possibility might be something built on top of the latest stuff from ARM, but a solution of that sort wouldn't solve a fab capacity problem if such a problem exists.
AMD has, in my opinion, taken prudent action by buying ATI. When you look at the Phenom architecture, there are some aspects of it that make it seem substantially better than Intel's Core architecture. The fact that it has an L3 cache that is shared by all cores makes it so that multiple threads attempting to perform shared-memory type parallel processing can avoid having to access RAM to share data. (For instance, a Core 2 Quad has two dies each of which has an L2 cache that is shared by the 2 cores on that die. If a user is running 4 processes, and they are attempting to share memory, the RAM access is likely to be the bottleneck. Intel's next generation chips are supposed to have an L3 that is shared among all cores like the AMD model.) I'm not sure why the Phenom isn't performing as well as a cursory look at the architecture makes me think it should, but then again, I don't know of any benchmarks that focus on the kinds of things that the Phenom architecture should be good at. (I'd like to see a benchmark that includes a lot of database activity or something that uses GSL in a multithreaded app to do some huge matrix calculations or something.) Perhaps the lackluster performance has to do with the much smaller cache sizes that the AMD chips use. I somewhat doubt that explains it all, however. I was originally going to post that the usually smaller set-associativity factor of the AMD caches makes them faster than the Intel ones, but I realized that I have no idea if any of this is true of the Phenom/Core 2 comparison. Does anyone know any details about how the Phenom caches work compared to the Core 2 architecture ones?
I've been using PostgreSQL pretty heavily for the last year or so, and since the version of Gentoo I was originally using had 8.0.12 as the default version, and I am now using 8.2.6, I can say that it has improved tremendously just through those two minor releases. I have also not found the documentation to be significantly lacking. It is well organized and covers everything you need for even a fairly heavily used system.
I haven't tried doing any of the more esoteric things with it that the documentation suggests can be done; one thing that I have not tried is installing any of the alternative languages that can be used to write embedded functions. (The list of available languages is pretty impressive, including some that do not initially seem to be semantically even remotely similar to PL/PgSQL.)
I've currently got a database running whose data takes up about 25 GB on disk, and queries that involve using a good portion of this data seem to have scaled well versus queries that use only a tiny fraction. In moving from one machine to another, I've used the utilities that come with it to dump and restore the database, (and I use them regularly to do backups), and they work well.
The only thing that ever makes me covet MySQL is the ability to use different storage types for tables. PostgreSQL only has one type of table storage, but any performance issues I have had have been curable in a way other than changing table types.
#postgresql on freenode is quite useful; fairly frequently, there are representatives of the core development group hanging around, and they have always been willing to answer questions and explain how stuff works.
One bit of performance advice: if you can get 3 drives, one for the data, one for the indexes, and one for the write-ahead logs, performance will improve drastically. Also, the default settings for the tweakable operational variables are not really set well for speed or efficiency. I'm not sure what kind of system they would be good for, but if you read the documentation, you can glean reasonable settings fairly easily and they make a huge difference. Overall, I have found it to be very reliable. I have never had any problems with this system that were related to the system itself; everything that I have had to deal with has been because I initially didn't know what I was doing and had created some fairly awful database designs.
I wonder why nobody has yet released a serious game that runs from a bootable DVD on a regular PC. Yes, they'd have to include a plethora of hardware drivers, but they'd have unfettered access to the hardware on the machine -- they can set up the interrupts to do as little as possible to keep latencies down and optimize everything for the performance of their game. Then they wouldn't have to worry about support issues that arise as a result of the host OS being compromised by malware and such. Game companies need to know a lot about the hardware in order to make things perform well anyway, so while they would have to put a lot of effort into making the game run natively on the hardware instead of on top of an OS, it wouldn't be impossible. People could save data on flash drives with little additional financial commitment. Or, if they so desire, they could save data on a normal hard drive. I wonder if the reason they haven't done this is because they are afraid of having to support all the hardware possibilities or if they aren't doing it because it makes one of their "partners" unhappy somehow.
To me, it seems like there are some real advantages for the gamer -- games with better performance and potentially better reliability, and they wouldn't have to worry about their regular computer use functions interfering with their ability to play games. This would be basically the same model as console gaming, except with superior hardware and improved ability to control "user experience" for the game developer.
This is essentially what Microsoft is trying to do with the XBox line, except the game developers would be holding all the cards instead of the company making the console.
I'm not necessarily advocating this model of game development; it is just interesting to consider the pros and cons of such a model becoming widely used. Anyone wish to discuss the potential ramifications of this idea taking off as a trend?
My personal opinion on this topic is somewhere in between "college is pointless" and "college is worthwhile." I tend to lean more toward "college is worthwhile" because having the experience of grinding through the formal frameworks of your field imparts some perspective and intuition fodder. By "intuition fodder", I imply that ideas all come from somewhere; having a large quantity of examples of problems solved in various ways in one's head allows connections to be made between a problem at hand and a similar problem that one has seen before. While the solution to such a problem may still not be obvious, (the problems are usually different enough that a solution cannot merely be cross-applied), a formal background provides conceptual origins from which solutions might arise. It is easy to underestimate the influence on one's abilities that such a background provides. After years of post-college experience, it isn't often easy to pinpoint where one's inspiration comes from, but in my case, when I am successful in figuring out how I came up with something, it can usually be traced back to some part of a theoretical framework. This is the reason they make you learn calculus when you are in a CS program, even though most software engineers probably don't use it directly.
Math in general is full of logical elements that can be applied to a very wide variety of things. Understanding differentiation and integration will change the way you think about a *lot* of things in a subtle manner. Any process you witness where something happens some amount per unit of time you will have more accurate intuition about. It can give you a better feel for things like "thresholds" and "break-even" points and the nature of the "end game" scenario that will apply to something that is happening, be it economic change, the eventual position of the burrito you are sliding out of your styrofoam take home box, why people frequently leave the last 1/2 inch of beer in the bottom of a glass, etc... All this despite the fact that most people rarely calculate integrals after they're done taking calculus. This sort of enhancement of basic understanding is the real reason for education. The problems you solve in school are there to give you the experiences that will enhance this understanding, not because you need to remember the answers afterward.
All this being said, I'm not sure that there is a huge advantage to going to a place like MIT over any of the numerous state schools that have 4 year programs in something. I agree with the many other posters who have mentioned that the largest difference at the undergrad level is the peers with whom you will associate. I have gone to a few different schools of widely varying prestige, (from community colleges to top 10 universities), and I can say with confidence that you will find brilliant, inspiring people at all of them, and also people who just want the piece of paper with as little fuss as they can get away with.
My advice: figure out how to find the kind of people who will inspire you to greatness, because after college, you'll have to do this anyway or you will end up sitting on the sofa watching some dumb TV show wondering how life became so lame. Then, continuously challenge yourself in school -- take the hard classes. You'll learn more and they won't actually be harder because they'll be more interesting. It is more fun to spend 10 hours trying to figure something out than it is to spend 10 hours grinding out some assignment whose intellectual content you absorbed in the first 3 minutes of studying. Those "GPA boosters" always waste a whole bunch of time because they have to make them hard somehow, but since they figure the people taking them don't have the aptitude for the real stuff, they substitute a bunch of foolishness for the stuff that will actually teach you something. (Those "physics for poets" type classes are the major culprits here -- avoid those like the plague and always take the class for people majoring in the subject if at all poss
Can you post some links to stuff that explain some formal verification techniques?
Strange.. They were present when I submitted. Guess I should have hit preview first.
For me, there are two main reasons that I use Linux as much as I can. The first, and most important one, is that I trust it. Having used Linux since before kernel version 1.0, I can say that I have never (knowingly) had a problem due to the fact that what I wanted to accomplish ran contrary to what the provider wanted me to be able to do with the software. This kind of business-motivated software behavior thing happens in Windows all the time. I remember way back in the day when I was trying to run something under DR DOS on a friend's machine and, upon execution, a message was returned that said something like, "this requires MS-DOS 5.0 or greater." After breaking out a hex editor, (probably part of Norton Utilities), I discovered that this executable contained a text string that was something like "MS-DOS v.5.0 Copyright, Microsoft Corporation". I wondered if it would work on DR DOS to some degree anyway, so I modified command.com, (I believe using the same hex editor) to produce the version string that the program was looking for instead of the DR DOS version information. The program, (I think it might have been Desqview), worked fine after that. Completely fine. As in: no apparent incompatibility whatsoever. This made me realize that there is more to commercial software than just making something work. I have been suspicious of Microsoft software ever since, and I have realized that running commercial software effectively gets you involved in the software business wether you like it or not. Running commercial software means that your success is subject to the motivations of entities that have absolutely nothing directly to do with your problem at hand. While this doesn't amount to much of a problem under casual circumstances, if you are deciding what to use for a project that is going to take up most of your time for the next couple of years, I honestly can't imagine why anyone would choose a platform that is even remotely likely to have any unnecessary obstacles to success. And before someone mentions all the ways in which these obstacles can be avoided / worked around, let me point out that they are still unnecessary and can be crossed completely off the list of potential problems when you choose to simply not get involved in something that has nothing to do with your intentions. Having to account for the unforeseen possibility of something being "nearly impossible" on a platform is much easier, (and usually more deterministic), when computer science determines what is possible instead of the vendor's business. While Linux has had many problems over the years, I have been repeatedly impressed by how quickly show-stopping problems get fixed in the open-source community. I can't really say that I've had any serious problems with the core functional parts of a Linux system, (aside from hardware drivers); everything that really *needs* to work right does so reliably. Hardware drivers, however, still remain as the biggest negative about open-source operating systems for me. I see this as being the most significant obstacle to open-source ubiquity that remains. (The "fit and finish" type stuff is being addressed, and while it isn't up to the level, say, of OS X, it has improved greatly, and is now quite usable. I'd prefer a Gnome or KDE based UI over the WinXP UI any day.) The biggest reason that I want to see Linux more widely used is so that hardware manufacturers get positively involved in the driver issue. Once they realize that 95% of what is required to make a usable product is the effort involved in implementation, (for instance: debugging, interoperation details, etc.., not the simple knowledge of what to implement), I expect that more will start cooperating with the open source people who are trying to write drivers for their hardware. We've seen some of this already, but obviously there are still those holding out in the belief that keeping their hardware specs private will preserve the exclusivity of whatever they have accomplished in the hardware realm. Perhaps this is a side effect of typical management belief that this data somehow has a large amount of inherent value, like the constituents of their "team".
When one reads TFA, one discovers that the article is talking about a difference in enrollment around 1% from 2006 to 2007. They also only count programs at institutions that also have PhD programs. (While most of the good schools do have PhD programs, there are some fine institutions that do not.) I'd tag this as !news; it doesn't say anything significant.
You wouldn't happen to be the same Jack9 that used to play Planetside a couple years ago are you?
Keep in mind that most of the missiles that would be fired at an aircraft were designed a long time ago. All but the newest generation do not have CCD sensors in them. The heat seeking missiles need to detect wavelengths in the mid to far IR wavelength range (4um to ~10um) or so, and none of the normal semiconductor stuff can see those wavelengths. Until relatively recently, the detectors were all cryogenically cooled single pixel or quadrant detectors. The seeker heads on those consist of a rotating head (inside a dome transparent to the wavelengths in question -- I can't remember off the top of my head, but I think they're sapphire usually,) on to which the detector is mounted on a bracket that can be swept over an angle between straight forward and some amount off axis. When the missile is in flight and the bracket is pointing straight forward, the missile is dead on, as the signal fades, it turns the bracket off to the side a bit to sweep a cone out in front of the missile. At some rotational angle, it gets the maximum signal, and so it knows to turn in that direction. The head spins at a few thousand RPMs. Those seeker heads are marvels of engineering if you ask me. And yes, many of the missiles do not store their cryogenic material internally -- for those that are fired from a plane, the LN2 or whatever is usually piped into the missile when it is armed, and it can only stay armed for so long before the cold stuff is gone and the missile will be unable to seek. Most of the ways of confusing such a missile involve identifying the missile type, and therefore the seeker head, and pulsing a laser at it out of phase with the signal from the hot parts of the target so that the missile spirals off in the wrong direction. So no, you can't defeat it with a pinhole. :)
I seriously doubt it. I don't know who designed the BAE system, but I used to work for one of the companies competing with BAE for the contract to make the laser for this system. I know (personally) the people that designed the competing laser, and probably all of the people that know how to put one of the lasers together. I seriously doubt they are pawns of the Israeli government. There are a few systems like this that exist in various forms, (or have existed.) I'm not sure which one will be flying on that AA jet, but I'm pretty sure that what is flying is BAE's version of the same thing I worked on. It might even be the laser that I worked on bolted to a bunch of BAE hardware. I know that the entity overseeing the progress of the project is the USAF. Every time they showed up to scrutinize the project, they were all wearing USAF uniforms. Actually, the way I found out about the company originally, (which led me to applying for a job there,) was through a friend of mine who was in the USAF and was associated somehow with the project. He had visited the company for one of those meetings, and told me, rather humorously, not to let them know how I found out about the company because he kept dosing off during the meetings and didn't want me to look like the friend of the dude who kept falling asleep. (Those meetings *are* pretty long and arduous.)
I had an enlightening conversation with a Comcast CSR a couple months ago about this. I posted notes about it shortly thereafter. I'm sure they are thinking that they can get away with saying that they don't block anything because they aren't technically blocking certain types of connections or connections to certain ports. They are sending RSTs to new and existing connections inbound to the subscriber's IP after a certain number of inbound connections per unit time have been detected. (I think it is measured per unit time -- there may be some sort of weighting scheme that favors some ports or detected connection contents or something else, but for me, the relevant metric involved how many SSH connections I could make to my machine at home from school before new ones failed and all the established ones stopped responding.) There is absolutely nothing beneficial to anyone's user experience about blocking my very low bandwidth ssh sessions. Particularly since they don't just send RSTs to the new connections, but all the existing ones as well, unencrypted P2P apps are essentially worthless if you are subject to this kind of TCP meddling. I'm not a P2P expert, but every P2P network I have ever used would blow the inbound connection quota in a few seconds. You'd be lucky to be able to do anything except leach. See previous comment and pseudotranscript of conversation with Comcast guy here: http://yro.slashdot.org/comments.pl?sid=287993&cid=20477309
There is more to it than that, methinks. Part of the problem with the relationship between America and cars is that a huge percentage of the population of America simply can't live without a car, or at least some way to regularly get a ride. When the nation decided to build the critical transportation infrastructure around highways, the atmosphere in America was one of rather naive optimism, (people didn't think about overcrowding or limited fossil fuels or people who couldn't drive for one reason or another,) and it became normal for companies to build suburban developments in such a way that cars were the only effective mode of transportation. Now, we are stuck with that kind of system -- riding a bike anywhere except in the most bike-friendly areas involves substantial risk to life and limb because one can rarely avoid riding through places designed for cars. Combine that with the fact that during that time, as the parent stated, people who rode bikes were assumed to be either kids, broke, or a serious bicycle racer, and we have a situation where nobody wants to fund improvements to non-car thoroughfares. I don't think this is going to get any better until driving cars becomes out of financial reach for most people.
It is fairly obvious to me at this point that advertisers are going to have to totally change how they do things. First of all, most of the people I know *cannot stand* advertisements that can't easily be ignored. The last thing I want is more distractions. Getting anything done is hard enough as it is. Reading websites that have obnoxious flashing ads on them arouses in me nothing but a desire to wage war on the asshats that are trying to prevent me from doing what I intended to do, which almost certainly had nothing to do with the advertiser or whatever they are touting. I don't think people will be willing to pay more for TV -- the good content occupies a very small percentage of channel hours. Between PVRs and growing intolerance of distraction, I don't see any future for the traditional advertising businesses. Then again, I am also strange in that I just don't care about a lot of popular culture. My wife, on the other hand, seems to just totally ignore commercials that aren't entertaining. When I have casually complained about TV and all the commercials, she has mentioned that she has seen a growing trend in commercials that attempt to amuse viewers instead of pathetically trying to ram things into people's heads. I have to admit, some of the Geico commercials are pretty humorous the first couple of times you see them. I suppose in the case of the "Mr. Jiggy Fly" Geico commercial, among other Geico commercials, the ad must have worked in my case -- I did end up changing car insurance companies, which I probably wouldn't have done had the commercials been just another patronizing attempt to get me to believe that product X is far superior than all other options and will probably improve my sex life.
Just an FYI to anyone in the general Seattle area who has Comcast as their cable provider, (or possibly anywhere else, but I don't have information about anywhere else): there is an unadvertized cable package called something like "the cable TV sucks ultra mega basic package" which costs, I believe, $1.99/month if you already have cable internet service. For years, I got basic cable because I wanted the cable internet service, and the price of the internet service goes up by about the same amount as the cost of basic cable if you don't subscribe to any TV packages. What I found, through asking a bunch of questions to a Comcast rep, is that there is a very cheap spartan cable package that allows you to still pay $44.99 or 49.99 or whatever it is for cable internet while only costing $1.99/month for the TV package. This seems to be by far the cheapest way to get internet access from Comcast in this area. (I haven't been particularly fond of Comcast's internet service since they started injecting RST-flagged packets into TCP sessions whenever you have exceeded your incoming connection quota for some unit of time, but it is my best realistic option at the moment.)
I agree with the sentiment expressed in many of the posts about this article, (that Andy Grove has been ignoring a lot of the very real developments that have occurred in the last few decades), but I do think he is right in that there are many aspects of the American biomedical research culture that are holding back advances significantly. One post specifically mentioned the fact that generally it is not considered OK to die during a study of a possible cure for something that is fatal anyway. This makes no sense to me. If I was in a situation where I had a few months to live, and this was more or less a certainty, I would be pissed if legal ramifications prevented me from participating in what might be a very enlightening study even if it had little chance of curing me and risked killing me. I would simply rather die a little bit sooner having given the world valuable data than die a little bit later without causing any additional benefit. As I understand it, this kind of thing happens frequently due to researchers' justifiable reluctance to risk getting sued. I also believe, as a consumer of a few different prescription pharmaceuticals, that drug companies are deliberately not releasing potentially useful formulations because they are afraid that doing so would erode their existing drug businesses. (Think about "extended release" formulations of just about any drug currently on the market -- if there isn't one yet, it is because the patent on the original hasn't expired yet. Once the patent expires, expect that the "extended release" version will be coming out shortly, having already been tested and approved, accompanied by a massive advertising campaign that tries to tell everyone, doctors included, why the old version sucks. The fact that our government encourages this behavior is appalling. It makes voting for anything a bittersweet experience.
For me, fiction wasn't particularly inspirational. I was mostly intrigued by the automation power of computers. Since computers automate that which, at the lowest level, is mathematics, I was naturally inclined to attempt to learn mathematical techniques for tackling problems because I could then devise a machine that would tackle those problems for me with great speed and proficiency. So essentially, it was the computers themselves and their capabilities that inspired my interest in maths. Science was also fundamentally attractive to me because it presented a way to discover truth with little foreknowledge. And since it involved a methodology, it could also be automated to a substantial degree. Of course, the "wow" factor of things like fighter planes, nuclear warheads, solar cells: ("look! it does work for free!"), and postage stamp sized ICs with bajillions of internal components also contributed to my interest in science. I eventually got a BS in physics and I am currently working on the core BS requirements for a CS degree in pursuance of CS graduate school, so one could say that my interest in these subjecs is significant.
To answer your question more specifically, even though I can't stand to watch it now, Star Trek:TNG offered a look at a possible future society that was attractive to me. I suppose the general benevolence of the characters and the "mission" combined with the reverence the characters showed toward those who were knowledgable and proficient made me think that a future that transcends the usual social, economic, and political BS that our society is riddled with might be possible. I think I had a more indulgent imagination back then.
I have a cable modem connection from Comcast. Being in the habit of working from places other than home and trying to sync files and whatnot with my machines at home, I was very irritated when they started doing this. I noticed it immediately. After I had made a few SSH connections, (around 5 or so,) all of my open SSH connections would cease responding, and I couldn't successfully connect again for a substantial period of time -- maybe 30 minutes or so. After doing some research, I discovered that Comcast was using some sort of system from Sandvine. (http://www.sandvine.com/) I don't really have proof that Sandvine is providing whatever they are using to do this, but the behavior seems consistent with the type of capability that Sandvine claims to offer.
I was able to work around this by keeping an SSH session open with a tunnel to the machine in question, (thereby not triggering the Sandvine gear by only having initiated one connection over the timeframe.) One day, I was irritated enough by the situation to attempt to communicate my displeasure to Comcast and see if they would do anything about it. Being a regular customer trying to maintain the level of service I previously had from a large company whose greed steadily increases, you can guess what happened. I saved the chat session. I'm not going to post it, because they probably have it stored somewhere as well, and will sue me into oblivion if I post it, but I will provide the relevant details.
I asked the guy why they started disconnecting my TCP sessions every time I try to make more than a very small number of connections in a decently-sized period of time. He replied something like: " I know of no recent changes to the connection protocol that would do what you're describing. "
Thinking this was a slightly strange answer, I forwarded him links to all my research about Comcast's use of Sandvine gear that is advertised as having the capabilities that I suspected they were using, and he didn't say anything. I told him that this was causing me substantial difficulties and asked if they would disable this "feature" from my account. He said he couldn't do that, and that if I wanted a "remote session" that I should upgrade to the "Comcast Workplace" service that was almost twice the price of my current service. I then very clearly asked him something like: "so, basically, if I want to continue having Comcast as my internet provider, I have to pay twice as much as I am now, (already about $50/month) or I have to deal with Comcast messing with my TCP sessions?" He said something like, "Pretty much, yeah."
So there you have it. It isn't a myth -- they've reduced the value of your service without telling you, instead just letting you find out by having to debug your stuff, and then lie to you about having done it until you shove it in their face that you know what is going on.
I'd jump ship immediately, but Verizon has decided that they aren't going to support DSL at my address, so I'm either going to have to pay a lot more, live with this crap, or move. I haven't fully decided what I'm going to do yet. Needless to say, I'm extremely pissed off.
To add to the RIAA bashing, I would like to point out that most recordings are compressed, (dynamic range reduced,) when a CD master is made so that the album will still sound "good" on a crappy car stereo in heavy traffic with the windows down. To make it worse, radio stations compress the recording even more for the same reasons. (Possibly also because standard FM doesn't have the dynamic range anyway? Haven't done the math on that possibility.) Compressing the dynamics of music makes it sound a lot more like amorphous noise to me. I hate the fact that I can't even buy a non-castrated recording of anything via any form of retail. It would be nice if independant, self-distributing artists would make available uncompressed mixes of their work, so that those of us who care about nuance and detail can actually hear it via our expensive headphones.
I actually spent a considerable amount of my employer's money researching this particular question as it applied to other types of devices. (TECs, instead of CPUs, with big heat sinks on them.) After doing a lot of tests with different thermal interface materials, (different types of thermal grease, Arctic Silver 5 included,) we concluded that Arctic Silver 5 really was the best, but not by much. The best performances of Arctic Silver 5 were less than 10% better than the best performances of regular thermal grease (Thermagon 2500). The manner of application, however, did make a large difference. We tested the razor blade method, (using it like a spatula to apply a thin, even layer,) versus the blob in the middle, (and sometimes 4 small blobs near the corners,) methods. The blob methods won every time. Whomever mentioned that the air bubbles are difficult to get rid of is correct -- if one tries this experiment with two microscope slides so that one can see the grease pattern, it quickly becomes obvious that a patient attempt at causing a central blob to spread throughout the interface moving in one direction only, (outward,) produces the best results. It is also helpful to try the experiment with the slides to see exactly how long one has to apply consistent pressure and very small movements to get the material to spread over the whole interface evenly. (Hint: it's way longer than you might think -- best results were obtained by placing a weight on the heat sink and leaving it there overnight. I suppose the spring on a CPU heat sink would provide the same results if it is stiff enough.) With regard to the optimal amount required for the interface, the manufacturer's instructions were an excellent guide. The manufacturer's spec sheets usually indicate the mass of thermal interface material required for some area of interface with a specific interface thickness. Note that the interface thickness that one is able to achieve is limited ultimately by the flatness of the surfaces being mated by the thermal interface material. Patience in making the thermal material cover and then flow out the sides of the interface is also important. So, if you really want optimal performance, buy a heat sink with a CPU-contact side flatness that is very good. (Or find a way to polish it yourself using a granite table or some other extremely flat surface.) Use Arctic Silver 5, and *do* do the bit where the instructions tell you to rub the stuff on the heat sink and massage it around, wipe the excess off, then apply your blob to the CPU. My tests suggested that the pre-rubbing bit was not just extra work. I think this is probably because aluminium and copper both gain surface oxide layers very readily, so one is usually not contacting metal to CPU with thermal grease in between. (One is usually contacting very thin metal oxide layer to the heat source, and the oxide layer has much higher thermal impedance than the metal itself.) Arctic Silver 5 has nitride particles in it, whose presence in the mixture I can only imagine to be useful as a very very fine abrasive that is actually hard enough to remove aluminum oxide from heat sinks. And use the blob method of thermal material deposition. Practice with microscope slides first so you can get a feel for what is happening in the interface where you can't normally see it.