That's only part of the story. Ford was in favour of the bailouts, even though they didn't take them, because if GM and Chrysler had gone out of business then a lot of other companies that supplied parts to GM and Chrysler would also have gone out of business. Many of them also supplied parts to Ford, so Ford would have suffered a sudden supply chain shortage and they didn't think that they'd be able to survive this. That's what 'too big to fail' means: that if the company fails, then it will affect significantly more than just that one company. Ford was in the strange position of not being able to survive the sudden failure of their competition.
I don't think I've ever published in a journal that didn't have proofreaders. They're about the only service of value that they offer - the peer reviews are all done by volunteers (and I've been on both sides of that), but the proof reading has to be done by paid staff. On the other hand, I know several freelance proofreaders and they're really not very expensive, in comparison with the claimed publication costs.
Then holy crap is he good. (Field-dependent, of course.)
Contributing to two journal papers in a year doesn't mean writing them. It depends a lot on the structure of the research group. Some favour having PhD students work on something by themselves with some input from their advisor, others have them working in larger groups, or focussing on their own part of a big project but contributing to many others.
I know there's extra costs and all, but I don't earn near that as a postdoc.... Where is this that you pay that kind of money for a student?
A rule of thumb for a PhD student is that they cost about twice as much as their stipend. This includes expenses for travel, their fees, and so on. And if you don't earn near that as a postdoc, you might want to consider moving universities. It's about half way up our junior postdoc salary scale, which starts at about $42K at current exchange rates.
When were these 'good old days'? There's a story from shortly after the founding of Sun. They got a visit from IBM with a set of patents that they claimed Sun infringed. They sat the patent lawyers down and explained why for each patent it was either invalid or didn't apply. The Nazgul replied that they were probably right, but they could come back with another seven patents that Sun did infringe, and fighting them in court would be far more expensive than Sun could afford. Sun signed a cross-licensing agreement with IBM. This was the early '80s.
That would be nice. In the UK, you can usually pick things up from the depot. The post office charges 50p extra to pick it up from your local post office, but lets you pick it up from the depot for free. The depots tend to be out of town, typically a 20 minute drive and totally inaccessible if you don't have a car.
You both make it sound like leaving packages outside a house is normal behaviour. I've only seen this happen once, to a former neighbour, and it was stolen. The solution was simple: call the delivery company, ask them to show you your signature on the form, when they can't ask them to replace the item. If it's not signed for by you, there's no proof of delivery and so the responsibility for the loss belongs the last person in the chain who can verify receipt, which is the delivery company.
A big problem for companies like UPS and FedEx is that they offer a service that's convenient for senders, not receivers. When I order a delivery from one of my local supermarkets, I specify a one or two hour timeslot when my delivery will take place. If I can't take time away from work, then I can schedule it in a weekend or evening. On the other hand, if I order something from Amazon then they will send it to me by a service that gives me a two day window. This doesn't actually end up being cheaper for the courier, because I'm often not in the first time they try to deliver things, so they have to come back the next day. If they could tell me in advance the one hour window when a package will arrive, that would be significantly more convenient. If they could let me choose a time, that would be even better.
While true, so is an abacus. Printing even a simple integrated circuit is a long way beyond the capabilities of a state of the art 3D printer. The original 8086 was produced on a 3um process. A 3D printer with that kind of accuracy simply doesn't exist, and that's assuming you could use the same printing techniques for ICs (the chemicals involved are quite corrosive and so you'd need some quite specialised equipment). One of the reasons Intel has maintained such a dominant position is that IC production is one of the fields where economies of scale really, really matter.
Insightful? No. I have an FPGA dev board on my desk. The dev board costs around $8K, the FPGA alone can be bought in small quantities for about $4K. We use it for experimental processor design. It can run our MIPS64-based softcore at about 100MHz (drawing around 40W) and there's enough space on die for 4-8 cores. You can't run a processor on one that is competitive with a cheap ARM processor (except if you configure the FPGA for a single algorithm, then you can't run general-purpose code on it), let alone one with 'all the power and capabilities we want'. FPGAs are cool, but they're no substitute for ASICs.
Most projects are bad ideas or don't appeal widely enough to be worth funding. The point of Kickstarter is to cut out the middle man between people who want a product or service and people willing to provide it. It isn't meant to fund things that there is no market for, it is meant to directly connect the funding for things with the existence of a market.
This is a platform that was not designed to have a scheduler carving up resources
Uh, what? The ARM architecture was designed for Acorn's line of 32-bit desktop computers, which shipped with a multitasking OS from the start. Now, it wasn't preemptively multitasking, but the only difference between cooperative multitasking and preemptive from a hardware perspective is that you need (relatively) cheap timer interrupts to enable preemptive multitasking, and ARM has always had this.
The cost of context switching boils down to a small number of things:
The cost of delivering the timer interrupt (interrupts on ARM are cheap and have a small number of aliased registers to play with so the code in the interrupt handler can be simple)
The size of the register set that needs to be saved (ARM is 15 32-bit GPRs, which can be saved and loaded in a single multi-cycle instruction, only x86 has a smaller register set)
The cost of TLB flushes and refills required during the switch (ARM has a tagged TLB, so you only need to invalidate any TLB entries when you recycle an ASID)
In summary, the orifice that you are talking out of is not your mouth.
And it's up to well-written apps to provide hints. OS X / iOS has the NSCache interface, which allows you to store references to objects that may be destroyed in response to memory pressure (unless they are in use right now). This is important, because it allows an application to grow to use RAM when it's available, without dying horribly when it isn't.
I did some work on one of Mark's other books, so he still periodically sends me copies of any that contain snippets that I wrote. I've got the second edition lying around somewhere (I think it contains a total of about three pages of text I wrote, no idea if there's anything of mine in the third edition). Most of the stuff is pretty timeless, but as I recall this book also talks about various configuration things. One of the reasons I switched from Linux to FreeBSD is that these have a habit of changing every six months in the Linux world without actually improving.
Interactions are only a problem when the two chemicals mix. Aviation gas reacts quite easily with oxygen, but it's safe because it's stored in tanks and kept isolated from the air. Hydrogen requires a flame to ignite, just like most fuels, and because it's much lighter than air it's very safe in a balloon that isn't in an enclosed space because any leakage will escape upwards quickly (and diffuse even more quickly).
And the reason why Michele wants to take Dell private is so he can do some radical things to it. So who knows? Give the man the benefit of a doubt, grab some popcorn, and see what happens
He's already CEO and Chairman. If he can take the company in new profitable directions, then there's no incentive for the other shareholders to sell because they they'd be giving up their shares just before they become more valuable. Their only sensible choice is to get as much money as possible for the shares now, since, if it goes private, they'll never be able to own part of Dell again.
The rigidity isn't the problem, the leakage is. A normal hydrogen balloon equal pressure inside and out, but the gaps in the membrane are large enough that hydrogen molecules can pass. I suppose that making something airtight is easier than making something vacuum-tight, so if the aerogel were rigid enough to support one atmosphere of pressure on all sides, then it may give cheap dirigibles...
Not really. The problem is in the sealing. If you had a membrane that could withstand the pressure of air on one side, vacuum on the other, and not allow the air to seep in, without adding more than a negligible amount of weight, then you could just use the same material to make hydrogen balloons.
Actually, in 2003 Opera was not that obscure. The wikipedia statistics have 94.43% for IE. Netscape and Mozilla each had about 2.5% and Opera had 0.66%. Back then, on any *NIX system, Netscape was old, Mozilla was crashy (and slow), Opera was fast. On Macs, the same was roughly true, except you had IE 5.5, which had transparent PNG support (which Windows lacked), but implemented an old draft of CSS standard. On Windows you also had IE, which was ubiquitous, but Opera was the best choice technically. They also released binaries for FreeBSD and NetBSD, on a few architectures. It was sufficient of an improvement over the competition that I actually paid for it.
Why do you keep bringing up weight, when I'm talking about diameter? A typical bike helmet is not very dense at all, it is usually a plastic cover over foam. If you want to dig out the statistics, they were a Slashdot story a couple of years ago, condensed from data from the USA, but I don't remember which government department issued them.
In a traditional filesystem, yes. The mirroring happens at the block device level, and so it is completely unaware of the semantics of the filesystem and will duplicate anything, potentially overwriting good data with bad if the filesystem is corrupted. Worse, unless the drive fails catastrophically, you're liable to either duplicate single-block errors or to be unable to tell which copy of a block is the damaged one. ZFS fixes the second of these problems with block-level checksums, so it can tell which disk has errors. It also makes the mirroring infrastructure partially aware of the filesystem layout, so it shouldn't duplicate filesystem corruption, however it will happily copy user errors. For example, if your word processor corrupts a document as it saves it, then there's nothing ZFS can do about that (unless you have an earlier snapshot). And, of course, if there's a bug in the filesystem driver, all bets are off.
Mirroring, as the grandparent says, is not a substitute for proper backups. One of the most common reasons for restoring from backups is accidental deletion. Even a filesystem with 100% reliability won't protect you against this.
Oh, I'm sorry, I thought we were talking about safety. I didn't realise this was a fashion discussion. The statistics I'm quoting were published about two years ago, so the century in question would be the 21st.
Slashdot Mirror
but this? FFS
No, this ZFS. FFS is much older and does a lot less.
That's only part of the story. Ford was in favour of the bailouts, even though they didn't take them, because if GM and Chrysler had gone out of business then a lot of other companies that supplied parts to GM and Chrysler would also have gone out of business. Many of them also supplied parts to Ford, so Ford would have suffered a sudden supply chain shortage and they didn't think that they'd be able to survive this. That's what 'too big to fail' means: that if the company fails, then it will affect significantly more than just that one company. Ford was in the strange position of not being able to survive the sudden failure of their competition.
I don't think I've ever published in a journal that didn't have proofreaders. They're about the only service of value that they offer - the peer reviews are all done by volunteers (and I've been on both sides of that), but the proof reading has to be done by paid staff. On the other hand, I know several freelance proofreaders and they're really not very expensive, in comparison with the claimed publication costs.
Then holy crap is he good. (Field-dependent, of course.)
Contributing to two journal papers in a year doesn't mean writing them. It depends a lot on the structure of the research group. Some favour having PhD students work on something by themselves with some input from their advisor, others have them working in larger groups, or focussing on their own part of a big project but contributing to many others.
I know there's extra costs and all, but I don't earn near that as a postdoc.... Where is this that you pay that kind of money for a student?
A rule of thumb for a PhD student is that they cost about twice as much as their stipend. This includes expenses for travel, their fees, and so on. And if you don't earn near that as a postdoc, you might want to consider moving universities. It's about half way up our junior postdoc salary scale, which starts at about $42K at current exchange rates.
When were these 'good old days'? There's a story from shortly after the founding of Sun. They got a visit from IBM with a set of patents that they claimed Sun infringed. They sat the patent lawyers down and explained why for each patent it was either invalid or didn't apply. The Nazgul replied that they were probably right, but they could come back with another seven patents that Sun did infringe, and fighting them in court would be far more expensive than Sun could afford. Sun signed a cross-licensing agreement with IBM. This was the early '80s.
That would be nice. In the UK, you can usually pick things up from the depot. The post office charges 50p extra to pick it up from your local post office, but lets you pick it up from the depot for free. The depots tend to be out of town, typically a 20 minute drive and totally inaccessible if you don't have a car.
You both make it sound like leaving packages outside a house is normal behaviour. I've only seen this happen once, to a former neighbour, and it was stolen. The solution was simple: call the delivery company, ask them to show you your signature on the form, when they can't ask them to replace the item. If it's not signed for by you, there's no proof of delivery and so the responsibility for the loss belongs the last person in the chain who can verify receipt, which is the delivery company.
A big problem for companies like UPS and FedEx is that they offer a service that's convenient for senders, not receivers. When I order a delivery from one of my local supermarkets, I specify a one or two hour timeslot when my delivery will take place. If I can't take time away from work, then I can schedule it in a weekend or evening. On the other hand, if I order something from Amazon then they will send it to me by a service that gives me a two day window. This doesn't actually end up being cheaper for the courier, because I'm often not in the first time they try to deliver things, so they have to come back the next day. If they could tell me in advance the one hour window when a package will arrive, that would be significantly more convenient. If they could let me choose a time, that would be even better.
While true, so is an abacus. Printing even a simple integrated circuit is a long way beyond the capabilities of a state of the art 3D printer. The original 8086 was produced on a 3um process. A 3D printer with that kind of accuracy simply doesn't exist, and that's assuming you could use the same printing techniques for ICs (the chemicals involved are quite corrosive and so you'd need some quite specialised equipment). One of the reasons Intel has maintained such a dominant position is that IC production is one of the fields where economies of scale really, really matter.
Insightful? No. I have an FPGA dev board on my desk. The dev board costs around $8K, the FPGA alone can be bought in small quantities for about $4K. We use it for experimental processor design. It can run our MIPS64-based softcore at about 100MHz (drawing around 40W) and there's enough space on die for 4-8 cores. You can't run a processor on one that is competitive with a cheap ARM processor (except if you configure the FPGA for a single algorithm, then you can't run general-purpose code on it), let alone one with 'all the power and capabilities we want'. FPGAs are cool, but they're no substitute for ASICs.
Most projects are bad ideas or don't appeal widely enough to be worth funding. The point of Kickstarter is to cut out the middle man between people who want a product or service and people willing to provide it. It isn't meant to fund things that there is no market for, it is meant to directly connect the funding for things with the existence of a market.
I think you're confusing it with a derived tool.
This is a platform that was not designed to have a scheduler carving up resources
Uh, what? The ARM architecture was designed for Acorn's line of 32-bit desktop computers, which shipped with a multitasking OS from the start. Now, it wasn't preemptively multitasking, but the only difference between cooperative multitasking and preemptive from a hardware perspective is that you need (relatively) cheap timer interrupts to enable preemptive multitasking, and ARM has always had this.
The cost of context switching boils down to a small number of things:
In summary, the orifice that you are talking out of is not your mouth.
And it's up to well-written apps to provide hints. OS X / iOS has the NSCache interface, which allows you to store references to objects that may be destroyed in response to memory pressure (unless they are in use right now). This is important, because it allows an application to grow to use RAM when it's available, without dying horribly when it isn't.
I did some work on one of Mark's other books, so he still periodically sends me copies of any that contain snippets that I wrote. I've got the second edition lying around somewhere (I think it contains a total of about three pages of text I wrote, no idea if there's anything of mine in the third edition). Most of the stuff is pretty timeless, but as I recall this book also talks about various configuration things. One of the reasons I switched from Linux to FreeBSD is that these have a habit of changing every six months in the Linux world without actually improving.
Interactions are only a problem when the two chemicals mix. Aviation gas reacts quite easily with oxygen, but it's safe because it's stored in tanks and kept isolated from the air. Hydrogen requires a flame to ignite, just like most fuels, and because it's much lighter than air it's very safe in a balloon that isn't in an enclosed space because any leakage will escape upwards quickly (and diffuse even more quickly).
And the reason why Michele wants to take Dell private is so he can do some radical things to it. So who knows? Give the man the benefit of a doubt, grab some popcorn, and see what happens
He's already CEO and Chairman. If he can take the company in new profitable directions, then there's no incentive for the other shareholders to sell because they they'd be giving up their shares just before they become more valuable. Their only sensible choice is to get as much money as possible for the shares now, since, if it goes private, they'll never be able to own part of Dell again.
The rigidity isn't the problem, the leakage is. A normal hydrogen balloon equal pressure inside and out, but the gaps in the membrane are large enough that hydrogen molecules can pass. I suppose that making something airtight is easier than making something vacuum-tight, so if the aerogel were rigid enough to support one atmosphere of pressure on all sides, then it may give cheap dirigibles...
Not really. The problem is in the sealing. If you had a membrane that could withstand the pressure of air on one side, vacuum on the other, and not allow the air to seep in, without adding more than a negligible amount of weight, then you could just use the same material to make hydrogen balloons.
I don't think I've ever seen a bicycle helmet that will protect your face. What sort do you wear?
Actually, in 2003 Opera was not that obscure. The wikipedia statistics have 94.43% for IE. Netscape and Mozilla each had about 2.5% and Opera had 0.66%. Back then, on any *NIX system, Netscape was old, Mozilla was crashy (and slow), Opera was fast. On Macs, the same was roughly true, except you had IE 5.5, which had transparent PNG support (which Windows lacked), but implemented an old draft of CSS standard. On Windows you also had IE, which was ubiquitous, but Opera was the best choice technically. They also released binaries for FreeBSD and NetBSD, on a few architectures. It was sufficient of an improvement over the competition that I actually paid for it.
Why do you keep bringing up weight, when I'm talking about diameter? A typical bike helmet is not very dense at all, it is usually a plastic cover over foam. If you want to dig out the statistics, they were a Slashdot story a couple of years ago, condensed from data from the USA, but I don't remember which government department issued them.
In a traditional filesystem, yes. The mirroring happens at the block device level, and so it is completely unaware of the semantics of the filesystem and will duplicate anything, potentially overwriting good data with bad if the filesystem is corrupted. Worse, unless the drive fails catastrophically, you're liable to either duplicate single-block errors or to be unable to tell which copy of a block is the damaged one. ZFS fixes the second of these problems with block-level checksums, so it can tell which disk has errors. It also makes the mirroring infrastructure partially aware of the filesystem layout, so it shouldn't duplicate filesystem corruption, however it will happily copy user errors. For example, if your word processor corrupts a document as it saves it, then there's nothing ZFS can do about that (unless you have an earlier snapshot). And, of course, if there's a bug in the filesystem driver, all bets are off.
Mirroring, as the grandparent says, is not a substitute for proper backups. One of the most common reasons for restoring from backups is accidental deletion. Even a filesystem with 100% reliability won't protect you against this.
Oh, I'm sorry, I thought we were talking about safety. I didn't realise this was a fashion discussion. The statistics I'm quoting were published about two years ago, so the century in question would be the 21st.