Google say "a drive is considered to have failed if it was replaced as part of a repairs procedure". This along with some other paragraphs in the reports strongly implies that they consider a drive to be failed as soon as it drops out of a raid set* regardless of what proportion of the data on that drive is recoverable. Their report doesn't seem to distinguish between drives that are problematic but mostly readable and drives that are dead.
My experience echos that of the GP in that most of the time drives i've seen go bad become excruciatingly slow and/or report read/write errors and/or suddenly lock up while remaining mostly readable. Even the few drives that appear completely dead can often be recovered with a controller board swap.
I don't have any personal experience with SSD failures to comment there.
* I am making the assumption here that google's servers use raid on the basis that most decent servers do and that the google server pics I can find show more than one drive. Seaching for google raid doesn't seem to find anything relavent either way just crap about police raids.
did FOSS really save IBM? I was under the impression that what saved IBM was reinventing themselves as a "soloutions" company. They may use a bit of FOSS in their solutions and contribute a bit to the projects they use but I never got the impression it was critical to them.
As for HP the real question IMO is would they be able to monetize their contributions. Contributing to FOSS buys you experience with the software (useful for consultancy) and mindshare with the projects leaders (useful if you are trying to build a solution on top of the software and want it steering in your direction) but ultimately to make money you need a product or service to sell, preferably one that is hard to duplicate.
My experiance with ikea furniture is that they like to use cardboard. Specifically instead of using the normal melamine coated chipboard. they use a construction that has a melamine surface, then a thin hardboard layer. Then a cardboard honeycomb core. It's almost certainly lighter and cheaper than chipboard but it also poses a major headache if you want to modify the furniture in any way.
Since you are using british terminology i'm assuming you are reffering to the UK, I can't seem to find anything in TFA about where the care is located.
Avoid any nasty surprises.
While I'd agree your list of things to check for is good for those buying cars in general I don't think they are the most important things in this case.
I doubt anyone cares too much about the mileage on a car like this. You don't buy a car like this because of the mileage or lack thereof. You buy it to own a price of history. Did cars of that age even have an odometer?
I'm not sure why anyone would file a SORN for such an old car when they could just get a tax disc. It's not like you have to pay road tax.
It's being sold by a major auctioneer so one would expect them to have done the legal checks to see if it was stolen etc.
I'd think the main things to check would be the mechanical condition of the car. Whether it has a MOT and if not how much it would cost to restore it to the point it gets through one and that sort of thing.
As others have said at least in the case of DSL the latency doesn't come from the copper line itself. It comes from the signal processing used to squeeze relatively huge amounts of data down a line designed for analog voice.
In particular when using forward error correction techniques (which are pretty much essential when you are running close to the limit of a channels capacity) it often makes sense to interleave frames. That way if there is a burst of noise it's impact will be spread across multiple frames and will be correctable whereas in a single frame the same noise burst would have been an uncorrectable error. However interleaving frames necessarily adds latency.
It's possible to turn off interleaving on ADSL but afaict most providers are reluctant to do so.
The biggest problem of copper is not latency, it's that you have to lay the fucking cable.
The biggest problem with copper is the low bandwidth and high interference susceptibility of long copper runs, especially long runs of cheap phone pair. Cable TV coax offers more bandwidth but that bandwidth is shared and much of the cable's bandwidth is taken up by cable TV services.
My guess would be that they did see it beforehand but didn't appreciate how much smaller the rooms were because it was either empty or dressed to look bigger than it really is.
There is a micro connector for USB 3 but it's much wider (it's formed of two sections, one with the low/full/high speed connections compatible with previous USB versions and a seperate portion for the superspeed connections) and phones don't really have any need for USB3 speeds at the moment.
Afaict there is already a soloution to that, boots that fit on the rear of the connector and provide both a strain relief and an anti-snag cover for the rear of the locking tab. On the downside they often make it harder to unplug the cables as you must compress the plastic of the boot when depressing the tab on unlock the connector.
It is difficult to tell how durable an item is. By the time you know how durable something is in the real world it is likely to be discontinued or silently changed. Some people base their decisions on brand trust. Unfortunately if there are financial problems or a need to look good to the stock market it's very tempting for a manufacturer to sell out their brand for a quick buck.
My experience is it's very rare that the connector itself gets busted. A far bigger problem is solder joints and/or PCB tracks getting busted as the connector moves relative to the PCB.
Re:For the uninformed like me....
on
Arduino Goes ARM
·
· Score: 1
I can't see why IO would become a problem
Compatibility between 5V logic and 3.3V logic is a somewhat complex area. Some 5V devices can accept 3.3V signals and some 3.3V devices are tolerant of 5V signals but many aren't (on both sides). So much of the time you end up having to building level translation circuits to interface 5V stuff to 3.3V stuff.
A further complication is that most IO on microcontrollers has independent direction control for every pin. So if you want to make a fully general level translation circuit you have to 1: find a bidirectional buffer chip with individual direction lines (or build the equivilent of one out of tristate buffers) and 2: find enough IO to control the direction of all those buffers. It's doable but I'm not sure if it's doable at a cost (in both size and money) that would be acceptable to the ardunio guys. I certainly don't see any evidence of it on the photo in TFA
Some discussion implementations (slashdot's D2 being one of them) make it easy to accidently lose a post by navigating away from the page or sometimes even by actions on the page. Thanks to the heavy use of javascript the auto-refilling of the text boxes that some browsers do when you use the back button doesn't work with D2 either.
Yes one could use notepad but that doesn't have a spell checker.
And not everyone uses a browser recent enough to have a spell checker.
Afaict it's not the fiber itself that is expensive It's the laser diodes, photodiodes, precision connections to the fiber, protection against fractured fibers from kinked cables and so on that make a high speed fiber system expensive. Especially if it has to be made "idiot-proof".
TOSLINK (optical version of S/PDIF) is indeed cheap but that is because the low speeds let them get away with REALLY low grade optical components (including polymer fiber that is less prone to damage than glass fiber.
With the rate power consumption of CPU cores has been increasing
You are a few years behind the curve. Since the pentium D CPU TDP (roughly power consumption under the highest normal load) has stayed pretty much flat while core counts and performance of individual cores have gone up (despite the drop in clockspeeds)
The power consumption per core has been going down in recent times. The pentium D 965 extreme has two cores, a clockspeed of 3.73 GHz and a TDP of 130W. The i7-990x has six cores.a nominal clockspeed of 3.47 GHz (plus turbo boost) and a TDP of 130W.
Comparing the two processors (made about 5 years apart) the later processor has the same TDP, three times as many cores and slightly lower clockspeed (but each core has far better performance at a given clockspeed)
The problem with that approach is that if foocorp doesn't have any relavent patents of their own and barcorp comes along and claims that foocorp violates their patents then barcorp has foocorp over a barrel. They may choose to grant a license at a reasonable rate, they may not. If they don't then afaict there is nothing foocorp can do about it.
OTOH if foocorp and barcorp both have relavent patent portfolios then a balance comes into play. Since neither side wants to be destroyed they have to agree not to destroy each other.
Quite likely neither is relavent. At least here individual research groups aren't charged for electricity and labour is pretty much a fixed cost. As for aircon that isn't really an issue at this scale.
As was seen this year in Fukushima. Even if you didn't get cancer or be made infertile (afaik people not working in the station did not suffer these risks) you were probably not very happy about what happened if you lived in the evacuation zone.
Nuclear isn't the only industry whose accidents can displace a community on a near permanent basis.
And that is just direct impacts. If global warming from fossil fuel usage causes large sea level rises or shifts in what land is arable that could displace a lot of people to.
Opensource and open standards are different things
But closely intertwined. In particular "open" standards can be divided into two main categories.
Open and unencumbered standards where anyone can implement the standard without having to sign license agreements or pay royalties. Open but encumbered standards where anyone can buy and read a copy but to implement the standard legally you have to sign a patent license agreement and pay license fees.
Of couse there is the further complication that thanks to how patents work it is impossible to be sure a standard is completely free of patent encumberenaces. You can only say that there are no known encumberances. Even being older than the typical lifetime of a patent isn't surefire protection in some countries.
as long as I can still look at my archives in 10-20-50 years.
And the best way to maximise the chance of that is to have the widest support possible for said standard. FOSS projects generally can't afford to pay patent license fees and even if they could the patent licenses are generally incompatible with major FOSS licenses. Sometimes FOSS projects implement encumbered standards anyway (either illegally or through basing the project in a country where the encumberance doesn't apply) but a few legal threats could easilly cause the project to pull the feature (yes you can go on using the old version for a while but as the underlying platforms move on that is likely to get harder and harder).
1: do you replace the entire infrasturcture with fiber or do you use existing (coax or twisted pair) infrastrucutre for part of the system? 2: if you do replace the whole infrastucture with fiber do you give each customer their own fiber right back to the exchange? do you have fiber switches out in the field? do you build a passive fiber network with TDMA or CSMA? do you use WDM to give multiple customers independent channels in the same fiber?
The answers chosen to these questions are likely to have a big impact on both the cost of the deployment and the speeds deliverable. In particular a deployment where you only replace the lines to the street is likely to be far cheaper than anything else but also give far lower speeds.
The reason this post is stupid is that infrastructure is long term. When you go to the trouble of sending out a crew to dig up and put fiber in the ground your putting in an infrastructure asset that should have a 15-30 year lifespan. The fact that can average machine can't saturate it today means we're being forward thinking.
True to an extent but remember there is nothing stopping the provider from changing how the fiber is used without changing the fiber itself.
Another alternative is to check the object through a method after construction, which a lot of STL objects do, but that's kind of messy.
Kind of messy but given that the whole point of constructors is that they are called automatically so the object is never in an undefined state you are going to have this fundamental problem.
a third alternative is to have the constructor create the object in an empty form (which can generally be done without memory allocation) and then fill it with data through regular functions.
. If you say that your car goes x mph and does y mpg I have to go to google or wolfram alpha to convert those numbers to km/h and L/100km.
Adding to the fun you have to be careful with mpg conversions because brits use miles per imperial gallon while americans use miles per US gallon but both use the term mpg.
IMO it should pay dividends the same as common stock and should convert to common stock on the death of the holder.
And it shouldn't be used for 100% of compensation, just a sufficiant proportion of it to give the employee a reason to care about the long term future of the company.
64GB system memory was $1500 a few months ago. I think it is below $1000 now.
The ram itself seems to be $880 (four kits each with 2x8GB registered ECC modules from newegg) but you also have to consider that to actually put that much ram in one system rules out use of a normal desktop platform because desktop processors (when used with currently available memory modules) simply can't support that much memory. Nor can many low end server platforms.
That means that you will likely have to spend a lot more money on your CPU and motherboard to get a platform that supports your 64GB of memory (the cheapest option appears to be G34 but it's still about $500 and the 8 slow cores you get may not be a good fit for your workload)
Slashdot Mirror
Google say "a drive is considered to have failed if it was replaced as part of a repairs procedure". This along with some other paragraphs in the reports strongly implies that they consider a drive to be failed as soon as it drops out of a raid set* regardless of what proportion of the data on that drive is recoverable. Their report doesn't seem to distinguish between drives that are problematic but mostly readable and drives that are dead.
My experience echos that of the GP in that most of the time drives i've seen go bad become excruciatingly slow and/or report read/write errors and/or suddenly lock up while remaining mostly readable. Even the few drives that appear completely dead can often be recovered with a controller board swap.
I don't have any personal experience with SSD failures to comment there.
* I am making the assumption here that google's servers use raid on the basis that most decent servers do and that the google server pics I can find show more than one drive. Seaching for google raid doesn't seem to find anything relavent either way just crap about police raids.
did FOSS really save IBM? I was under the impression that what saved IBM was reinventing themselves as a "soloutions" company. They may use a bit of FOSS in their solutions and contribute a bit to the projects they use but I never got the impression it was critical to them.
As for HP the real question IMO is would they be able to monetize their contributions. Contributing to FOSS buys you experience with the software (useful for consultancy) and mindshare with the projects leaders (useful if you are trying to build a solution on top of the software and want it steering in your direction) but ultimately to make money you need a product or service to sell, preferably one that is hard to duplicate.
My experiance with ikea furniture is that they like to use cardboard. Specifically instead of using the normal melamine coated chipboard. they use a construction that has a melamine surface, then a thin hardboard layer. Then a cardboard honeycomb core. It's almost certainly lighter and cheaper than chipboard but it also poses a major headache if you want to modify the furniture in any way.
Since you are using british terminology i'm assuming you are reffering to the UK, I can't seem to find anything in TFA about where the care is located.
Avoid any nasty surprises.
While I'd agree your list of things to check for is good for those buying cars in general I don't think they are the most important things in this case.
I doubt anyone cares too much about the mileage on a car like this. You don't buy a car like this because of the mileage or lack thereof. You buy it to own a price of history. Did cars of that age even have an odometer?
I'm not sure why anyone would file a SORN for such an old car when they could just get a tax disc. It's not like you have to pay road tax.
It's being sold by a major auctioneer so one would expect them to have done the legal checks to see if it was stolen etc.
I'd think the main things to check would be the mechanical condition of the car. Whether it has a MOT and if not how much it would cost to restore it to the point it gets through one and that sort of thing.
As others have said at least in the case of DSL the latency doesn't come from the copper line itself. It comes from the signal processing used to squeeze relatively huge amounts of data down a line designed for analog voice.
In particular when using forward error correction techniques (which are pretty much essential when you are running close to the limit of a channels capacity) it often makes sense to interleave frames. That way if there is a burst of noise it's impact will be spread across multiple frames and will be correctable whereas in a single frame the same noise burst would have been an uncorrectable error. However interleaving frames necessarily adds latency.
It's possible to turn off interleaving on ADSL but afaict most providers are reluctant to do so.
The biggest problem of copper is not latency, it's that you have to lay the fucking cable.
The biggest problem with copper is the low bandwidth and high interference susceptibility of long copper runs, especially long runs of cheap phone pair. Cable TV coax offers more bandwidth but that bandwidth is shared and much of the cable's bandwidth is taken up by cable TV services.
My guess would be that they did see it beforehand but didn't appreciate how much smaller the rooms were because it was either empty or dressed to look bigger than it really is.
There is a micro connector for USB 3 but it's much wider (it's formed of two sections, one with the low/full/high speed connections compatible with previous USB versions and a seperate portion for the superspeed connections) and phones don't really have any need for USB3 speeds at the moment.
Afaict there is already a soloution to that, boots that fit on the rear of the connector and provide both a strain relief and an anti-snag cover for the rear of the locking tab. On the downside they often make it harder to unplug the cables as you must compress the plastic of the boot when depressing the tab on unlock the connector.
It is difficult to tell how durable an item is. By the time you know how durable something is in the real world it is likely to be discontinued or silently changed. Some people base their decisions on brand trust. Unfortunately if there are financial problems or a need to look good to the stock market it's very tempting for a manufacturer to sell out their brand for a quick buck.
My experience is it's very rare that the connector itself gets busted. A far bigger problem is solder joints and/or PCB tracks getting busted as the connector moves relative to the PCB.
I can't see why IO would become a problem
Compatibility between 5V logic and 3.3V logic is a somewhat complex area. Some 5V devices can accept 3.3V signals and some 3.3V devices are tolerant of 5V signals but many aren't (on both sides). So much of the time you end up having to building level translation circuits to interface 5V stuff to 3.3V stuff.
A further complication is that most IO on microcontrollers has independent direction control for every pin. So if you want to make a fully general level translation circuit you have to 1: find a bidirectional buffer chip with individual direction lines (or build the equivilent of one out of tristate buffers) and 2: find enough IO to control the direction of all those buffers. It's doable but I'm not sure if it's doable at a cost (in both size and money) that would be acceptable to the ardunio guys. I certainly don't see any evidence of it on the photo in TFA
Some discussion implementations (slashdot's D2 being one of them) make it easy to accidently lose a post by navigating away from the page or sometimes even by actions on the page. Thanks to the heavy use of javascript the auto-refilling of the text boxes that some browsers do when you use the back button doesn't work with D2 either.
Yes one could use notepad but that doesn't have a spell checker.
And not everyone uses a browser recent enough to have a spell checker.
Afaict it's not the fiber itself that is expensive It's the laser diodes, photodiodes, precision connections to the fiber, protection against fractured fibers from kinked cables and so on that make a high speed fiber system expensive. Especially if it has to be made "idiot-proof".
TOSLINK (optical version of S/PDIF) is indeed cheap but that is because the low speeds let them get away with REALLY low grade optical components (including polymer fiber that is less prone to damage than glass fiber.
With the rate power consumption of CPU cores has been increasing
You are a few years behind the curve. Since the pentium D CPU TDP (roughly power consumption under the highest normal load) has stayed pretty much flat while core counts and performance of individual cores have gone up (despite the drop in clockspeeds)
The power consumption per core has been going down in recent times. The pentium D 965 extreme has two cores, a clockspeed of 3.73 GHz and a TDP of 130W. The i7-990x has six cores.a nominal clockspeed of 3.47 GHz (plus turbo boost) and a TDP of 130W.
Comparing the two processors (made about 5 years apart) the later processor has the same TDP, three times as many cores and slightly lower clockspeed (but each core has far better performance at a given clockspeed)
The problem with that approach is that if foocorp doesn't have any relavent patents of their own and barcorp comes along and claims that foocorp violates their patents then barcorp has foocorp over a barrel. They may choose to grant a license at a reasonable rate, they may not. If they don't then afaict there is nothing foocorp can do about it.
OTOH if foocorp and barcorp both have relavent patent portfolios then a balance comes into play. Since neither side wants to be destroyed they have to agree not to destroy each other.
Quite likely neither is relavent. At least here individual research groups aren't charged for electricity and labour is pretty much a fixed cost. As for aircon that isn't really an issue at this scale.
As was seen this year in Fukushima. Even if you didn't get cancer or be made infertile (afaik people not working in the station did not suffer these risks) you were probably not very happy about what happened if you lived in the evacuation zone.
Nuclear isn't the only industry whose accidents can displace a community on a near permanent basis.
http://en.wikipedia.org/wiki/Centralia,_Pennsylvania
And that is just direct impacts. If global warming from fossil fuel usage causes large sea level rises or shifts in what land is arable that could displace a lot of people to.
Opensource and open standards are different things
But closely intertwined. In particular "open" standards can be divided into two main categories.
Open and unencumbered standards where anyone can implement the standard without having to sign license agreements or pay royalties.
Open but encumbered standards where anyone can buy and read a copy but to implement the standard legally you have to sign a patent license agreement and pay license fees.
Of couse there is the further complication that thanks to how patents work it is impossible to be sure a standard is completely free of patent encumberenaces. You can only say that there are no known encumberances. Even being older than the typical lifetime of a patent isn't surefire protection in some countries.
as long as I can still look at my archives in 10-20-50 years.
And the best way to maximise the chance of that is to have the widest support possible for said standard. FOSS projects generally can't afford to pay patent license fees and even if they could the patent licenses are generally incompatible with major FOSS licenses. Sometimes FOSS projects implement encumbered standards anyway (either illegally or through basing the project in a country where the encumberance doesn't apply) but a few legal threats could easilly cause the project to pull the feature (yes you can go on using the old version for a while but as the underlying platforms move on that is likely to get harder and harder).
The real questions are
1: do you replace the entire infrasturcture with fiber or do you use existing (coax or twisted pair) infrastrucutre for part of the system?
2: if you do replace the whole infrastucture with fiber do you give each customer their own fiber right back to the exchange? do you have fiber switches out in the field? do you build a passive fiber network with TDMA or CSMA? do you use WDM to give multiple customers independent channels in the same fiber?
The answers chosen to these questions are likely to have a big impact on both the cost of the deployment and the speeds deliverable. In particular a deployment where you only replace the lines to the street is likely to be far cheaper than anything else but also give far lower speeds.
The reason this post is stupid is that infrastructure is long term. When you go to the trouble of sending out a crew to dig up and put fiber in the ground your putting in an infrastructure asset that should have a 15-30 year lifespan. The fact that can average machine can't saturate it today means we're being forward thinking.
True to an extent but remember there is nothing stopping the provider from changing how the fiber is used without changing the fiber itself.
Another alternative is to check the object through a method after construction, which a lot of STL objects do, but that's kind of messy.
Kind of messy but given that the whole point of constructors is that they are called automatically so the object is never in an undefined state you are going to have this fundamental problem.
a third alternative is to have the constructor create the object in an empty form (which can generally be done without memory allocation) and then fill it with data through regular functions.
Classes, inheritance, templates, virtual functions etc are all language features not runtime library features.
. If you say that your car goes x mph and does y mpg I have to go to google or wolfram alpha to convert those numbers to km/h and L/100km.
Adding to the fun you have to be careful with mpg conversions because brits use miles per imperial gallon while americans use miles per US gallon but both use the term mpg.
IMO it should pay dividends the same as common stock and should convert to common stock on the death of the holder.
And it shouldn't be used for 100% of compensation, just a sufficiant proportion of it to give the employee a reason to care about the long term future of the company.
Be aware that Yahoo Japan is actually a seperate company and only about a third of it is owned by Yahoo.
64GB system memory was $1500 a few months ago. I think it is below $1000 now.
The ram itself seems to be $880 (four kits each with 2x8GB registered ECC modules from newegg) but you also have to consider that to actually put that much ram in one system rules out use of a normal desktop platform because desktop processors (when used with currently available memory modules) simply can't support that much memory. Nor can many low end server platforms.
That means that you will likely have to spend a lot more money on your CPU and motherboard to get a platform that supports your 64GB of memory (the cheapest option appears to be G34 but it's still about $500 and the 8 slow cores you get may not be a good fit for your workload)