It won't disable it, because that would break a load of existing code, but it may modify the microcode to replace a single FMA micro op with a sequence of slower micro ops.
And some of them were security critical. Remember having to disable hyperthreading because it let one thread snoop on another's cache? Or having to disable transactional memory because it allowed arbitrary physical memory manipulation from an untrusted process.
I found GPS directions were a good way of getting to know my way around when I moved here, but I looked at the planned route before I set off (walking or cycling) and was then able to look more at my surroundings. The phone in my pocket would tell me which turning to take, and so I'd get to know the routes very quickly and not need to look at the map. After a week, I wasn't bothering to set the directions. Having the GPS also gave me more confidence to explore - I could wander around in a random direction and know that I'd never actually get lost. Just like any other tool, the benefits of a satnav depend entirely on how you use it.
We bought a new house last year and it was a few months before it was on most maps[1]. The number of delivery drivers that got lost was incredible. We gave them clear directions from the nearest main road, but most of them couldn't manage to follow them. Our road is just past a large office block and so all they needed to do was get to that office building (been there for decades, on all of the maps) and then follow the road around. We also told them which turning to take off the road that led to that one. It appears that most of them could not read street signs and didn't know whether they were heading towards or away from the city centre.
[1] OpenStreetMap had it before we looked at it, Google still doesn't. I still find the comment that OSM is a parody from the Google Maps lead amusing - apparently looking pretty is far more important than having accurate data.
A lithium-ion battery is basically a bomb with a small circuit saying 'don't explode, don't explode, don't explode'. They're banned from aircraft holds because the don't-explode circuits turn out not to be as reliable as previously thought. It amazes me that I'm allowed to carry a few of them onto a plane, but not a small bottle of water (though I can buy one at an overprices shop, or I can buy something a lot more flammable in Duty Free).
Google doesn't 'sell' your personal information, because they consider it an asset. They do, however, sell the right to put adverts in front of people matching certain criteria. This has been shown to allow third parties to infer information that Google has about you. The first attack demonstrating this placed adverts for a pizza place with a free pizza code and asked for it to target people in a specific region who identified as gay. Users then went to the pizza company web site, entered their name, address, and coupon, and suddenly someone else had a database of people that Google knew were in a certain area and gay - a very useful took for hate groups. That was years ago and there are now far more subtle attacks that give better information.
You can get 384GB(6x64) of ECC server ram for $5,202 on newegg...
So you can get 384GB of RAM for a bit more than 8TB of NVMe flash (not sure what the XPoint pricing is going to be).
For a 2K to 5K difference...the ram memory beats all...and it is not THAT much more expensive.
It's about 20 times as expensive per GB. You're arguing that the better latency and throughput of the RAM is going to outweigh the increased capacity of the NVRAM. That's by no means clear. If your working set fits entirely into that 384GB RAM cache, then the RAM will definitely be faster, but your working set is 1-4TBs (not that uncommon for a SAN device) then the RAM solution is going to be a lot slower than one with 128GB of RAM and 4TB of NVMe flash. For a smaller array, that $5K price difference is enough to replace all of the spinning rust disks with NVRAM, which will give you even better performance.
If you want under 2K...what about a 2xM.2 pcie card(~$100-$150) with two 1TB Samsung PRO 950 M.2 cards for ~$1400? Run that in raid0 and you would probably be equal or better then the current Xpoint offering spec wise.
Except that the one place that XPoint does seem to do a lot better than flash is in random reads, which is the main performance bottleneck for a cache device.
The endurance listed is 30 drive writes per day for the 375GB model. That's 11.25TB/day, or about 130MB/s sustained 24/7 writes. A cache device should be spending 90% of its time being read, rather than written (or it's not doing that good a job as a cache[1]), which means that even if you're reading from it at about half of its peak rate every second of every day and getting then it's going to last its rated lifetime. Additionally, for a cache device, we really don't care if it burns out. When it does, we don't lose any data, we just take a performance hit until it can be replaced (we've had flash L2 ARC devices die quite a few times), but we really do care a lot about random read performance from the cache drive while it's working.
[1] I just checked the ARC stats on one of our machines. We're getting over 98% hit rates from the ARC and only about 55% from the L2 ARC, so we'd probably wear out one of these a bit faster. That said, this machine has 256GB of RAM and is only using 320GB of L2 ARC, so it's not too surprising that the L2 ARC is not doing a great job: once the working set goes past about 150-200GB (ARC size), it's going to grow past 500GB (ARC + L2 ARC size) pretty quickly.
I don't know if you have a comprehensible tax code over there, but over here most people just pay someone else (or some website) to do their taxes, and if that person (etc.) could get them another $100 in refund they'd gladly pay them $50 to do it.
Almost no one here pays an accountant to do their taxes. The vast majority simply use PAYE (Pay As You Earn), whereby their employer deducts the tax at source (banks will also withhold tax on interest automatically) and they get a statement each year. Self-employed people and people with other sources of income file a tax return, but for most people this is very simple (income in this box, expenses in this one, hit submit). The only people who get an accountant to do their taxes are people with a limited liability company or two (these must be done by an accountant) or people with so much money that they can't keep track of it themselves. For most people, the possible savings are far less than the cost of having a professional file their taxes, so there's little incentive to do so.
Maybe you feel you're getting your money's worth. We certainly are not.
That's probably a problem that you should try to fix.
How many individuals "do the right thing" and pay extra taxes beyond what they are legally required to pay?
It depends on what you mean by 'legally required to pay'. The amount of tax that I pay is the amount that you get by taking my salary and multiplying the parts of it in different tax brackets by the tax rates. There are a huge number of tax avoidance schemes that I could use to reduce my tax burden, but I've received a lot of benefits from living in a functioning society with a working social safety net and I can quite easily afford the taxes, so I'd rather just pay them. I doing so, I am not in a minority, this is precisely what most other people in the UK do.
If this is your definition of paying more than you are 'legally required to pay', then most people do, but most large corporations don't.
O'Reilly books have always been bad, but they were also cheap. You could get an O'Reilly book for under half the cost of one from a competing publisher, so you'd overlook the poor writing and poor editing (and there were a few gems that were fairly well written). The change is that the basis for comparison is not expensive books from another publisher, it's free blog articles online. O'Reilly books look really expensive compared to free, and the price difference isn't reflected in the quality.
Neither extreme is a panacea, but modern compilers are good at inlining, so there's little performance difference between the two and small functions do make sense when each is an individual step in the algorithm. The function that calls a number of small functions then reads like an abstract description of the algorithm that you're implementing and people only need to see the details if they want to. What you really want is a literate programming system where you can embed default detail levels into the code and have the IDE so that when someone opens your code for the first time they just see comments describing the steps of the algorithm and high-level flow control, but can easily unhide any of the details. Small functions give you an approximation of that.
Because, in general (and there are a few exceptions), they're poorly written, very poorly edited (typos and incomprehensible sentences scattered throughout) and written by people who either don't understand the subject or don't understand how to explain things (or, in some cases, both). O'Reilly got into its current position by undercutting other publishers, which it did by cutting costs throughout the process. Unfortunately, this included paying authors and editors less, which shows in their output.
He is claiming that Beowulf clusters are one of the biggest contributions to computer science by NASA. Beowulf clusters are commodity hardware combined with open source software that predates NASA's involvement. NASA's contribution was to assemble the bits and give it a name. They weren't the first organisation to build a cluster using that software. There was almost nothing new in the software, network topology, or hardware in their design, yet this is what he's claiming as evidence of NASA's great contributions to computer science.
I'm pretty sure your phone won't die "unexpectedly"
I updated the apps on my phone before a flight a year ago and found that there was a bug in the SMS app that prevented the phone from sleeping. I lost 30% of the charge on the way to the airport. If I hadn't noticed, it would have been flat long before I tried to board my connecting flight. I don't know what their procedure is to handle this, but I'd imagine that it involves printing paper boarding passes, which requires them to keep all of the associated infrastructure.
So, what you're saying is, NASA's biggest contribution to computer science was giving a name to a combination of software written by other people and pre-dating NASA's involvement?
Consider IIWW Eastern Front. Soviets won, despite taking roughly 2x as high losses as Germany. Simply, Soviets had more than 2x the power to throw at them.
That's actually a better example than you think. The Soviets lost a lot more men and tanks, but their tanks were a lot cheaper to produce (which is why they were destroyed more often).
This isn't a new issue with drones. I recall a debriefing from the Iraq war where an Apache pilot was asked what he thought the cost of the missiles that he fired and the value of the convoy of trucks that he destroyed were. Hopefully the trucks contained something important to the enemy's war effort, because otherwise he'd spent a few million dollars to destroy a few tens of thousands of dollars. One of the strategies that's been core military doctrine since the US-Soviet conflict in Afghanistan (and Vietnam before that) has been to try to make your enemy outspend you by a large margin. A superpower like the USA might be able to afford to spend $100 for every $1 you spend, but starts to struggle when it has to spend $1,000 or even $10,000 for every $1 that you spend.
Every cluster in every data center in the world today is a direct descendent of NASA's design
There are very few clusters in datacentres. Most distributed compute jobs in datacentres use things like MapReduce or similar, which (unlike Beowulf) are specifically designed around high-latency, high-bandwidth interconnects. There are clusters in supercomputing centres. They overwhelmingly use MPI, which predates NASA's Beowulf system by three years and was an integral part. MPI was developed with NSF and ARPA funding. NASA built a famous commodity system using it, but they weren't the first, nor were they the only group doing so. The other core part of Beowulf, PVM, is even older (1989) and is also not a NASA technology.
Beowulf definitely succeeded in becoming a Slashdot meme, but its impact on cluster computing was far less. I've worked on a number of systems that people referred to as 'beowulf clusters,' but not a single one was actually running the Beowulf software.
among the most noteworthy achievements was the invention of clusters
If that's among their most noteworthy achievements, it doesn't say much. Clusters were a pretty obvious development: they're basically an approximation of existing supercomputer designs using commodity hardware, and come with all of the limitations that you'd expect. If NASA hadn't invented them, someone else would have done, just as multiple people have replicated the high-availability features of mainframes on commodity hardware with VMs.
Slashdot Mirror
It won't disable it, because that would break a load of existing code, but it may modify the microcode to replace a single FMA micro op with a sequence of slower micro ops.
And some of them were security critical. Remember having to disable hyperthreading because it let one thread snoop on another's cache? Or having to disable transactional memory because it allowed arbitrary physical memory manipulation from an untrusted process.
I found GPS directions were a good way of getting to know my way around when I moved here, but I looked at the planned route before I set off (walking or cycling) and was then able to look more at my surroundings. The phone in my pocket would tell me which turning to take, and so I'd get to know the routes very quickly and not need to look at the map. After a week, I wasn't bothering to set the directions. Having the GPS also gave me more confidence to explore - I could wander around in a random direction and know that I'd never actually get lost. Just like any other tool, the benefits of a satnav depend entirely on how you use it.
We bought a new house last year and it was a few months before it was on most maps[1]. The number of delivery drivers that got lost was incredible. We gave them clear directions from the nearest main road, but most of them couldn't manage to follow them. Our road is just past a large office block and so all they needed to do was get to that office building (been there for decades, on all of the maps) and then follow the road around. We also told them which turning to take off the road that led to that one. It appears that most of them could not read street signs and didn't know whether they were heading towards or away from the city centre.
[1] OpenStreetMap had it before we looked at it, Google still doesn't. I still find the comment that OSM is a parody from the Google Maps lead amusing - apparently looking pretty is far more important than having accurate data.
A lithium-ion battery is basically a bomb with a small circuit saying 'don't explode, don't explode, don't explode'. They're banned from aircraft holds because the don't-explode circuits turn out not to be as reliable as previously thought. It amazes me that I'm allowed to carry a few of them onto a plane, but not a small bottle of water (though I can buy one at an overprices shop, or I can buy something a lot more flammable in Duty Free).
It's especially weird, as I'm not allowed to put a laptop in checked luggage because they don't allow large lithium-ion batteries in the hold.
Most plans let you finance the phone for 24 months at 15-50% ARP, yet people are still gullible enough to think it's a subsidised price.
Google doesn't 'sell' your personal information, because they consider it an asset. They do, however, sell the right to put adverts in front of people matching certain criteria. This has been shown to allow third parties to infer information that Google has about you. The first attack demonstrating this placed adverts for a pizza place with a free pizza code and asked for it to target people in a specific region who identified as gay. Users then went to the pizza company web site, entered their name, address, and coupon, and suddenly someone else had a database of people that Google knew were in a certain area and gay - a very useful took for hate groups. That was years ago and there are now far more subtle attacks that give better information.
The first article I read that promised MRAM real soon now was around 2008. I'm not holding my breath.
You can get 384GB(6x64) of ECC server ram for $5,202 on newegg...
So you can get 384GB of RAM for a bit more than 8TB of NVMe flash (not sure what the XPoint pricing is going to be).
For a 2K to 5K difference...the ram memory beats all...and it is not THAT much more expensive.
It's about 20 times as expensive per GB. You're arguing that the better latency and throughput of the RAM is going to outweigh the increased capacity of the NVRAM. That's by no means clear. If your working set fits entirely into that 384GB RAM cache, then the RAM will definitely be faster, but your working set is 1-4TBs (not that uncommon for a SAN device) then the RAM solution is going to be a lot slower than one with 128GB of RAM and 4TB of NVMe flash. For a smaller array, that $5K price difference is enough to replace all of the spinning rust disks with NVRAM, which will give you even better performance.
If you want under 2K...what about a 2xM.2 pcie card(~$100-$150) with two 1TB Samsung PRO 950 M.2 cards for ~$1400? Run that in raid0 and you would probably be equal or better then the current Xpoint offering spec wise.
Except that the one place that XPoint does seem to do a lot better than flash is in random reads, which is the main performance bottleneck for a cache device.
The endurance listed is 30 drive writes per day for the 375GB model. That's 11.25TB/day, or about 130MB/s sustained 24/7 writes. A cache device should be spending 90% of its time being read, rather than written (or it's not doing that good a job as a cache[1]), which means that even if you're reading from it at about half of its peak rate every second of every day and getting then it's going to last its rated lifetime. Additionally, for a cache device, we really don't care if it burns out. When it does, we don't lose any data, we just take a performance hit until it can be replaced (we've had flash L2 ARC devices die quite a few times), but we really do care a lot about random read performance from the cache drive while it's working.
[1] I just checked the ARC stats on one of our machines. We're getting over 98% hit rates from the ARC and only about 55% from the L2 ARC, so we'd probably wear out one of these a bit faster. That said, this machine has 256GB of RAM and is only using 320GB of L2 ARC, so it's not too surprising that the L2 ARC is not doing a great job: once the working set goes past about 150-200GB (ARC size), it's going to grow past 500GB (ARC + L2 ARC size) pretty quickly.
Charitable donations are difficult to claim in the UK. Pension payments are take from pre-tax income automatically via the PAYE system.
I don't know if you have a comprehensible tax code over there, but over here most people just pay someone else (or some website) to do their taxes, and if that person (etc.) could get them another $100 in refund they'd gladly pay them $50 to do it.
Almost no one here pays an accountant to do their taxes. The vast majority simply use PAYE (Pay As You Earn), whereby their employer deducts the tax at source (banks will also withhold tax on interest automatically) and they get a statement each year. Self-employed people and people with other sources of income file a tax return, but for most people this is very simple (income in this box, expenses in this one, hit submit). The only people who get an accountant to do their taxes are people with a limited liability company or two (these must be done by an accountant) or people with so much money that they can't keep track of it themselves. For most people, the possible savings are far less than the cost of having a professional file their taxes, so there's little incentive to do so.
Maybe you feel you're getting your money's worth. We certainly are not.
That's probably a problem that you should try to fix.
How many individuals "do the right thing" and pay extra taxes beyond what they are legally required to pay?
It depends on what you mean by 'legally required to pay'. The amount of tax that I pay is the amount that you get by taking my salary and multiplying the parts of it in different tax brackets by the tax rates. There are a huge number of tax avoidance schemes that I could use to reduce my tax burden, but I've received a lot of benefits from living in a functioning society with a working social safety net and I can quite easily afford the taxes, so I'd rather just pay them. I doing so, I am not in a minority, this is precisely what most other people in the UK do.
If this is your definition of paying more than you are 'legally required to pay', then most people do, but most large corporations don't.
O'Reilly books have always been bad, but they were also cheap. You could get an O'Reilly book for under half the cost of one from a competing publisher, so you'd overlook the poor writing and poor editing (and there were a few gems that were fairly well written). The change is that the basis for comparison is not expensive books from another publisher, it's free blog articles online. O'Reilly books look really expensive compared to free, and the price difference isn't reflected in the quality.
Did you really just say that make is arcane, but XML is straightforward?
Neither extreme is a panacea, but modern compilers are good at inlining, so there's little performance difference between the two and small functions do make sense when each is an individual step in the algorithm. The function that calls a number of small functions then reads like an abstract description of the algorithm that you're implementing and people only need to see the details if they want to. What you really want is a literate programming system where you can embed default detail levels into the code and have the IDE so that when someone opens your code for the first time they just see comments describing the steps of the algorithm and high-level flow control, but can easily unhide any of the details. Small functions give you an approximation of that.
Because, in general (and there are a few exceptions), they're poorly written, very poorly edited (typos and incomprehensible sentences scattered throughout) and written by people who either don't understand the subject or don't understand how to explain things (or, in some cases, both). O'Reilly got into its current position by undercutting other publishers, which it did by cutting costs throughout the process. Unfortunately, this included paying authors and editors less, which shows in their output.
He is claiming that Beowulf clusters are one of the biggest contributions to computer science by NASA. Beowulf clusters are commodity hardware combined with open source software that predates NASA's involvement. NASA's contribution was to assemble the bits and give it a name. They weren't the first organisation to build a cluster using that software. There was almost nothing new in the software, network topology, or hardware in their design, yet this is what he's claiming as evidence of NASA's great contributions to computer science.
I'm pretty sure your phone won't die "unexpectedly"
I updated the apps on my phone before a flight a year ago and found that there was a bug in the SMS app that prevented the phone from sleeping. I lost 30% of the charge on the way to the airport. If I hadn't noticed, it would have been flat long before I tried to board my connecting flight. I don't know what their procedure is to handle this, but I'd imagine that it involves printing paper boarding passes, which requires them to keep all of the associated infrastructure.
So, what you're saying is, NASA's biggest contribution to computer science was giving a name to a combination of software written by other people and pre-dating NASA's involvement?
Consider IIWW Eastern Front. Soviets won, despite taking roughly 2x as high losses as Germany. Simply, Soviets had more than 2x the power to throw at them.
That's actually a better example than you think. The Soviets lost a lot more men and tanks, but their tanks were a lot cheaper to produce (which is why they were destroyed more often).
This isn't a new issue with drones. I recall a debriefing from the Iraq war where an Apache pilot was asked what he thought the cost of the missiles that he fired and the value of the convoy of trucks that he destroyed were. Hopefully the trucks contained something important to the enemy's war effort, because otherwise he'd spent a few million dollars to destroy a few tens of thousands of dollars. One of the strategies that's been core military doctrine since the US-Soviet conflict in Afghanistan (and Vietnam before that) has been to try to make your enemy outspend you by a large margin. A superpower like the USA might be able to afford to spend $100 for every $1 you spend, but starts to struggle when it has to spend $1,000 or even $10,000 for every $1 that you spend.
Every cluster in every data center in the world today is a direct descendent of NASA's design
There are very few clusters in datacentres. Most distributed compute jobs in datacentres use things like MapReduce or similar, which (unlike Beowulf) are specifically designed around high-latency, high-bandwidth interconnects. There are clusters in supercomputing centres. They overwhelmingly use MPI, which predates NASA's Beowulf system by three years and was an integral part. MPI was developed with NSF and ARPA funding. NASA built a famous commodity system using it, but they weren't the first, nor were they the only group doing so. The other core part of Beowulf, PVM, is even older (1989) and is also not a NASA technology.
Beowulf definitely succeeded in becoming a Slashdot meme, but its impact on cluster computing was far less. I've worked on a number of systems that people referred to as 'beowulf clusters,' but not a single one was actually running the Beowulf software.
among the most noteworthy achievements was the invention of clusters
If that's among their most noteworthy achievements, it doesn't say much. Clusters were a pretty obvious development: they're basically an approximation of existing supercomputer designs using commodity hardware, and come with all of the limitations that you'd expect. If NASA hadn't invented them, someone else would have done, just as multiple people have replicated the high-availability features of mainframes on commodity hardware with VMs.