big storage isn't that interesting a problem - anyone can do it, and at fairly modest cost. there are a few smop's if you want really big and/or really fast storage, but they're clearly solvable, and affordable.
it's quite astonishing that these researchers, who are otherwise well-reputed, have missed the whole point of government sponsorship of super-* facilities: to do what can't be done otherwise. mostly, that means running traditional supercomputer jobs, those that are tightly coupled. people who have loosely-coupled jobs have long ago bailed from the supercomputing arena, and have been building their own clusters. similarly, there's no unique advantage to centralizing data storage, and a huge disadvantage (bottlenecks in and out).
I have to wonder whether Markoff badly munged the intent of the Gray/Bell paper, since the way he presents it is internally inconsistent. that is: the gov should spend huge bucks on massive centralized storage, but computing should be decentralized ala grids. oops, how is all that compute power supposed to move data to/from the three national data repositories?
perhaps the central problem here is the fallacy shared by grid-o-philes: that networking is getting dramatically faster. take a look at your own network: if you are lucky enought to have gigabit to the desktop, when did that upgrade happen (probably 100 upgrade happen? what kind of speed did you get on your last big download?
I've experienced a speedup of something between 10 and 50x in the past, say, 10 years. that's pathetic, when compared to the speedup we all have experienced in CPU power, memory size/speed, and disk size/speed.
there's no Moore's Law of networking: no n^2 process to keep accelerating (unlike die or disk densities). yes, there are technological improvements, and yes, you can gang cables together to scale bandwidth almost linearly. no such help for latency, though. and technological improvements are neither infinite nor increasing. that means that the network is becoming more of a bottleneck, not less.
fixed or random delays work find - as long as you don't leak info about the key. this is mentioned in the paper.
the attack is a nice proof-of-concept, but almost totally impractical. notice that their "lan" was a completely private switch, with just the attack and target machines on it. if your lan is anything like mine, there's huge variance that will defeat the attack. not to mention that any other activity on the target machine will make the attack harder, or any kind of traffic shaping (they were doing 140 probes per second). not to mention simply turning on blinding.
it's great to see this vuln demonstrated in the lab, but really no need to worry about it.
ahhh, right. so why is it that people have continued to choose unsafe, broken, archaic platforms for so many decades? sorry, it's simply not valid to claim that laziness is the reason.
it's ridiculous to lump unix and windows into the same category. though it really makes an excellently condescending slam!
this is revisionism. for instance, the PM interface was specified by IBM - as a gratuitously different version of existing interfaces. the code was most definitely done by Msft, though (I was there).
basically, IBM committee-designed OS/2 into oblivion, and Msft simply gave up on it, since they had two routes around: windows on the low-end, and NT OS/2 (yep) on the high end. discarding ia32 in favor of risc chips and microkernels was pretty fashionable at the time.
Sun is just as anal-retentive/control-freak/lawyer-driven as Microsoft, only less successful. Both of them are rather pitiful.
the really odd thing about this development is that Sun inexplicably thinks that the enemy of its enemy is yet another enemy. well, I guess it's also mystifying that they think anyone gives a damn about Solaris or their hardware. as if the freeness of SO6 would make anyone pay good money for Sun's crappy hardware.
let's be honest, the states AG's are just slimeball politicos owned by companies who weren't good enough (as evaluated by the market/customer) to compete with msft. make no mistake: their "remedy" is nothing short of nationalizing windows.
democracy has become a motherhood/apple-pie thing - it's not even a concept, it's more of a nebulous emotional state. why has it gained this unquestioned positive spin? democracy is mostly a way for incapacitating government, for avoiding putting too much power in one place. all elections are marketing competitions. sure, occasionally there's some politico who actually has something on his agenda besides being elected/reelected. but the nature of democracy, at least parlimentary forms, means that the few principled participants will be inherently dilluted by useless, photogenic seatwarmers.
this is great for globalization, since ineffectual government avoids doing anything dramatic to multinationals, except the usual extortion/tax.
what's missing? the real goal should be liberty, not democracy. sure, democracy might be a means to liberty, but it's NOT THE GOAL. they're orthogonal - liberty is about policy (principles); democracy is about mechanism.
there's a "meta-politics" that's not being discussed - that's why this is such a fuzzy topic. why is terrorism wrong? what is the real conflict between the West and Taliban-style fundamentalism? the principle of individual liberty - that if you want to live a Wahabi life, you're perfectly free to do so in the West. you just can't coerce someone else into doing it. liberty/non-coercion is what we should be talking about, not democracy.
and this is relevant to the undercurrent of discussion about how the net will effect society in the future. it's obvious that strong crypto, peer-to-peer, net-communities are powerful forces that, in the absence of some kind of apocalypse of talibanhood, will become dominant. they have a sense of historic inevitability. they're also profoundly liberty-based, self-organizing, non-coercive. even anti-authoritarian. and globalizing.
but how can that be? wasn't seattle supposed to be the rise of a non-hierarchical, self-organized political force devoted to overthrowing globalization? there's a contradiction there: absence of hierarchically imposed limits are what permits these anti-globalization people to demonstrate. (and demonstration != democracy!) the anti-globalization freaks are opposed to commerce being the "working fluid" of globalization. it's not the multinationals that they oppose, it's the fact that MN's are based on an international currency market that in effect makes my 8-hours of labor in the West incomparable to 8 hours of labor by someone in the 3rd world. this seems irrational to me, or at least based on principles I don't share (ie, more "from each according to his ability" rather than "to each according to the market price of his ability").
if online/crypto is a globalizing force, it's not necessarily going to cause a redistribution of wealth, or a replacement of property as the measure of wealth. and that's the tip of another iceberg - that some people want ideas to become as ownable as property; not surprising, these "idea hegemonists" are large, Western, multinational corporations...
don't be stupid: no one is talking about a second latency. a decently configured machine will never see more than a couple hundred ms event latency; the question is how much performance it's worth sacrificing to get that down to 1ms.
it's all very well to say that you want to trade
5% of normal performance for a 200% improvement in latency. but why does anyone need better latency? afaikt, the latency here is strictly for people who want to do RT audio effects. this has nothing to do with audio playback, which has no latency sensitivity (because of buffering). this also has nothing to do with "feel", since humans are terribly slow, and cannot possibly feel the difference between 5 and 10ms.
I hope that Linus will look at whether these patches hurt the normal case. "normal" means things like kernel compilation, not just an arbitrary latency measure and dbench (one of the least realistic benchmarks possible!)
there are good reasons to be skeptical of all-out premptiveness: it will unavoidably lower throughput in easy-to-define cases. any intro OS text will talk about optimal scheduling, where 'optimal' requires a definition of throughput or some other metric. preemptive kernels will context switch more, and will probably interfere with the natural 'batching' that happens when a big job runs for a while. think about caches: you never want to switch unless you must.
this is not an argument against low-latency! it's an arguement against lowest latency as an absolute; we need to set a target (5ms would be fine imo) and meet it. going beyond such a goal will hurt the normal case.
beowulf isn't just clustering, it's compute-intensive clustering. that means that run-of-the-mill 100bT cards will likley not be enough - think Myri2000, SCI, Quadrics, gigabit.
whitebox machines also tend to be truely horribly configured - you probably don't want UMA video stealing dram bandwidth, for instance. and even for a smallish cluster, it's smart to get better power supplies and fans, simply because the noname stuff has short expected lifespans.
they suck, basically. a 4-way Xeon is sitting
on a miserable 800 MB/s (450 sustained) bus,
so unless your code is totally cache-friendly,
the CPUs will starve. and if your code is so
cache-friendly, you should probably be looking
at different hardware in the first place.
it's easy to dislike Microsoft - after all, they are remarkably successful at offering obviously flawed and limited products. yes, they clearly use network effects to promote their products, which is exactly what any good business does. their actions are quite possibly too agressive, agressive enough to infringe on certain laws. but they are not immoral. the moral aspect of this issue is the willingness of companies like Sun, Oracle and Netscape to resort to politics when they can't compete, when their products are worse than Microsoft's. you may disagree on how the nation's antitrust laws apply to Microsoft, but why should it be appropriate for these failed competitors to get the cops to beat up on Microsoft? it's clearly not just a matter of "enforcing the law" or some such tripe - after all, Intel is free to permute its pinouts in a blatantly anticompetitive manner, or simply refused to lisence them.
if Sun/Oracle/Netscape were serious, they'd be asking for updated or more specific laws. immoral companies routinely use stupid politicians, gullible judges and ridiculous laws to abuse their competitors. routine doesn't make right.
it's disingenuous to pretend that you don't realize there's no clear distinction between "abusing the power of a monopoly" and simple good business. how far do you think msft should bend over to make their competition thrive?
in reality, as opposed to the courts or press, competitors to msft have been stupid and ineffectual. and msft has done a remarkable job of producing products that are eminently affordable, and get the job done for hundeds of millions of users. backwards compatibility is msft's main product, and they do it well.
if msft is broken up, then clearly Intel should, so that the Celeron product line can be uncrippled. and bundling rdram with CPUS! that's obviously abusing their monopoly position - DOJ should require them to support PC2100 immediately!
you have to work hard to find any tangible indication of how fast this machine is. Sun hasn't submitted any spec scores, which is certainly a bad sign. I found 20/21 spec95 buried in the page for the bare processor. that's HORRIBLE, since you have to work pretty hard to find a PC that slow these days. maybe a Celeron system; even so, a cheaper Duron would be significantly faster.
amusing: takes PC133 sdram, but runs the bus at only 83 or 100 MHz.
it's critical to realize that these boxes, while impressive, are mainframes, not supercomputers. for instance, if you look at the 64-way E10K Stream score, you realize each CPU is seeing a measly 150 MB/s or so! these are clearly machines designed mainly for driving big IO, not running applications that need more than trivial memory bandwidth.
so maybe we've learned some things. progress does mean that the frontier looks different, even that the frontier is gone. in concrete terms, 30 years of Unix has given us wonderfully practical systems; wasn't that the point? after all, this is an applied field, not purely abstract, nor purely artisanal.
MSFT would be smart to shift just certain products north. hell, how about just the free one, on which a 90% tarriff would be uh, $dumb... actually, it would be hillarious if they shifted part of the company (office or windows). Canada would work fine, but then again, India might too.
Slashdot Mirror
it's quite astonishing that these researchers, who are otherwise well-reputed, have missed the whole point of government sponsorship of super-* facilities: to do what can't be done otherwise. mostly, that means running traditional supercomputer jobs, those that are tightly coupled. people who have loosely-coupled jobs have long ago bailed from the supercomputing arena, and have been building their own clusters. similarly, there's no unique advantage to centralizing data storage, and a huge disadvantage (bottlenecks in and out).
I have to wonder whether Markoff badly munged the intent of the Gray/Bell paper, since the way he presents it is internally inconsistent. that is: the gov should spend huge bucks on massive centralized storage, but computing should be decentralized ala grids. oops, how is all that compute power supposed to move data to/from the three national data repositories? perhaps the central problem here is the fallacy shared by grid-o-philes: that networking is getting dramatically faster. take a look at your own network: if you are lucky enought to have gigabit to the desktop, when did that upgrade happen (probably 100 upgrade happen? what kind of speed did you get on your last big download? I've experienced a speedup of something between 10 and 50x in the past, say, 10 years. that's pathetic, when compared to the speedup we all have experienced in CPU power, memory size/speed, and disk size/speed.
there's no Moore's Law of networking: no n^2 process to keep accelerating (unlike die or disk densities). yes, there are technological improvements, and yes, you can gang cables together to scale bandwidth almost linearly. no such help for latency, though. and technological improvements are neither infinite nor increasing. that means that the network is becoming more of a bottleneck, not less.
fixed or random delays work find - as long as you don't leak info about the key. this is mentioned in the paper.
the attack is a nice proof-of-concept, but almost totally impractical. notice that their "lan" was a completely private switch, with just the attack and target machines on it. if your lan is anything like mine, there's huge variance that will defeat the attack. not to mention that any other activity on the target machine will make the attack harder, or any kind of traffic shaping (they were doing 140 probes per second). not to mention simply turning on blinding.
it's great to see this vuln demonstrated in the lab, but really no need to worry about it.
USF? I don't think so: top500 lists around 1 TF total at UFL, but .ca can beat that by 20-50% without counting any of the many beowulf clusters.
ahhh, right. so why is it that people have continued to choose unsafe, broken, archaic platforms for so many decades? sorry, it's simply not valid to claim that laziness is the reason.
it's ridiculous to lump unix and windows into the same category. though it really makes an excellently condescending slam!
this is revisionism. for instance, the PM interface was specified by IBM - as a gratuitously different version of existing interfaces. the code was most definitely done by Msft, though (I was there).
basically, IBM committee-designed OS/2 into oblivion, and Msft simply gave up on it, since they had two routes around: windows on the low-end, and NT OS/2 (yep) on the high end. discarding ia32 in favor of risc chips and microkernels was pretty fashionable at the time.
Sun is just as anal-retentive/control-freak/lawyer-driven as Microsoft, only less successful. Both of them are rather pitiful.
the really odd thing about this development is that Sun inexplicably thinks that the enemy of its enemy is yet another enemy. well, I guess it's also mystifying that they think anyone gives a damn about Solaris or their hardware. as if the freeness of SO6 would make anyone pay good money for Sun's crappy hardware.
let's be honest, the states AG's are just slimeball politicos owned by companies who weren't good enough (as evaluated by the market/customer) to compete with msft. make no mistake: their "remedy" is nothing short of nationalizing windows.
this is great for globalization, since ineffectual government avoids doing anything dramatic to multinationals, except the usual extortion/tax.
what's missing? the real goal should be liberty, not democracy. sure, democracy might be a means to liberty, but it's NOT THE GOAL. they're orthogonal - liberty is about policy (principles); democracy is about mechanism.
there's a "meta-politics" that's not being discussed - that's why this is such a fuzzy topic. why is terrorism wrong? what is the real conflict between the West and Taliban-style fundamentalism? the principle of individual liberty - that if you want to live a Wahabi life, you're perfectly free to do so in the West. you just can't coerce someone else into doing it. liberty/non-coercion is what we should be talking about, not democracy.
and this is relevant to the undercurrent of discussion about how the net will effect society in the future. it's obvious that strong crypto, peer-to-peer, net-communities are powerful forces that, in the absence of some kind of apocalypse of talibanhood, will become dominant. they have a sense of historic inevitability. they're also profoundly liberty-based, self-organizing, non-coercive. even anti-authoritarian. and globalizing.
but how can that be? wasn't seattle supposed to be the rise of a non-hierarchical, self-organized political force devoted to overthrowing globalization? there's a contradiction there: absence of hierarchically imposed limits are what permits these anti-globalization people to demonstrate. (and demonstration != democracy!) the anti-globalization freaks are opposed to commerce being the "working fluid" of globalization. it's not the multinationals that they oppose, it's the fact that MN's are based on an international currency market that in effect makes my 8-hours of labor in the West incomparable to 8 hours of labor by someone in the 3rd world. this seems irrational to me, or at least based on principles I don't share (ie, more "from each according to his ability" rather than "to each according to the market price of his ability").
if online/crypto is a globalizing force, it's not necessarily going to cause a redistribution of wealth, or a replacement of property as the measure of wealth. and that's the tip of another iceberg - that some people want ideas to become as ownable as property; not surprising, these "idea hegemonists" are large, Western, multinational corporations...
Linux has been running in it for some time, quite possibly beating *BSD. good talk on this at OLS.
humans are slow. you CANNOT perceive 5ms.
I hope that Linus will look at whether these patches hurt the normal case. "normal" means things like kernel compilation, not just an arbitrary latency measure and dbench (one of the least realistic benchmarks possible!)
there are good reasons to be skeptical of all-out premptiveness: it will unavoidably lower throughput in easy-to-define cases. any intro OS text will talk about optimal scheduling, where 'optimal' requires a definition of throughput or some other metric. preemptive kernels will context switch more, and will probably interfere with the natural 'batching' that happens when a big job runs for a while. think about caches: you never want to switch unless you must. this is not an argument against low-latency! it's an arguement against lowest latency as an absolute; we need to set a target (5ms would be fine imo) and meet it. going beyond such a goal will hurt the normal case.
he's right that the copyright should have been propogated. but this hardly counts as IP-theft: move along, nothing to see, move along...
beowulf isn't just clustering, it's compute-intensive clustering. that means that run-of-the-mill 100bT cards will likley not be enough - think Myri2000, SCI, Quadrics, gigabit.
whitebox machines also tend to be truely horribly configured - you probably don't want UMA video stealing dram bandwidth, for instance. and even for a smallish cluster, it's smart to get better power supplies and fans, simply because the noname stuff has short expected lifespans.
they suck, basically. a 4-way Xeon is sitting
on a miserable 800 MB/s (450 sustained) bus,
so unless your code is totally cache-friendly,
the CPUs will starve. and if your code is so
cache-friendly, you should probably be looking
at different hardware in the first place.
it's easy to dislike Microsoft - after all, they are remarkably successful at offering obviously flawed and limited products. yes, they clearly use network effects to promote their products, which is exactly what any good business does. their actions are quite possibly too agressive, agressive enough to infringe on certain laws. but they are not immoral. the moral aspect of this issue is the willingness of companies like Sun, Oracle and Netscape to resort to politics when they can't compete, when their products are worse than Microsoft's. you may disagree on how the nation's antitrust laws apply to Microsoft, but why should it be appropriate for these failed competitors to get the cops to beat up on Microsoft? it's clearly not just a matter of "enforcing the law" or some such tripe - after all, Intel is free to permute its pinouts in a blatantly anticompetitive manner, or simply refused to lisence them.
if Sun/Oracle/Netscape were serious, they'd be asking for updated or more specific laws. immoral companies routinely use stupid politicians, gullible judges and ridiculous laws to abuse their competitors. routine doesn't make right.
in reality, as opposed to the courts or press, competitors to msft have been stupid and ineffectual. and msft has done a remarkable job of producing products that are eminently affordable, and get the job done for hundeds of millions of users. backwards compatibility is msft's main product, and they do it well.
if msft is broken up, then clearly Intel should, so that the Celeron product line can be uncrippled. and bundling rdram with CPUS! that's obviously abusing their monopoly position - DOJ should require them to support PC2100 immediately!
amusing: takes PC133 sdram, but runs the bus at only 83 or 100 MHz.
it's critical to realize that these boxes, while impressive, are mainframes, not supercomputers. for instance, if you look at the 64-way E10K Stream score, you realize each CPU is seeing a measly 150 MB/s or so! these are clearly machines designed mainly for driving big IO, not running applications that need more than trivial memory bandwidth.
so maybe we've learned some things. progress does mean that the frontier looks different, even that the frontier is gone. in concrete terms, 30 years of Unix has given us wonderfully practical systems; wasn't that the point? after all, this is an applied field, not purely abstract, nor purely artisanal.
MSFT would be smart to shift just certain products north. hell, how about just the free one, on which a 90% tarriff would be uh, $dumb... actually, it would be hillarious if they shifted part of the company (office or windows). Canada would work fine, but then again, India might too.