I've been using 8.1 beta releases for a while now, and even compared to 8.0.x (which was really good), 8.,1 is very impressive and well worth the upgrade. Performance improvements alone are worth it (esp on SMP).
But the biggest thing to me in 8.1, which the blurb didn't mention, is native support for inheritance-based table partitioning optimizations, which is a huge performance win for large and/or ever-growing tables.
Re:you mean Redhat wont support my modified code!?
on
Open Source Not That Open?
·
· Score: 3, Informative
The same is generally true with RedHat or any of the other OSS companies. If you make some custom patches to, say, your Postfix mail server, and then you experience a bug/issue with your Apache server, they'll still help you out with the Apache problem. They'd probably help you with the modified Postfix too if you'd just keep it on the down-low.
Perhaps they saw that PostgreSQL was on the rise, and were hoping to prep themselves to pull a SCO and start trying to charge every company that uses PostgreSQL because it somehow violated some old copyright or patent on their recently acquired Ingres (and eventually figured out they couldn't do so, so now they're giving up on it). It isn't even remotely sane, and they'd have no case, but that didn't stop SCO from trying either.
I was never the type to be having a discussion about quantum phsyics that wouldn't get sneered at by real college students at the age of 8, but I was well ahead of my time, testing for post-high-school-level skills in all subjects at about that same age on the standardized US CAT/MAT/whatever tests. I'm very grateful now in my late 20's that my parents didn't push me or skip me up grades or anything like that. They *encouraged* me, by buying me whatever books or materials I wanted at various times, and a computer early on (and perhaps significantly, tried to keep me broad - encouraged music lessons and sports, etc). But they never pushed me to succeed or tried to get me to skip over grades in school. As a result I was mostly a slacker C student (do no homework, skip when possible, ace tests, impress teachers, barely pass on grade averages somehow, often with a little fudge-factor help from the teacher) in school for lack of challenge or motivation, and I gave up on college in the first semester, but I've turned out fine as a self-learner in the real world with a somewhat normal (if a bit nerdy and introverted) life. I hate to think what this kid's life is going to be like.
The last thing the free software community needs is the US government fucking it over with beauracracy and red tape and project proposals and grants, etc. The best thing the governments of the world can do to encourage and promote the free software movement is to officially adopt open standards (open protocols, open document formats, etc) for all official business. Don't screw over a good thing by trying to play parent to it. We get by fine on our own thanks.
I'd really like to see something on OpenOffice's scale and compatibility level done in more of an AJAX style, as daunting as that is, rather than see it be an XPCOM component. I don't like being locked into Firefox any more that I like being locked into IE. Ultimately we should be free of lockin everywhere, with the only requirement being that the user platform conform to certain open standards. XPCOM quibble aside, I very much look forward to the future you describe. I think the key component is that Google needs to start bundling both secure and public remote storage with their Google Accounts. Public storage can be searched and viewed by anyone, and might even include straight up static web content hosting in addition to serving heirarchies of user published files. Private storage would be securely encrypted on the client side with a key you keep on a USB device or something, and only sent over the network and stored in encrypted form. It could all be done in an open standards way with a little work.
Google can do this without it costing them as much disk space as it would seem, because a whole lot of the public content is compressable and/or redundant with the public content of other users (if 400 users store the same file to public storage, even under different filenames, they can all be stored as a reference to the same content using hash lookups). Private storage would have to be fairly limited in size, perhaps with an option to buy more space.
Sun has lost datacenter shares to Linux, not to Microsoft. Windows just isn't even in remotely the same ballpark as *nix for the kinds of things most people deploy *nix for in datacenters. I've never really heard of any significant cases of people migrating significant services from *nix to windows in the datacenter, other than "business" windows desktop services like company email, company file sharing volumes, etc. At most companies that matter, internal business services are just a small thing running in the corner somewhere compared to whatever domain-specific thing it is they really do with most of their hardware.
Even on the business desktop services side, I suspect we're (finally) seeing the light at the end of the tunnel. As more US states, foreign governments, and eventually the US feds adopt document standards like OpenDocument that OOo uses and start embracing the idea that government software must be open-source, the effect will filter down to private business. First to those that contract with the government directly, and then to businesses that in turn contract with them, etc. The net effect of that change will be that the typical corporate desktop will be running OpenOffice, Firefox, Evolution/Thunderbird/Sunbird/etc (or similar in nature/compatibility) software, and the data being interchanged will be flowing in open formats on open protocols (even if, at least initially, the desktop OS itself is still Windows).
At that point the momentum builds strongly for converting the backend business services off of Windows servers and onto Linux, and off of Windows and onto something better (maybe a future better Linux corporate desktop, or OS/X for x86, or god knows what).
From the technical details available, it seems they are actually transmitting on multiple channels at a time. They're making a parallel bus out of the airwaves, so to speak. There's a control frequency that informs the other side what to look for on which of the sidebands. It's not revolutionary, it's just a rehash of other radio ideas that have been around for a while, but rehashed in a new and potentially useful way.
So basically, we've got a relatively low-power low-bitrate transmitter, transmitting "words" split up into bits on several of these relatively low-bitrate channels at once to achieve a remarkable-sounding aggregate bitrate for the amount of power being used (or if you prefer, for the acceptable s/n ratio)
This brings to mind something I see in wired communications all the time. Seems all communication technologies (but especially the various cables and busses in computers) go through this cyclic parallel/serial dance. It goes like this:
1) We push the data rate on our serial line as high as we can go, until it's just not practical for us to make it any faster, because we're outpacing fundamental tech. So we invent a parallel technology with N (8, 16, 32, 64,...) channels operating at the old speed/N/2, giving us twice the bandwidth we had before. 2) We push the data rate on our parallel line as high as we can go, until it's just not practical for us to make it any faster, because we're outpacing the fundamental tech. So we invent a serial technology with 1 channel running at a blazingly-fast data rate compared to that serial technology. 3) Goto 1
In reality, our fundamental ability to push data (either parallel or serial) slowly gets better over time regardless of these switches. But for some reason (maybe not rational), we jump back and forth as we progress. I guess it's a kind of meme-boredom or something. Or maybe just the companies that set the standards do this to force cyclic upgrades.
But while we'll fight all day long switching back and forth between parallel and serial copper cabling for data tranmissions (and the inherent problem, the reason parallel isn't just flat out superior, is because of inductance/capacitance issues specific to putting several copper conductors in parallel)... nobody seems to be interested in parallel fiber. Fiber doesn't have any issues going parallel (other than cost). Why hasn't Cisco gone and strapped 32 fiber strands together into a special connector for proprietary 32Gig-Ethernet that sends 4 bytes at a time in parallel? It might be expensive, but you know someone would buy it versus trying to trunk 32 individual Gig-E ports.
So basically, if it takes you 5 minutes to write a brief product description, and you churn through them all day, you're making $7.80, which is better than minimum wage. Not a good proprosition if you're clueless about auto parts and have to research everything as described, but I don't think that's the intended optimal target for completing the task (although it someone's dumb enough to spend half an hour or more per description for a crappy hourly wage, they're more than welcome). The optimal target to take up that task is someone who already knows a lot about car parts. Chances are if you're an Autozone (auto parts store chain) employee, you could get most of the descriptions done in under 5 minutes with little to no side research, because you already have the domain-specific knowledge. That's the guy who will be drawn to answer that question.
So the key to making effective money at this scheme is to skip tasks that you don't think you're "better than average" at - kinda like the job marketplace in real life.
The original post never mentioned ASICs that I saw. But in any case ASIC vs FPGA isn't all that relevant to the article, whereas FPGA vs Generic CPU is very relevant (and isn't at all odd). As the post you replied to said, if you do the math and it appears you can offload an operation you would normally do on your general purpose CPU to an FPGA and get the results back sooner than you could have calculated it on the CPU, it's a win (hell even if the net times are identical, you've freed up some general purpose cpu idle time for other things).
The idea of sticking one or more FPGAs into a machine via an I/O bus certainly has merit. I think the main issue is that we don't have compiler toolchains, libraries, and kernels ready to take advantage of it in an intelligent way. The biggest problem is that the FPGA computations need to be able to fallback to the general purpose CPU, which has an entirely different instruction set. A method that might be used, for example, would be to wrap things up such that in the application source code you have two functions with identical call signatures and supposedly identical behavior - one is for the cpu, the other is for fpga offloading. Then the runtime linker and the kernel can work magic together and schedule applications any time they need on the FPGAs and dynamically cause an application to fallback to its cpu code as well.
Actually, the naming story is more complicated than that.
The SunOS 4.x series was just "SunOS", and it was more BSD-based.
The SunOS 5.x series has always also been called Solaris. In the SunOS 5.x series, SunOS is the name of either just the kernel, or the kernel plus the very basics of the operating system, depending on who you listen to, whereas the Solaris brand name refers to the entire "Operating Environment", including SunOS 5.x and a lot of other add-ons like NIS+, LDAP services, CDE, etc, etc. I don't think they have ever sold a stripped down SunOS 5.x, only a Solaris Operating Environment 2.x which happens to contain SunOS 5.x.
And then with the release of Solaris 2.7 (SunOS 5.7) they began marketing it as "Solaris 7", presumably for some dumb marketing reason. Since then that's been the case, so for any given release from 2.7 onwards, the possible alternate namings are:
Solaris 2.X = SunOS 5.X = "Solaris X"
2.7 of course turned out to be an absymal failure of an operating environment, so it's quite fitting that it was the launching point of a marketing-driven renaming policy. Whereas 2.6 was a solid stable platform for many years, building on the successes of 2.4 and 2.5.1, the 2.7 release is where they introduced transitional 64-bit support, and poorly. It wasn't until 2.8 that things settled down again. Many customers unfortunately still have "Solaris 7" installed here and there in production, and it's a nightmare to support. What we can learn from this is never buy/install the first release of an OS right after a major overhaul of its guts to support a new architecture.
For the exact same reasons that I prefer the GPL to every other open source license. If you release code under a BSD-ish license, Microsoft can co-opt your work into a proprietary product directly without playing the same open source game that you are.
He's not trying to be truly generous and nice, give me a break. With the market penetration he has in the westernized/1st world countries, his only room for expansion of revenue happens to be to open up new markets in places that happen to have serious malaria problems. Throwing some chump change at malaria research and making some PR announcements about it helps win hearts, minds, contracts, and pocketbooks in those countries.
1) What gaurantee do we have that the.mobi admins will really track the important (especially emerging) standards correctly? If a new phone comes out that supports some special new wireless web standard that they haven't heard about yet, and I design my site to the new spec, will they drop my domain?
2) Who said that domain names have anything to do with the web, or at least standard uses of http? Perhaps I want to register a.mobi domain and offer up a service based on my own custom xml api over http, which gets hit by some java software I sell for cellphones. It won't comply with their known standards, but it is a mobile phone network service. Perhaps I'm offering something completely different, like ssh service over port 443 for proxy bypassers, etc..
3) Are they really going to take a zone xfer of all the hostnames within a client.mobi domain, and portscan them all to find services that should be machine-verified for standards compliance? If so, what ports do you check and which do you not? Is it ok to offer noncompliant web-services over port 888?
DNS is a lookup service for IP addresses in general. IP addresses are used for many things besides displaying standards-compliant content for standard browsers via http on some port or other. Some people seem to think that browsers are the only thing that generate IP traffic anymore.
The Unix guys finally figure out how to move past symlinks to something better (private per-process inherited namespaces and bind() overlay mounts ala Plan 9 - coming to a Linux box near you soon), and now Windows starts implementing it for the first time (well.lnk links were kinda like half-baked symlinks, so they were halfway there before this announcement anyways).
Jesus just shut the fuck up and admit you're wrong.
There's a dude out there who will sign a contract and give anyone on the planet $1 million in hard cash if they only demonstrate anything paranormal in a truly correct scientific way. Nobody, not one single person among the innumerable charlatans on this planet, has stepped forward and made a decent effort to obtain the prize. That speaks volumes. Sometimes, abscence of proof really is proof of abscence. At the very least, it is proof that even if there are "paranormal" effects out there that nobody understands, none of the current crop of practicioners are anything but frauds.
I can't believe I'm even having this conversation on this site. Where did rationality go?
There's still a lot of fundamental work outside of proprietary drivers and keeping up with the latest specs and all that. The NUMA slab allocator in this release, for example, qualifies as that type of thing. So does the spinlock consolidation and the lockless file descriptor thing from the Changelog. The Linux kernel is a huge, huge project. We could freeze all hardware and standards in the world so that there was nothing left to do but refine the kernel we have today, and it would still take another 5+ years before the changes settled down to "oh well, that's as good as we can get it".
He's probably referring to the book by Roger Penrose and Martin Gardner (both brilliant minds, to be sure) called "The Emperor's New Mind: Concerning Computers, Minds, and the Laws of Physics". In that book, Penrose makes a rather detailed argument for the fact that the essence of human-ness, whatever that is (consciousness, creativity, free will, etc...), the thing that we are desperately trying and failing to create with strong-AI efforts (AI that *really* can operate on a human level, not the stuff that gets marketed as "AI"), depends on some quantum effects in the brain. I haven't actually read the book (but I should), only seen reviews of it.
I'm guessing he's just getting very technical about an idea that I've also always harbored about human intelligence vs AI - that a lot of the "magic" of how the human brain works boils down to have a good random number generator deep down inside every neuron that has a small chance to perturb the neuron's normal output and give you something unexpected. A quantum probability effect somewhere within the neuron would fit the bill nicely. On the scale and complexity of the brain, that small probability of "error" (as opposed to how an otherwise relatively rigid neural network would perform) could give rise to creativity, and also break loops (a problem detailed by Hofstated in the GEB and MMT books - how can you construct a peice of (sufficiently complex) software such that it will never get stuck in an infinite loop regardless of the inputs? You really can't - but a true AI needs that capability, or it will get stuck often. Small random perturbations in neuron decisions gaurantees that you won't keep re-executing the same response to the same feedbacked stimulus over and over and over)
Not to mention it's not a good idea to play with the gene pool on a global basis. A seemingly beneficial genetic fix might turn out to have unintended bad consequences that we don't realize until perhaps generations later. Imagine if we toyed with our genes to make the whole population AIDS-immune, and a few years later it turns out that this change made us highly susceptible to some other drastic and unpredictable issue. Imagine that the new issue quickly wiped virtually everyone who had the modification. This doesn't apply just to gene therapy (which would be almost impossible to do uniformly to every human alive on the planet), but also to unnatural genetic selection at birth. Gattaca-style screening to promote certain genes and discourage others could the same effect - tending over time to make the whole population a genetic monoculture (at least in the case of a few important genes, which might be all it takes to get us wiped out).
Don't get me wrong, I think that genetic experimentation and modification are the only way forward for the human race in the long run. Natural selection and evolution simply move too slowly to give us a high enough probability of truly long-term survival, and the era is upon us now where we should be taking the reigns from mother nature and directing ourselves towards a new future. But I think it is important that the future of gene-control happen in a distributed, loosely-controlled, highly-localized and private fashion. In that way, each seemingly positive genetic decision we make (say, to turn on a certain normally dormant gene in newborns and gain 30% more intelligence on average) will probably only be made to a small portion of the population initially, and spread slowly over the course of generations based on observation of it's true long term worth and of course a form of natural selection whereby those that have it tend to succeed in human society. That way if it is found that the new intelligence gene mod turns out to make us more succeptable to some new form of mad cow disease, we won't be at risk of losing such a huge portion of our population while we correct that little problem.
Part of the plan is the ability to mechanically translate Perl5 to equivalent Perl6 without any human intervention.
I take great offense at your insinuation that by using Perl I'm foolish, irresponsible, short sighted, and self serving, or that I would try to lock my employer into anything. I think you've brainwashed yourself into hating perl after seeing too many examples of bad perl. There is good perl, and there are good perl programmers. I'm nearing completion right now on an enterprise-grade perl application. It will probably finish at ~12,000 lines of code, god knows how many multiples of that the same task would take in many other languages or environments.
It's all written to a clear, consistent style. It's got correct OO where appropriate, and code re-use in the form of non-OO libraries where appropriate. It runs on several distinct *nix environments, some of which were installed a decade ago and have never been patched. It was developed in about 1.5 man years (1 real year). It's wide open, with a public (within the company) Subversion server tracking all changes, and a public Bugzilla server for defect tracking. Plans are in the works to bring other capable people in the group up to speed on the architecture so that they can extend and support it, and to make sure the company can reliably do so long after I or anyone else is gone.
I'll say it one more time, for people who are too fucking foolish, irresponsible, short-sighted, and self-serving to see past their own prejudices:
Perl is a badass language, don't blame the language for the code you've seen some crappy programmers generate with it. And definitely don't accuse me of personal and/or corporate irresponsibility, dipshit.
I can see how one can equate certain forms of self-similarity with fractals, but I don't think net/car traffic or disk failure rates qualify just because they tend to clump up in spots. Now if, for instance, there was an "event" every 5 days that caused a ~5% higher rate of failure than the average, and every 25 days (or every 5th of those events), it was 25% higher than normal (and every 125 days it was 125% higher than normal, and the pattern aggregated indefinitely like that), then I could see calling that pattern "fractal" in nature. But clumping around events like hot days doesn't sound like a fractal pattern to me.
Then again, I've never read anything about fractal patterns in failure analysis, so what do I know. Just sounds like fishy use of a cool-sounding term to me.
Slashdot Mirror
I've been using 8.1 beta releases for a while now, and even compared to 8.0.x (which was really good), 8.,1 is very impressive and well worth the upgrade. Performance improvements alone are worth it (esp on SMP).
But the biggest thing to me in 8.1, which the blurb didn't mention, is native support for inheritance-based table partitioning optimizations, which is a huge performance win for large and/or ever-growing tables.
The same is generally true with RedHat or any of the other OSS companies. If you make some custom patches to, say, your Postfix mail server, and then you experience a bug/issue with your Apache server, they'll still help you out with the Apache problem. They'd probably help you with the modified Postfix too if you'd just keep it on the down-low.
Perhaps they saw that PostgreSQL was on the rise, and were hoping to prep themselves to pull a SCO and start trying to charge every company that uses PostgreSQL because it somehow violated some old copyright or patent on their recently acquired Ingres (and eventually figured out they couldn't do so, so now they're giving up on it). It isn't even remotely sane, and they'd have no case, but that didn't stop SCO from trying either.
I was never the type to be having a discussion about quantum phsyics that wouldn't get sneered at by real college students at the age of 8, but I was well ahead of my time, testing for post-high-school-level skills in all subjects at about that same age on the standardized US CAT/MAT/whatever tests. I'm very grateful now in my late 20's that my parents didn't push me or skip me up grades or anything like that. They *encouraged* me, by buying me whatever books or materials I wanted at various times, and a computer early on (and perhaps significantly, tried to keep me broad - encouraged music lessons and sports, etc). But they never pushed me to succeed or tried to get me to skip over grades in school. As a result I was mostly a slacker C student (do no homework, skip when possible, ace tests, impress teachers, barely pass on grade averages somehow, often with a little fudge-factor help from the teacher) in school for lack of challenge or motivation, and I gave up on college in the first semester, but I've turned out fine as a self-learner in the real world with a somewhat normal (if a bit nerdy and introverted) life. I hate to think what this kid's life is going to be like.
The last thing the free software community needs is the US government fucking it over with beauracracy and red tape and project proposals and grants, etc. The best thing the governments of the world can do to encourage and promote the free software movement is to officially adopt open standards (open protocols, open document formats, etc) for all official business. Don't screw over a good thing by trying to play parent to it. We get by fine on our own thanks.
I'd really like to see something on OpenOffice's scale and compatibility level done in more of an AJAX style, as daunting as that is, rather than see it be an XPCOM component. I don't like being locked into Firefox any more that I like being locked into IE. Ultimately we should be free of lockin everywhere, with the only requirement being that the user platform conform to certain open standards. XPCOM quibble aside, I very much look forward to the future you describe. I think the key component is that Google needs to start bundling both secure and public remote storage with their Google Accounts. Public storage can be searched and viewed by anyone, and might even include straight up static web content hosting in addition to serving heirarchies of user published files. Private storage would be securely encrypted on the client side with a key you keep on a USB device or something, and only sent over the network and stored in encrypted form. It could all be done in an open standards way with a little work.
Google can do this without it costing them as much disk space as it would seem, because a whole lot of the public content is compressable and/or redundant with the public content of other users (if 400 users store the same file to public storage, even under different filenames, they can all be stored as a reference to the same content using hash lookups). Private storage would have to be fairly limited in size, perhaps with an option to buy more space.
Sun has lost datacenter shares to Linux, not to Microsoft. Windows just isn't even in remotely the same ballpark as *nix for the kinds of things most people deploy *nix for in datacenters. I've never really heard of any significant cases of people migrating significant services from *nix to windows in the datacenter, other than "business" windows desktop services like company email, company file sharing volumes, etc. At most companies that matter, internal business services are just a small thing running in the corner somewhere compared to whatever domain-specific thing it is they really do with most of their hardware.
Even on the business desktop services side, I suspect we're (finally) seeing the light at the end of the tunnel. As more US states, foreign governments, and eventually the US feds adopt document standards like OpenDocument that OOo uses and start embracing the idea that government software must be open-source, the effect will filter down to private business. First to those that contract with the government directly, and then to businesses that in turn contract with them, etc. The net effect of that change will be that the typical corporate desktop will be running OpenOffice, Firefox, Evolution/Thunderbird/Sunbird/etc (or similar in nature/compatibility) software, and the data being interchanged will be flowing in open formats on open protocols (even if, at least initially, the desktop OS itself is still Windows).
At that point the momentum builds strongly for converting the backend business services off of Windows servers and onto Linux, and off of Windows and onto something better (maybe a future better Linux corporate desktop, or OS/X for x86, or god knows what).
From the technical details available, it seems they are actually transmitting on multiple channels at a time. They're making a parallel bus out of the airwaves, so to speak. There's a control frequency that informs the other side what to look for on which of the sidebands. It's not revolutionary, it's just a rehash of other radio ideas that have been around for a while, but rehashed in a new and potentially useful way.
So basically, we've got a relatively low-power low-bitrate transmitter, transmitting "words" split up into bits on several of these relatively low-bitrate channels at once to achieve a remarkable-sounding aggregate bitrate for the amount of power being used (or if you prefer, for the acceptable s/n ratio)
This brings to mind something I see in wired communications all the time. Seems all communication technologies (but especially the various cables and busses in computers) go through this cyclic parallel/serial dance. It goes like this:
1) We push the data rate on our serial line as high as we can go, until it's just not practical for us to make it any faster, because we're outpacing fundamental tech. So we invent a parallel technology with N (8, 16, 32, 64,
2) We push the data rate on our parallel line as high as we can go, until it's just not practical for us to make it any faster, because we're outpacing the fundamental tech. So we invent a serial technology with 1 channel running at a blazingly-fast data rate compared to that serial technology.
3) Goto 1
In reality, our fundamental ability to push data (either parallel or serial) slowly gets better over time regardless of these switches. But for some reason (maybe not rational), we jump back and forth as we progress. I guess it's a kind of meme-boredom or something. Or maybe just the companies that set the standards do this to force cyclic upgrades.
But while we'll fight all day long switching back and forth between parallel and serial copper cabling for data tranmissions (and the inherent problem, the reason parallel isn't just flat out superior, is because of inductance/capacitance issues specific to putting several copper conductors in parallel)... nobody seems to be interested in parallel fiber. Fiber doesn't have any issues going parallel (other than cost). Why hasn't Cisco gone and strapped 32 fiber strands together into a special connector for proprietary 32Gig-Ethernet that sends 4 bytes at a time in parallel? It might be expensive, but you know someone would buy it versus trying to trunk 32 individual Gig-E ports.
Wow, I've wandered way off topic
So basically, if it takes you 5 minutes to write a brief product description, and you churn through them all day, you're making $7.80, which is better than minimum wage. Not a good proprosition if you're clueless about auto parts and have to research everything as described, but I don't think that's the intended optimal target for completing the task (although it someone's dumb enough to spend half an hour or more per description for a crappy hourly wage, they're more than welcome). The optimal target to take up that task is someone who already knows a lot about car parts. Chances are if you're an Autozone (auto parts store chain) employee, you could get most of the descriptions done in under 5 minutes with little to no side research, because you already have the domain-specific knowledge. That's the guy who will be drawn to answer that question.
So the key to making effective money at this scheme is to skip tasks that you don't think you're "better than average" at - kinda like the job marketplace in real life.
The original post never mentioned ASICs that I saw. But in any case ASIC vs FPGA isn't all that relevant to the article, whereas FPGA vs Generic CPU is very relevant (and isn't at all odd). As the post you replied to said, if you do the math and it appears you can offload an operation you would normally do on your general purpose CPU to an FPGA and get the results back sooner than you could have calculated it on the CPU, it's a win (hell even if the net times are identical, you've freed up some general purpose cpu idle time for other things).
The idea of sticking one or more FPGAs into a machine via an I/O bus certainly has merit. I think the main issue is that we don't have compiler toolchains, libraries, and kernels ready to take advantage of it in an intelligent way. The biggest problem is that the FPGA computations need to be able to fallback to the general purpose CPU, which has an entirely different instruction set. A method that might be used, for example, would be to wrap things up such that in the application source code you have two functions with identical call signatures and supposedly identical behavior - one is for the cpu, the other is for fpga offloading. Then the runtime linker and the kernel can work magic together and schedule applications any time they need on the FPGAs and dynamically cause an application to fallback to its cpu code as well.
Actually, the naming story is more complicated than that.
The SunOS 4.x series was just "SunOS", and it was more BSD-based.
The SunOS 5.x series has always also been called Solaris. In the SunOS 5.x series, SunOS is the name of either just the kernel, or the kernel plus the very basics of the operating system, depending on who you listen to, whereas the Solaris brand name refers to the entire "Operating Environment", including SunOS 5.x and a lot of other add-ons like NIS+, LDAP services, CDE, etc, etc. I don't think they have ever sold a stripped down SunOS 5.x, only a Solaris Operating Environment 2.x which happens to contain SunOS 5.x.
And then with the release of Solaris 2.7 (SunOS 5.7) they began marketing it as "Solaris 7", presumably for some dumb marketing reason. Since then that's been the case, so for any given release from 2.7 onwards, the possible alternate namings are:
Solaris 2.X = SunOS 5.X = "Solaris X"
2.7 of course turned out to be an absymal failure of an operating environment, so it's quite fitting that it was the launching point of a marketing-driven renaming policy. Whereas 2.6 was a solid stable platform for many years, building on the successes of 2.4 and 2.5.1, the 2.7 release is where they introduced transitional 64-bit support, and poorly. It wasn't until 2.8 that things settled down again. Many customers unfortunately still have "Solaris 7" installed here and there in production, and it's a nightmare to support. What we can learn from this is never buy/install the first release of an OS right after a major overhaul of its guts to support a new architecture.
Now we just need to work on reverse-engineering their secret ultrasonic communications so that we can find out what they plan to do with us.
For the exact same reasons that I prefer the GPL to every other open source license. If you release code under a BSD-ish license, Microsoft can co-opt your work into a proprietary product directly without playing the same open source game that you are.
He's not trying to be truly generous and nice, give me a break. With the market penetration he has in the westernized/1st world countries, his only room for expansion of revenue happens to be to open up new markets in places that happen to have serious malaria problems. Throwing some chump change at malaria research and making some PR announcements about it helps win hearts, minds, contracts, and pocketbooks in those countries.
1) What gaurantee do we have that the
2) Who said that domain names have anything to do with the web, or at least standard uses of http? Perhaps I want to register a
3) Are they really going to take a zone xfer of all the hostnames within a client
DNS is a lookup service for IP addresses in general. IP addresses are used for many things besides displaying standards-compliant content for standard browsers via http on some port or other. Some people seem to think that browsers are the only thing that generate IP traffic anymore.
Sounds like competition for MySQL then
The Unix guys finally figure out how to move past symlinks to something better (private per-process inherited namespaces and bind() overlay mounts ala Plan 9 - coming to a Linux box near you soon), and now Windows starts implementing it for the first time (well
Shouldn't that be a WORKGROUP\Administrator-Kit or something? There is no root in Windows. Stop stealing unix terminology you clowns.
Jesus just shut the fuck up and admit you're wrong.
There's a dude out there who will sign a contract and give anyone on the planet $1 million in hard cash if they only demonstrate anything paranormal in a truly correct scientific way. Nobody, not one single person among the innumerable charlatans on this planet, has stepped forward and made a decent effort to obtain the prize. That speaks volumes. Sometimes, abscence of proof really is proof of abscence. At the very least, it is proof that even if there are "paranormal" effects out there that nobody understands, none of the current crop of practicioners are anything but frauds.
I can't believe I'm even having this conversation on this site. Where did rationality go?
There's still a lot of fundamental work outside of proprietary drivers and keeping up with the latest specs and all that. The NUMA slab allocator in this release, for example, qualifies as that type of thing. So does the spinlock consolidation and the lockless file descriptor thing from the Changelog. The Linux kernel is a huge, huge project. We could freeze all hardware and standards in the world so that there was nothing left to do but refine the kernel we have today, and it would still take another 5+ years before the changes settled down to "oh well, that's as good as we can get it".
He's probably referring to the book by Roger Penrose and Martin Gardner (both brilliant minds, to be sure) called "The Emperor's New Mind: Concerning Computers, Minds, and the Laws of Physics". In that book, Penrose makes a rather detailed argument for the fact that the essence of human-ness, whatever that is (consciousness, creativity, free will, etc...), the thing that we are desperately trying and failing to create with strong-AI efforts (AI that *really* can operate on a human level, not the stuff that gets marketed as "AI"), depends on some quantum effects in the brain. I haven't actually read the book (but I should), only seen reviews of it.
I'm guessing he's just getting very technical about an idea that I've also always harbored about human intelligence vs AI - that a lot of the "magic" of how the human brain works boils down to have a good random number generator deep down inside every neuron that has a small chance to perturb the neuron's normal output and give you something unexpected. A quantum probability effect somewhere within the neuron would fit the bill nicely. On the scale and complexity of the brain, that small probability of "error" (as opposed to how an otherwise relatively rigid neural network would perform) could give rise to creativity, and also break loops (a problem detailed by Hofstated in the GEB and MMT books - how can you construct a peice of (sufficiently complex) software such that it will never get stuck in an infinite loop regardless of the inputs? You really can't - but a true AI needs that capability, or it will get stuck often. Small random perturbations in neuron decisions gaurantees that you won't keep re-executing the same response to the same feedbacked stimulus over and over and over)
Not to mention it's not a good idea to play with the gene pool on a global basis. A seemingly beneficial genetic fix might turn out to have unintended bad consequences that we don't realize until perhaps generations later. Imagine if we toyed with our genes to make the whole population AIDS-immune, and a few years later it turns out that this change made us highly susceptible to some other drastic and unpredictable issue. Imagine that the new issue quickly wiped virtually everyone who had the modification. This doesn't apply just to gene therapy (which would be almost impossible to do uniformly to every human alive on the planet), but also to unnatural genetic selection at birth. Gattaca-style screening to promote certain genes and discourage others could the same effect - tending over time to make the whole population a genetic monoculture (at least in the case of a few important genes, which might be all it takes to get us wiped out).
Don't get me wrong, I think that genetic experimentation and modification are the only way forward for the human race in the long run. Natural selection and evolution simply move too slowly to give us a high enough probability of truly long-term survival, and the era is upon us now where we should be taking the reigns from mother nature and directing ourselves towards a new future. But I think it is important that the future of gene-control happen in a distributed, loosely-controlled, highly-localized and private fashion. In that way, each seemingly positive genetic decision we make (say, to turn on a certain normally dormant gene in newborns and gain 30% more intelligence on average) will probably only be made to a small portion of the population initially, and spread slowly over the course of generations based on observation of it's true long term worth and of course a form of natural selection whereby those that have it tend to succeed in human society. That way if it is found that the new intelligence gene mod turns out to make us more succeptable to some new form of mad cow disease, we won't be at risk of losing such a huge portion of our population while we correct that little problem.
Part of the plan is the ability to mechanically translate Perl5 to equivalent Perl6 without any human intervention.
I take great offense at your insinuation that by using Perl I'm foolish, irresponsible, short sighted, and self serving, or that I would try to lock my employer into anything. I think you've brainwashed yourself into hating perl after seeing too many examples of bad perl. There is good perl, and there are good perl programmers. I'm nearing completion right now on an enterprise-grade perl application. It will probably finish at ~12,000 lines of code, god knows how many multiples of that the same task would take in many other languages or environments.
It's all written to a clear, consistent style. It's got correct OO where appropriate, and code re-use in the form of non-OO libraries where appropriate. It runs on several distinct *nix environments, some of which were installed a decade ago and have never been patched. It was developed in about 1.5 man years (1 real year). It's wide open, with a public (within the company) Subversion server tracking all changes, and a public Bugzilla server for defect tracking. Plans are in the works to bring other capable people in the group up to speed on the architecture so that they can extend and support it, and to make sure the company can reliably do so long after I or anyone else is gone.
I'll say it one more time, for people who are too fucking foolish, irresponsible, short-sighted, and self-serving to see past their own prejudices:
Perl is a badass language, don't blame the language for the code you've seen some crappy programmers generate with it. And definitely don't accuse me of personal and/or corporate irresponsibility, dipshit.
I can see how one can equate certain forms of self-similarity with fractals, but I don't think net/car traffic or disk failure rates qualify just because they tend to clump up in spots. Now if, for instance, there was an "event" every 5 days that caused a ~5% higher rate of failure than the average, and every 25 days (or every 5th of those events), it was 25% higher than normal (and every 125 days it was 125% higher than normal, and the pattern aggregated indefinitely like that), then I could see calling that pattern "fractal" in nature. But clumping around events like hot days doesn't sound like a fractal pattern to me.
Then again, I've never read anything about fractal patterns in failure analysis, so what do I know. Just sounds like fishy use of a cool-sounding term to me.
But what does that have to do with a fractal?