Let's say a building has a lot of dsl and or cable lines now for some thing like this to work may need to have fiber put in or a big mess of multi IP routers hooded to banks of DSL / cable lines.
The incumbent telco is supporting this project with fiber pulls. The 'IT guy' is part ISP salesman on commission.
We've talked about this - my daughter would rather come to my office after school to do her homework than stay at home, near where the bus drops her off. She keeps asking if we can by a car with autopilot yet. I suspect she'll have her license before the legislature here allows them.
which don't deal with mass amounts of data that exceeds the memory limits by orders of magnitude (which is roughly the point it becomes a mild pain)
I was talking to an HPC friend this weekend at the ice cream parlor and he was telling me how their problem had no advantage on GPU processing because they were really memory-bound, not processing-bound.
He has a quad-rate Infiniband going into each machine (40Gbps) and a couple CPU's, and keeps them saturated (say 5Gbps per core).
Looking at TFA's expansion card, with a memory bandwidth of 320Gbps and 60 cores, that's only 4Gbps per core and what's worse, you can't push that much over the PCIe 3.0 bus (only 16GBps).
What they really need is a card with 8 cores and an Infiniband controller right on the die and DMA from one to the other. Then you could fill a housekeeping box chock full of slots with these things and only worry about pushing setup code and managing jobs over the PCIe bus from the mainboard. There's a market niche that needs filling, hardware dudes.
I was expecting 32 cores minimum in desktop CPUs by the start of this decade. All this new supercomputer stuff is well and good, but what about lots of cores for us mere mortals too?
You wouldn't like the speed of typical software on a 32-core CPU using the same transistor count (i.e. at the same cost) of the machine you're running now.
Cache sharing, NUMA access, etc. turn out to be tricky to get fast, right, and cheap. In the meantime, much of the existing software library can't even properly take advantage of a 6-core desktop chip, so all mere mortals would get today from a 32-core chip would be a slower machine.
The statement that a single analyst can eavesdrop on domestic communications without proper legal authorization is incorrect and was not briefed to Congress.
Of course not, the NSA believes it is operating under "proper legal authorization". Or at least it's deluded itself into thinking that.
The DNI does not say, "The statement that a single analyst can eavesdrop on domestic communications is incorrect and was not briefed to Congress."
Certainly. GPG offers authentication, like TLS, and of course security through encryption, but neither offer repudiation or perfect forward secrecy, which are essential features of OTR and ought to be for any protocol implementing causal chat protocols. They'd be inappropriate features for the problems GPG and TLS are solving.
In general, you don't want to be able to prove that Alice or Bob said something in an online chat at some point in the future - that's not the nature of chatting. And if Mallory captures their computing devices, you don't want him to be able to forge messages using their private keys that make it look like they said something they didn't.
What their protocol have that XMPP doesn't, or couldn't be extended to support?
It has TLS, which is a bad idea for chat. Unless you're taking a deposition, or something, where you want provable identity, most chat is expected to be ephemeral and reputable. Picture two people sitting quietly chatting in a secure room. That's the goal for most online chat.
You want to use OTR for most chat, not TLS. It offers repudiation as well as authentication, security, and perfect forward secrecy. It's even obnoxious about repudiation on the wire.
TLS is great for other stuff. It's just the wrong choice for online chat.
You're not smarter than the presenters in those videos, nor, unless you're a hot-shot constitutional attorney, are you smarter than a police apparatus that's trying to frame you. Here, in geek terms:
Any idiot can write ignorance like spinning the Earth backwards to save a girl.
Uh, he didn't spin the earth backwards. He exceeded the speed of light to go back in time - the mythical reference frame of the camera on Superman saw the earth "spin backwards" as the direction of time reversed.
I'm all for archaeology and historical preservation, but it's absurd to stick the land owner with the cost.
It could be handled that way, but with responsibility comes authority in a sane property rights regime. That is, the owners should be able to sell the archaeological finds to the highest bidder to recoup the dig costs or dispose of them if the costs cannot be recouped (that is, nobody finds the dig to be of value).
That would be the way to maximize the recovery of artifacts and have them make their way to museums. Sure, in the short term private collectors might have them, but that's not a lasting problem, especially compared to the age of most interesting artifacts.
It sounds like in Canada the owners have the responsibility but not the authority. That's just a way to socialize costs in an acute fashion and to reinforce the idea of weak property rights.
You need to work on changing the minds of the public, then maybe you'll see changes in the government.
The majority of the people will always go along with whatever sounds good at the moment, so those polls are hard to garner information from. Fun fact: in 1775, only 18% of the people favored separation from the King. Something like 15% favored staying British (the 'loyalists') and the rest just didn't care too much.
Coincidentally, a recent poll showed that 18% of Americans felt that the people would be better off if the US was broken up into smaller nations.
$10? That will outright pay for a 40W Cree equivalent with a ten year warranty.
Sadly, the rebates are on units $15 and higher. This was fine before the Crees came out, but now it's worth paying the $13 for the Crees in some cases. I just wish the HD would charge me $15 for a 60W Cree.
It's OK, I like the $30 LED can lights - I bought six of them to re-do the kitchen lighting. I tip-toed into CFL early on, but I can't wait to re-do the whole house in LED now. As far as I can tell, the Cree bulbs are perfect.
In the mean time, let the first adopters with lots of disposable income (and a lot less sense) buy them.
My silly power company is giving me $10 back on each of the LED's that I buy. They must not have enough generating power, because I'm getting 60W equivalent (2700K) LED bulbs for about the same cost as operating a 60W incandescent for a year. The ones that look just like an incandescent have a high WAF.
So, if the LED lasts a year, I've broken even. If it lasts more than that, I'm saving money. I think you have to have disposable income to not switch to LED, if you can get in on the rebates and have a minimum amount of capital to hold for a year.
In other cases (for example, if a shark is chewing on your leg), watching a visual representation of your brain isn't going to help much.
I tried one of those biofeedback devices as a kid - my father is a dentist and his patients back then were asking for it (fad?) as an alternative to other pain control methods, so he got one. It was a handheld unit basically the shape of a modern mouse (hrm, the Mac 128 could have learned something from those) that had a couple electrically conductive pads on them connected to a circuit with a speaker. To relax, the goal was to get the sound of the speaker to go down in pitch and tremolo. For the first few minutes it was pretty erratic, but then you could learn to get control of it. If you got yourself anxious, it would spike back up again (maybe it was based on perspiration?), so I think basically it was an anxiety measurer - effectively giving people a tool to measure their relaxation level. Presumably a relaxed person does not mind the pain as much.
Anyway, some of the patients used it for minor procedures (fillings and such). I don't think anybody used it for extractions or root canals. I was like, "hell, no, nitrous and lidocaine, please," but it was a neat toy anyway.
right, it's a self-clocking digital signal. But it's quinary, not binary, as the symbol space, the letter space, and the word space are all significant (as well as dot and dash, of course).
The other thing you can to is set the clock too high
From what I've read, overclocking is a sure way to find a corrupt SD, so you might be onto something there. Some have said that the UK units are more reliable when they have both UK and Chinese units side-by-side. I don't know if that's true.
and then run it 20% or so slower to get a safety margin
and if you forgot to factor in the 20%? Just a hunch.
After all, digital cameras, smartphones, MP3 players, etc. all get it right.
Exactly, which makes this so frustrating. Call me naive, but this seems like a solved problem.
Now, what definitely helps in both cases is to have a card that is faster than the bus.... I am using some 30MB/s cards, which should give plenty of safety margin as the Pi seems to level out at about 12MB/s.
Hrm, now from what I've seen it's people using the Class 4 cards who are having better luck than those on Class 6 or 10 cards. But, and my memory from a few years ago is rusty, but at the time I was doing some embedded linux development (just COTS parts) and I wound up going with Class 4 cards instead of 6 or 10 at the time because their actual transfer capability was often higher, at least among certain manufacturers. I forget why, but Class 10 was a "burst out the gates" kind of fast, that faltered shortly. Again, this is years-old r&d, it might have changed by now. I'm running some of those old cards in my Pis and so far so good, though.
Do you have a link to the reference that says it is a timing issue?
I tried looking just now with 'raspberry pi sd card "timing"' and found a bunch of references to it, but, sorry, not the analysis I was thinking of. Just for thread completeness, here's the lkml post talking about the USB problems.
So far I have not had issues with the SD slot, fortunately. Maybe people are using it wrong?
nor have I, but from what I read, some really competent people do, so I'm figuring it's a matter of time. I saw reports of "fine for two months and then corruption". I'm looking to try PXE booting them next, to minimize writes.
Apparently the SD card write timings are out of spec, and some SD cards are tolerant of this and some aren't. I suppose with a tolerant enough card one might never encounter a problem.
I've done it with one of those world's-cheapest four-port USB2 hubs that has the molded foot-long pigtail and the really square transparent case.
And I'm doing it now with this hub which was recommended for RPi noobs when I bought mine.
It does pain me to run a hub that's nearly as expensive as the computer, but I was going for the zero-frustration approach. "World's cheapest" does appeal to the low-cost solutions driver in me, though - got a link?
Slashdot Mirror
Is the landlord going to run the server farm?
Certainly not - the landlord will just rent space to the 'IT guy'.
Let's say a building has a lot of dsl and or cable lines now for some thing like this to work may need to have fiber put in or a big mess of multi IP routers hooded to banks of DSL / cable lines.
The incumbent telco is supporting this project with fiber pulls. The 'IT guy' is part ISP salesman on commission.
Sending the car to pick up the kids after school
We've talked about this - my daughter would rather come to my office after school to do her homework than stay at home, near where the bus drops her off. She keeps asking if we can by a car with autopilot yet. I suspect she'll have her license before the legislature here allows them.
which don't deal with mass amounts of data that exceeds the memory limits by orders of magnitude (which is roughly the point it becomes a mild pain)
I was talking to an HPC friend this weekend at the ice cream parlor and he was telling me how their problem had no advantage on GPU processing because they were really memory-bound, not processing-bound.
He has a quad-rate Infiniband going into each machine (40Gbps) and a couple CPU's, and keeps them saturated (say 5Gbps per core).
Looking at TFA's expansion card, with a memory bandwidth of 320Gbps and 60 cores, that's only 4Gbps per core and what's worse, you can't push that much over the PCIe 3.0 bus (only 16GBps).
What they really need is a card with 8 cores and an Infiniband controller right on the die and DMA from one to the other. Then you could fill a housekeeping box chock full of slots with these things and only worry about pushing setup code and managing jobs over the PCIe bus from the mainboard. There's a market niche that needs filling, hardware dudes.
I was expecting 32 cores minimum in desktop CPUs by the start of this decade. All this new supercomputer stuff is well and good, but what about lots of cores for us mere mortals too?
You wouldn't like the speed of typical software on a 32-core CPU using the same transistor count (i.e. at the same cost) of the machine you're running now.
Cache sharing, NUMA access, etc. turn out to be tricky to get fast, right, and cheap. In the meantime, much of the existing software library can't even properly take advantage of a 6-core desktop chip, so all mere mortals would get today from a 32-core chip would be a slower machine.
Of course not, the NSA believes it is operating under "proper legal authorization". Or at least it's deluded itself into thinking that.
The DNI does not say, "The statement that a single analyst can eavesdrop on domestic communications is incorrect and was not briefed to Congress."
See?
You can deal with Authentication using GPG.
Certainly. GPG offers authentication, like TLS, and of course security through encryption, but neither offer repudiation or perfect forward secrecy, which are essential features of OTR and ought to be for any protocol implementing causal chat protocols. They'd be inappropriate features for the problems GPG and TLS are solving.
In general, you don't want to be able to prove that Alice or Bob said something in an online chat at some point in the future - that's not the nature of chatting. And if Mallory captures their computing devices, you don't want him to be able to forge messages using their private keys that make it look like they said something they didn't.
What their protocol have that XMPP doesn't, or couldn't be extended to support?
It has TLS, which is a bad idea for chat. Unless you're taking a deposition, or something, where you want provable identity, most chat is expected to be ephemeral and reputable. Picture two people sitting quietly chatting in a secure room. That's the goal for most online chat.
You want to use OTR for most chat, not TLS. It offers repudiation as well as authentication, security, and perfect forward secrecy. It's even obnoxious about repudiation on the wire.
TLS is great for other stuff. It's just the wrong choice for online chat.
disclaimer: trusting the summary.
You're not smarter than the presenters in those videos, nor, unless you're a hot-shot constitutional attorney, are you smarter than a police apparatus that's trying to frame you. Here, in geek terms:
Any idiot can write ignorance like spinning the Earth backwards to save a girl.
Uh, he didn't spin the earth backwards. He exceeded the speed of light to go back in time - the mythical reference frame of the camera on Superman saw the earth "spin backwards" as the direction of time reversed.
OTOH a market approach to dinos doesn't make much sense because they ain't making any more of them.
They ain't making any more gold either, at least not in these parts for another seven billion years.
No, it would simply lead to maximizing disposal.
That's silly, nobody throws away things that have value. If they found a gold nugget instead of an artifact, would they throw it away or sell it?
I'm all for archaeology and historical preservation, but it's absurd to stick the land owner with the cost.
It could be handled that way, but with responsibility comes authority in a sane property rights regime. That is, the owners should be able to sell the archaeological finds to the highest bidder to recoup the dig costs or dispose of them if the costs cannot be recouped (that is, nobody finds the dig to be of value).
That would be the way to maximize the recovery of artifacts and have them make their way to museums. Sure, in the short term private collectors might have them, but that's not a lasting problem, especially compared to the age of most interesting artifacts.
It sounds like in Canada the owners have the responsibility but not the authority. That's just a way to socialize costs in an acute fashion and to reinforce the idea of weak property rights.
If their telco's lowered the price of calls, then perhaps folks would not mind if Internet calling was less available.
The number of people who think this story is really about Saudi Arabia needing the telco income is asymptotic to zero.
You need to work on changing the minds of the public, then maybe you'll see changes in the government.
The majority of the people will always go along with whatever sounds good at the moment, so those polls are hard to garner information from. Fun fact: in 1775, only 18% of the people favored separation from the King. Something like 15% favored staying British (the 'loyalists') and the rest just didn't care too much.
Coincidentally, a recent poll showed that 18% of Americans felt that the people would be better off if the US was broken up into smaller nations.
$10? That will outright pay for a 40W Cree equivalent with a ten year warranty.
Sadly, the rebates are on units $15 and higher. This was fine before the Crees came out, but now it's worth paying the $13 for the Crees in some cases. I just wish the HD would charge me $15 for a 60W Cree.
It's OK, I like the $30 LED can lights - I bought six of them to re-do the kitchen lighting. I tip-toed into CFL early on, but I can't wait to re-do the whole house in LED now. As far as I can tell, the Cree bulbs are perfect.
In the mean time, let the first adopters with lots of disposable income (and a lot less sense) buy them.
My silly power company is giving me $10 back on each of the LED's that I buy. They must not have enough generating power, because I'm getting 60W equivalent (2700K) LED bulbs for about the same cost as operating a 60W incandescent for a year. The ones that look just like an incandescent have a high WAF.
So, if the LED lasts a year, I've broken even. If it lasts more than that, I'm saving money. I think you have to have disposable income to not switch to LED, if you can get in on the rebates and have a minimum amount of capital to hold for a year.
In other cases (for example, if a shark is chewing on your leg), watching a visual representation of your brain isn't going to help much.
I tried one of those biofeedback devices as a kid - my father is a dentist and his patients back then were asking for it (fad?) as an alternative to other pain control methods, so he got one. It was a handheld unit basically the shape of a modern mouse (hrm, the Mac 128 could have learned something from those) that had a couple electrically conductive pads on them connected to a circuit with a speaker. To relax, the goal was to get the sound of the speaker to go down in pitch and tremolo. For the first few minutes it was pretty erratic, but then you could learn to get control of it. If you got yourself anxious, it would spike back up again (maybe it was based on perspiration?), so I think basically it was an anxiety measurer - effectively giving people a tool to measure their relaxation level. Presumably a relaxed person does not mind the pain as much.
Anyway, some of the patients used it for minor procedures (fillings and such). I don't think anybody used it for extractions or root canals. I was like, "hell, no, nitrous and lidocaine, please," but it was a neat toy anyway.
This does not make it analog.
right, it's a self-clocking digital signal. But it's quinary, not binary, as the symbol space, the letter space, and the word space are all significant (as well as dot and dash, of course).
The other thing you can to is set the clock too high
From what I've read, overclocking is a sure way to find a corrupt SD, so you might be onto something there. Some have said that the UK units are more reliable when they have both UK and Chinese units side-by-side. I don't know if that's true.
and then run it 20% or so slower to get a safety margin
and if you forgot to factor in the 20%? Just a hunch.
After all, digital cameras, smartphones, MP3 players, etc. all get it right.
Exactly, which makes this so frustrating. Call me naive, but this seems like a solved problem.
Now, what definitely helps in both cases is to have a card that is faster than the bus. ... I am using some 30MB/s cards, which should give plenty of safety margin as the Pi seems to level out at about 12MB/s.
Hrm, now from what I've seen it's people using the Class 4 cards who are having better luck than those on Class 6 or 10 cards. But, and my memory from a few years ago is rusty, but at the time I was doing some embedded linux development (just COTS parts) and I wound up going with Class 4 cards instead of 6 or 10 at the time because their actual transfer capability was often higher, at least among certain manufacturers. I forget why, but Class 10 was a "burst out the gates" kind of fast, that faltered shortly. Again, this is years-old r&d, it might have changed by now. I'm running some of those old cards in my Pis and so far so good, though.
Do you have a link to the reference that says it is a timing issue?
I tried looking just now with 'raspberry pi sd card "timing"' and found a bunch of references to it, but, sorry, not the analysis I was thinking of. Just for thread completeness, here's the lkml post talking about the USB problems.
thanks, I'll check my preferred provider of cheap Chinese electronics.
So far I have not had issues with the SD slot, fortunately. Maybe people are using it wrong?
nor have I, but from what I read, some really competent people do, so I'm figuring it's a matter of time. I saw reports of "fine for two months and then corruption". I'm looking to try PXE booting them next, to minimize writes.
Apparently the SD card write timings are out of spec, and some SD cards are tolerant of this and some aren't. I suppose with a tolerant enough card one might never encounter a problem.
Don't plug up tor with streaming video - use a VPN provider for that. You need an endpoint IP, not anonymity.
I've done it with one of those world's-cheapest four-port USB2 hubs that has the molded foot-long pigtail and the really square transparent case.
And I'm doing it now with this hub which was recommended for RPi noobs when I bought mine.
It does pain me to run a hub that's nearly as expensive as the computer, but I was going for the zero-frustration approach. "World's cheapest" does appeal to the low-cost solutions driver in me, though - got a link?
I expect this will get fixed eventually
Like the GP said, USB on the RPi is defective. There's an infamous lkml post outlining the problems.
The SD slot is defective too - a tremendous number of stories are out there about SD corruption. Too bad it's the only possible boot device.
oh, and the h.264 decoder is defective too. Blocking artifacts, GPU lockups.
People tell me the Beaglebone Black is what the RPi was supposed to be.