We have to hope/pray that Larry Ellison has a "Road to Damascus" event, and realises that a truly open Sparc system (not just CPU) might lead to Oracle being in a far stronger position than it already is, and seriously weaken his competition
I really hope not. At the moment, SPARC and Itanium are dead architectures and no one has to worry about abominations like register windows coming back. We have been able to kill a load of complexity in operating systems and compilers that had to deal with these things. The last thing that we need is return of the zombie architectures.
Have you looked at ARMv4? It lacks, for example, all atomic operations - you can't produce a multicore ARMv4 chip without some custom extensions (which you then need to support in toolchain parts that are mostly maintained by ARM employees, so good luck with that). It doesn't include an FPU. It doesn't include Thumb-2 (so your instruction density is going to suck). Add to that, LLVM doesn't support ARMv4 and I'm not sure GCC does anymore - if they do, don't expect it to be well tested, because even cheap embedded crap was ARMv5 and is moving to v6/v7.
Early versions of ARM made a lot of decisions that are quite painful to implement with a modern microarchitecture. Examples include predicated everything (including loads and stores), which has fun interactions with dependencies, load and store multiple, which basically have to be done in microcode and have very exciting interactions with exceptions, and making the PC a general-purpose register (which means that every instruction is a potential branch until you've completely decoded it and know what the destination register is).
Add on top of that the fact that the only conformance test suites for ARM are owned by ARM and are expensive to license. You're far better off starting with RISC-V than trying to implement an old version of a mainstream ISA.
There are two problems with that. The first is that negotiating even a simple ARM license is complex and time consuming. If all you want is an off-the-shelf SoC with an ARM core, then that's fine and you don't need to worry because someone else has done it, but if you want to produce a new SoC that has a generic CPU core and some domain-specific accelerator then that means you need a license and you're looking at two years to negotiate it. The second is that those ARM royalties add up quickly in low-volume markets. Micron is looking at moving to RISC-V, because they're in a very low margin business and the license fees for the 7 ARM cores that they put on the controller boards for their SSDs eat a big chunk of their profit. They don't need very fast cores, they don't care about the software ecosystem beyond having a half decent C compiler, and they don't care too much about power (the flash chips will be using a lot more than the controller).
Hopefully it costs Google more to retain that information than it's worth.
That's not how Google works. Any information that isn't valuable now, they assume may become valuable when they are able to aggregate it with other data.
So you do the same search regularly, and now Google knows where you live, where you work, and how much time you spend at work each day. Sure, no value in that information...
When you check the traffic congestion, they get probabilistic information about where you are, where you're going, and which route you'll take. It's not 100% accurate, because they don't know if you actually took that trip, but it's a good first approximation. They can then use this information both to target ads at you, to provide better traffic information to others to make them use the service, and to sell demographic information to businesses in your to and from locations.
I've never seen an online IQ test that was actually an IQ test and not some random set of questions with none of the relevant background analysis. I believe that the tests are distributed only in book form. If you really care, the books are quite cheap - they probably cost less than the value of your time sitting the test, at least.
Or maybe it would mean that you were as smart as Einstein. Or, at least, able to plagiarise Einstein, who did explain special relativity to his children and wrote down the explanation that he used. My father told me the same explanation when I was 11. There's a lot of maths, and the moment when you can work out the mass-energy equivalence formula from first principles requires a lot more maths than a typical child has, but that isn't needed to get an understanding of what Einstein showed any more than you need to understand Newton's laws to understand that he worked out how to calculate where a thrown object will land and the relationship between the mass of an object and how fast two things will fall together.
Einstein used a model of two trains moving towards each other, each with headlights on the front, and asked his children what would happen to the light. If you start with the speed of sound on a train, then you get to the answer that sound goes faster because the air is moving, so for the light to move faster you'd need some substrate to be moving. The rest falls fairly naturally out of there.
General relativity, in contrast, is horribly complex.
It's worth noting that the standard IQ tests have very low discrimination above 120 (for example, successive attempts at different IQ tests will give high variation for the same test taker). I'm therefore very suspicious of anything that attempts to correlate 120+ scores with anything if they only did one test. There are specially designed IQ tests that have high discrimination over 130 and very low discrimination below that (i.e. most people with an IQ under that are expected to get 0-1 questions right), but unless they did a sequence of tests with increasing discrimination for their range, any number above about 120 should be taken as suspicious and anything above 130 is noise.
Have you actually bought any consumer network equipment in the past decade? Most of the things I've bought handle https already. Even a cheap (under £10) TP-Link WiFi router does (via a fairly complex dance involving public DNS records). My ISP-provided router does with a self-signed cert that I have to explicitly mark as trusted (but which is then pinned). The manual config is only an issue if you're manually configuring your own intranet server, and if you're doing that then you should know what you're doing.
Slack is a bloated monstrosity that provides IRC and a few other things, using a combination of Node.js and Chromium to produce one of the largest and most memory-hungry desktop applications that you might ever need to run. Snap is Ubuntu's version of the old PC-BSD PBI installer, where each application comes with all of its dependencies and installs them in a directory so that the package maintainers don't have to worry about incompatible upgrades. The combination of the two allows Slack to consume even more resources, by not even sharing memory mappings for common libraries.
The goal of Slack is to minimise productivity, by consuming all available computing resources and all available attention. This combination allows it to consume even more resources, but unfortunately does nothing to increase the amount of time that people waste on Slack.
Neither? Really? You'd honestly prefer to have a 50% chance of having to wait longer to get your information?
Load times for web sites are now under a second - over 100ms is considered slow. I honestly couldn't care less if two sites that are otherwise equally ranked end up sending me to the one that loads in 500ms instead of 50ms.
What about other characteristics of bad sites? They can be slow, ugly, spammy, malware-laden.
A site that is spammy or malware-laden shouldn't make it into the search index at all. A site that is ugly may still have useful information - and often ugliness correlates quite strongly with utility for technical web pages, so I'd be very unhappy if a search engine decided that I wanted to look at pretty and information-light sites instead.
Nice straw man. Why is your 82-year-old father-in-law doing configuring web servers on his local network if he finds these things so difficult? Oh, right, he isn't, he's buying off-the-shelf equipment that handles this stuff automatically for him.
I wonder if that's actually true. On the one hand, they get a lot of training data for free. On the other hand, how well are the protected against adversarial training? If a botnet signs up for a million accounts, submits 20 photos with correct descriptions for 19 and incorrect ones for the remaining one, is it possible for the attacker to teach the Google algorithm some nonsense things? What if I'm more subtle and upload 100 pictures of dogs and tag them as dogs, but the pictures are carefully crafted to share more characteristics with other pictures of cars than of dogs? Can I train it to recognise all cars as dogs (last year researchers were able to alter a picture of a dog or a car by one pixel and have this system recognise it as the other). A system like this is only as good as its training set - remember that this is the same system that decided that all black people were gorillas because the only black face in its original training set of photos from Google employees was a gorilla at a zoo.
20-year-old, I might give you. Just. As long as it was a cheap and crappy machine from 20 years ago. 10 years? No chance. A 10-year-old machine is going to be at least a Core 2 Solo, which can handle line-rate TLS on a 100Mbit connection without consuming more than a fairly small amount of CPU. The RAM usage per TLS connection is tiny. It was an issue on machines with 4MB of RAM servicing a few hundred connections, but on your low-end VPS with 256MB of RAM it's trivial.
Most modern IoT devices have hardware AES, so aren't even doing most of the hard work in software, but even doing it entirely in software on something like a Cortex-M3 is very feasible at the kinds of network speeds that these devices can handle.
The last computer I used where SSL was a noticeable performance hit was a low-end 486sx. When Netflix can saturate two 40GigE network adaptors doing TLS on every connection, with commodity Intel processors, the argument that TLS is expensive needs to die.
It's due to apple's instance of posting characters like ' as unicode even if the site is not using unicode
Apple doesn't do this on sites not using unicode. Take a look at the HTML for this page and you will see a meta tag telling you that the encoding is UTF-8. The problem is that Slashdot explicitly advertises that it is unicode, but isn't.
The fact that it doesn't support unicode in 2017, when even my terminal does, is a secondary incompetence.
You really want to integrate this with the DHCP response (though that's also not authenticated in any way). The problem with.local is that names in that namespace are not guaranteed to be unique. mycomputer.local probably exists on hundreds of LANs and the point of a DNS cert is to prove that your endpoint is who it says it is.
A good first step would be for the DHCP response to include a root cert that can be used only for things on the current network. Ideally, you probably also want something integrated with mDNS so that devices that publish their names via mDNS can also publish their cert via the same mechanism and have other parties automatically reject names if the signing cert changes. Neither of these mechanisms is very secure, but they both probably better than nothing - at least they give you reasonable protection against passive eavesdroppers.
No, they issue certificates for domains, not IP addresses. If you want to get certificates for home network devices, then the simplest thing to do is set up a subdomain like home.example.com and point a public wildcard DNS record at a machine running acme-client. Configure all of the subdomains of that you want (e.g. printer.home.example.com) and have the deploy script push them to the things on your local network. On your local network, provide a DNS server via the DHCP reply which gives local addresses for printer.home.example.com, rather than the publicly routable one.
DDG uses advertising, but it decides the adverts to place based solely on the search terms, not on any tracking data. They do use cookies, but they document what every character in them means and you can opt out of them entirely, or provide the same flags in the URL.
I thought you were talking about quarantines in general, rather than in the specific case of ebola. Now that I look it up, I discover that there have been clinical trials of ebola vaccines in west Africa in the last couple of years. The husband of a friend is a tropical disease specialist and worked on the last Ebola outbreaks and described some of the operation to me: bottom line is that you're more likely to survive becoming infected in a quarantine area than out of it.
The materials costs of new clothes is tiny. The costs are either labour or capital costs of machinery. In the very poor places that are the recipients of second-hand clothes in large quantities, the cost of labour is very, very low. Shipping them fabric costs less than shipping them second-hand clothes (because it can be transported more densely in rolls) and the cost of making the fabric into clothes at the end is negligible, as is the cost of mass producing fabric. When your entire supply chain for both new and second-hand goods is dominated by the cost of transportation, there's little incentive for a second-hand market to exist.
Slashdot Mirror
We have to hope/pray that Larry Ellison has a "Road to Damascus" event, and realises that a truly open Sparc system (not just CPU) might lead to Oracle being in a far stronger position than it already is, and seriously weaken his competition
I really hope not. At the moment, SPARC and Itanium are dead architectures and no one has to worry about abominations like register windows coming back. We have been able to kill a load of complexity in operating systems and compilers that had to deal with these things. The last thing that we need is return of the zombie architectures.
Have you looked at ARMv4? It lacks, for example, all atomic operations - you can't produce a multicore ARMv4 chip without some custom extensions (which you then need to support in toolchain parts that are mostly maintained by ARM employees, so good luck with that). It doesn't include an FPU. It doesn't include Thumb-2 (so your instruction density is going to suck). Add to that, LLVM doesn't support ARMv4 and I'm not sure GCC does anymore - if they do, don't expect it to be well tested, because even cheap embedded crap was ARMv5 and is moving to v6/v7.
Early versions of ARM made a lot of decisions that are quite painful to implement with a modern microarchitecture. Examples include predicated everything (including loads and stores), which has fun interactions with dependencies, load and store multiple, which basically have to be done in microcode and have very exciting interactions with exceptions, and making the PC a general-purpose register (which means that every instruction is a potential branch until you've completely decoded it and know what the destination register is).
Add on top of that the fact that the only conformance test suites for ARM are owned by ARM and are expensive to license. You're far better off starting with RISC-V than trying to implement an old version of a mainstream ISA.
There are two problems with that. The first is that negotiating even a simple ARM license is complex and time consuming. If all you want is an off-the-shelf SoC with an ARM core, then that's fine and you don't need to worry because someone else has done it, but if you want to produce a new SoC that has a generic CPU core and some domain-specific accelerator then that means you need a license and you're looking at two years to negotiate it. The second is that those ARM royalties add up quickly in low-volume markets. Micron is looking at moving to RISC-V, because they're in a very low margin business and the license fees for the 7 ARM cores that they put on the controller boards for their SSDs eat a big chunk of their profit. They don't need very fast cores, they don't care about the software ecosystem beyond having a half decent C compiler, and they don't care too much about power (the flash chips will be using a lot more than the controller).
Hopefully it costs Google more to retain that information than it's worth.
That's not how Google works. Any information that isn't valuable now, they assume may become valuable when they are able to aggregate it with other data.
So you do the same search regularly, and now Google knows where you live, where you work, and how much time you spend at work each day. Sure, no value in that information...
China is a lot more business friendly
True, in a sense, but requirements like 50% state ownership are not very attractive to a lot of companies.
When you check the traffic congestion, they get probabilistic information about where you are, where you're going, and which route you'll take. It's not 100% accurate, because they don't know if you actually took that trip, but it's a good first approximation. They can then use this information both to target ads at you, to provide better traffic information to others to make them use the service, and to sell demographic information to businesses in your to and from locations.
I've never seen an online IQ test that was actually an IQ test and not some random set of questions with none of the relevant background analysis. I believe that the tests are distributed only in book form. If you really care, the books are quite cheap - they probably cost less than the value of your time sitting the test, at least.
Or maybe it would mean that you were as smart as Einstein. Or, at least, able to plagiarise Einstein, who did explain special relativity to his children and wrote down the explanation that he used. My father told me the same explanation when I was 11. There's a lot of maths, and the moment when you can work out the mass-energy equivalence formula from first principles requires a lot more maths than a typical child has, but that isn't needed to get an understanding of what Einstein showed any more than you need to understand Newton's laws to understand that he worked out how to calculate where a thrown object will land and the relationship between the mass of an object and how fast two things will fall together.
Einstein used a model of two trains moving towards each other, each with headlights on the front, and asked his children what would happen to the light. If you start with the speed of sound on a train, then you get to the answer that sound goes faster because the air is moving, so for the light to move faster you'd need some substrate to be moving. The rest falls fairly naturally out of there.
General relativity, in contrast, is horribly complex.
It's worth noting that the standard IQ tests have very low discrimination above 120 (for example, successive attempts at different IQ tests will give high variation for the same test taker). I'm therefore very suspicious of anything that attempts to correlate 120+ scores with anything if they only did one test. There are specially designed IQ tests that have high discrimination over 130 and very low discrimination below that (i.e. most people with an IQ under that are expected to get 0-1 questions right), but unless they did a sequence of tests with increasing discrimination for their range, any number above about 120 should be taken as suspicious and anything above 130 is noise.
Have you actually bought any consumer network equipment in the past decade? Most of the things I've bought handle https already. Even a cheap (under £10) TP-Link WiFi router does (via a fairly complex dance involving public DNS records). My ISP-provided router does with a self-signed cert that I have to explicitly mark as trusted (but which is then pinned). The manual config is only an issue if you're manually configuring your own intranet server, and if you're doing that then you should know what you're doing.
Currently 80% informative, 20% interesting. I was going for funny, but maybe I was closer to the mark than I thought...
Slack is a bloated monstrosity that provides IRC and a few other things, using a combination of Node.js and Chromium to produce one of the largest and most memory-hungry desktop applications that you might ever need to run. Snap is Ubuntu's version of the old PC-BSD PBI installer, where each application comes with all of its dependencies and installs them in a directory so that the package maintainers don't have to worry about incompatible upgrades. The combination of the two allows Slack to consume even more resources, by not even sharing memory mappings for common libraries.
The goal of Slack is to minimise productivity, by consuming all available computing resources and all available attention. This combination allows it to consume even more resources, but unfortunately does nothing to increase the amount of time that people waste on Slack.
Neither? Really? You'd honestly prefer to have a 50% chance of having to wait longer to get your information?
Load times for web sites are now under a second - over 100ms is considered slow. I honestly couldn't care less if two sites that are otherwise equally ranked end up sending me to the one that loads in 500ms instead of 50ms.
What about other characteristics of bad sites? They can be slow, ugly, spammy, malware-laden.
A site that is spammy or malware-laden shouldn't make it into the search index at all. A site that is ugly may still have useful information - and often ugliness correlates quite strongly with utility for technical web pages, so I'd be very unhappy if a search engine decided that I wanted to look at pretty and information-light sites instead.
Nice straw man. Why is your 82-year-old father-in-law doing configuring web servers on his local network if he finds these things so difficult? Oh, right, he isn't, he's buying off-the-shelf equipment that handles this stuff automatically for him.
I wonder if that's actually true. On the one hand, they get a lot of training data for free. On the other hand, how well are the protected against adversarial training? If a botnet signs up for a million accounts, submits 20 photos with correct descriptions for 19 and incorrect ones for the remaining one, is it possible for the attacker to teach the Google algorithm some nonsense things? What if I'm more subtle and upload 100 pictures of dogs and tag them as dogs, but the pictures are carefully crafted to share more characteristics with other pictures of cars than of dogs? Can I train it to recognise all cars as dogs (last year researchers were able to alter a picture of a dog or a car by one pixel and have this system recognise it as the other). A system like this is only as good as its training set - remember that this is the same system that decided that all black people were gorillas because the only black face in its original training set of photos from Google employees was a gorilla at a zoo.
20-year-old, I might give you. Just. As long as it was a cheap and crappy machine from 20 years ago. 10 years? No chance. A 10-year-old machine is going to be at least a Core 2 Solo, which can handle line-rate TLS on a 100Mbit connection without consuming more than a fairly small amount of CPU. The RAM usage per TLS connection is tiny. It was an issue on machines with 4MB of RAM servicing a few hundred connections, but on your low-end VPS with 256MB of RAM it's trivial.
Most modern IoT devices have hardware AES, so aren't even doing most of the hard work in software, but even doing it entirely in software on something like a Cortex-M3 is very feasible at the kinds of network speeds that these devices can handle.
The last computer I used where SSL was a noticeable performance hit was a low-end 486sx. When Netflix can saturate two 40GigE network adaptors doing TLS on every connection, with commodity Intel processors, the argument that TLS is expensive needs to die.
It's due to apple's instance of posting characters like ' as unicode even if the site is not using unicode
Apple doesn't do this on sites not using unicode. Take a look at the HTML for this page and you will see a meta tag telling you that the encoding is UTF-8. The problem is that Slashdot explicitly advertises that it is unicode, but isn't.
The fact that it doesn't support unicode in 2017, when even my terminal does, is a secondary incompetence.
You really want to integrate this with the DHCP response (though that's also not authenticated in any way). The problem with .local is that names in that namespace are not guaranteed to be unique. mycomputer.local probably exists on hundreds of LANs and the point of a DNS cert is to prove that your endpoint is who it says it is.
A good first step would be for the DHCP response to include a root cert that can be used only for things on the current network. Ideally, you probably also want something integrated with mDNS so that devices that publish their names via mDNS can also publish their cert via the same mechanism and have other parties automatically reject names if the signing cert changes. Neither of these mechanisms is very secure, but they both probably better than nothing - at least they give you reasonable protection against passive eavesdroppers.
No, they issue certificates for domains, not IP addresses. If you want to get certificates for home network devices, then the simplest thing to do is set up a subdomain like home.example.com and point a public wildcard DNS record at a machine running acme-client. Configure all of the subdomains of that you want (e.g. printer.home.example.com) and have the deploy script push them to the things on your local network. On your local network, provide a DNS server via the DHCP reply which gives local addresses for printer.home.example.com, rather than the publicly routable one.
Neither, you put them in the data set with equal rank and then in the UI pick a random ordering for each client.
DDG uses advertising, but it decides the adverts to place based solely on the search terms, not on any tracking data. They do use cookies, but they document what every character in them means and you can opt out of them entirely, or provide the same flags in the URL.
I thought you were talking about quarantines in general, rather than in the specific case of ebola. Now that I look it up, I discover that there have been clinical trials of ebola vaccines in west Africa in the last couple of years. The husband of a friend is a tropical disease specialist and worked on the last Ebola outbreaks and described some of the operation to me: bottom line is that you're more likely to survive becoming infected in a quarantine area than out of it.
The materials costs of new clothes is tiny. The costs are either labour or capital costs of machinery. In the very poor places that are the recipients of second-hand clothes in large quantities, the cost of labour is very, very low. Shipping them fabric costs less than shipping them second-hand clothes (because it can be transported more densely in rolls) and the cost of making the fabric into clothes at the end is negligible, as is the cost of mass producing fabric. When your entire supply chain for both new and second-hand goods is dominated by the cost of transportation, there's little incentive for a second-hand market to exist.