and once the limits of perception are reached (and probably before they're reached, honestly), perception will just improve:)
So maybe you can render enough polys to make something on a monitor look like a photograph, but can you compute enough video, sound, and AI, to make something thats as real as "the matrix" when you are bypassing your eyes altogether for sensory input ? Think of the staggering computation required to put you in a beleivable 3d world by stimulating the brain directly.
driveline losses for 2wd systems are typically 10-20%, most people just say 15%. if you dont beleive, me, go have your car dyno'd.
furthermore, the transmission/clutch assembly is not where all the losses occur, putting 30hp at the front of the driveshaft would still subject you to the losses of the driveshaft, coupllings, and more importantly the differential. Why do you think performance vehicles have temperature senders for differential fluids ? Because they turn some of your hard earned power into heat (and tiny metal filings:)
even in a perfect system, 30hp to the tires themselves would be a useless vehicle. even motorcycles deliver more power to the road.
this car is interesting because they're actually getting respectable performance out of it. im examining this site to see where they're "cheating". if they could make a 3500 pound electric vehicle that ran under 8 seconds for 0-60, seated 4, and had a 100mile range, then that'd be something. that'd be your average lame sedan. afaik, No one has made that car yet. once you start going to electrics, you need small, light, cheap, flimsy vehicles.
I'll never drive anything that looks/handles/performs like a geo metro just because it runs on batteries. They can't even make laptops run a long time on batteries:)
Humanity will always have conflict. People will always have agression.
Any book that talks about giving people drugs to treat their "Aggressive emotions or tendancies" had better do so in the style of a satirical vonnegut novel. I like my feelings just the way they are, thank you.
Fwiw, I'd much rather juice someone into a puff of red mist in a video game and laugh about how nasty it looked than do it in real life and spend the rest of my days grabbing the bars while i get porked from the rear.
While we're at it, as long as people are at home playing video games, they're not out raping someone i care about. Thats fine by me.
Because as evidenced over and over, judges are woefully stupid w.r.t. technology cases.
I think video conferencing and all that business is silly for court cases, especially if its only for technology cases. But I'm all for pickning some handful of judges and having them take a bunch of CS courses or _something_ so that they are at least qualified to talk intelligently about computing and technology with someone of 8th grade comprehension. once thats happened, i think it should be reasonable to allow any technology/computer crime cases to be moved to those districts.
Sega OUT of the console business, and instead doing what they do best. Making _software_.
Everyone knows that Sega has some of the best franchises ever, and that they're developed in house. Having sega as a developer on your platform means you're guaranteed at least a few good games - none of this playstation2 launch bullshit.
I love my dreamcast (first console i've owned since an 8bit nes) and while on one hand its disappointing to see the DC going away, its good that sega will have more focus on making their titles. Who cares what color the machine is as long as Sonic the hedgehog is still blue ? Plus, with the VF series on playstation hardware, maybe all those button mashing idiots (read: tekken fans:) will have some exposure to an interesting fighting game:)
As an avid console gamer, i think you are full of shit. I've been playing video games all my life and when i read shit like this i just want to kill the dumb
Java is a language. It happenes to be compiled to a VM which has many implementations on many platforms. It also has a huge honkin set of class libraries that make it so you dont have to invent the wheel for each app you write.
.NET is about distributed computing for the masses. The language you choose is irrelevant, the OS providing a.NET service is irrelevant.
Do you not see the fundamental advantage of being able to make _function calls_ across HTTP ?
When VS.NET ships suddenly (apart from the time it takes to get used to the changes from VB6->VB7) your average VB developer can write distributed reusable services. If shes not so hot at writing some peice, her friend the mighty C++ guru can cook her up a few components, which she can use pretty easily. Infact, she can _inherit_ from them if she likes and extend them in new ways.
So when the majority of future MS products are designed against this distributed computing architecture, "things should be pretty cool".
So, can you do distributed computing with java ? Certainly.
Is it easy ? Not so sure...
Is it designed from the ground up for massively distributed scalable services using the public internet as the connecting infrastructure ?
renders your file-system as 3d columns, where the ehight of the column has something to do with the size of the file/directory. You fly around in the fs and can click on things.
It is clearly not the case that all whites naturally assume they are better employees than blacks, just like it is clearly not the case that all whites have access to good education (have you ever been to the south ?), just like it is _clearly_ not the case that all members of your "Racial group" even give a shit about the educational progress or financial stratification of their race.
There is obviously racism in our field, because our field is made of PEOPLE, some of whom happen to be racist, or even have subconscious prejudicial tendancies.
The problem with this lawsuit is that it states that microsoft as a _company_ has a corporate policy of discrimination against blacks. This lawsuit isn't saying "there are some bad apples at MS" or "all the people we had to deal with were racist" - it says that "microsoft as a company is run by a secret good old boy network of black-hating people that turn otherwise unbiased managers into agents of the white devil".
How likely do you think that is ?
As someone who has been through Microsoft interviewer training, at what point in the MS hiring process do i get told about the secret "discriminate against blacks" rule ? It sure wasn't in my training packet. It wasn't anywhere listed in the whole section of stuff about actively but tastefully avoiding any questions or issues regarding ethnicity/religion/gender/preference/whatever. As interviewers there are whole lists of things we're not allowed to talk about and anything that could even be used to inferr financial history or caste or racial background or whatever is strictly taboo.
I've worked at a number of tech firms - there is more racial diversity at MS than anywhere else i've worked, both in general and in management.
Although I don't beleive in the government mandating AA programs and quotas, given a hiring choice where all things were equal, if one candidate clearly had an advantage growing up because of racial and financial history, and the other probably had to work _really_ hard against lots of financial and social obsticales, i'd want to try and hire the less priviledged of the two - the one that had to work harder to cover the same distance. But I like to root for the underdog. Unfortuneately, thanks to all the AA and PC stuff thats been forced on institutions, my understanding is that i am not allowed to take into consideration such issues when making those decisions.
Shame really - the very basis to judge who's had the harder climb, a method to really show minorities and those who are disenfranchised with "playing by the rules" that they _can_ win - pulled right out of the whole hiring process.
And for what its worth, I salute you for sticking with it to get to where you are now - I got glimpses of how the black kids at my HS who were really working hard to make the most of their education were treated by others - sometimes whites - but more often than not, the blacks from their very own neighborhoods that seemed to have given up on the system. It was disheartening all around, to say the least.
It's hard to say wether AA type programs or even optional preference given to minorities will do anything to help rekindle the spirit of the kids that see a tunnel with no light at the end. Part of me thinks that people see through AA and minority preference as a cop out, enforcing the idea that minorities _cant_ get ahead without help.
Whats the right answer ? The government says im not allowed to have a preference one way or the other. What do you say ? What should the hiring practice in the tech industry be ?
The E10k can be configured as a single-image 64 proc SMP machine, or 16 4 proc SMP machines, or several variations inbetween.
Linux is crap for single-image computing with lots of processors. Even Win2k is better (and win2k datacenter runs on the Unisys ES7000, 32 proc Win2k machine!)
Incidentally, the E10000 is a crap architecture for SMP computing - its basically a big ass backplane for 2-or-4-cpu boards and lcoal ram to plug into. But at the heart of things, its a bus connecting crossbar node cards. Not very scalable, because eventually you just have too much bus contention.. A 64-way crossbar would be practically impossible, and having 64 procs on the same cpu/memory bus doesn't work at all. So the E10k is a hybrid : 2 or 4 proc node cards with local ram using essentially the standard sun4u crossbar. Then each of these boards plugs into the "gigaplane" backplane. Non-local memory requests go out over the backplane.
The SGI O2k and 3800 are done right, comparatively. Thats why SGI can ship a 512proc single-image machine. No one else comes close.
There are problems with the SGI approach though, NUMA can be tricky to tune and some argue that if you've got to tune at all you might as well go straight MPP via MPI or PVM - i.e. beowulf, and get more or less infinite scalability, but little/no support from the OS for shared resources.
The cost of data storage isn't the physical capacity - its the management.
You should know that if you've heard of EMC. EMCs are practically self contained black-boxes of "poop data here and dont worry about it, ever". Some (all?) EMC systems phone home when they think there will be a problem. It is not uncommon for the first sign of disk failure in an emc to be the new disk arriving in the mail on the sysadmins desk!
Its not tricky to slap a bunch of drives together and get an assload of capacity. It is tricky to figure out how to keep 23523 18gb disks running if you just have an excel spreadsheet telling you which cabinet each disk sits in.
Ditto with IBM. The coolest thing i've ever seen are the multiple-arm tape storage libraries with the ADSM interfaces infront of them that make data archival and retreival pretty painless.
The key is managing data distribution when you've got an assload of data. This is one project that addresses that, among other things.
The point is scalable storage. RAID isn't scalable. It doesn't scale beyond the number of chips and cables you can stuff into one box. You can get really big boxes. You can get fibre chanel. You can get SANs. But eventually, it all runs out, and you need clusters with a distributed file-system.
This, and many other previous projects (and even a few products) are attempts at that holy grail: the file-system for loosely coupled clusters that gives people useful and familiar semantics, and scales well.
Personally i know i'd love to just be able to add capacity to my entire enterprise by plugging in a few more boxes with 30 gig ide disks in them, and have everything just figure itself out. Bam, i get more cpu power, more disk space, and more app serving capacity. Oh, and better fault tolerance.
Oh, and it all manages itself without ridiculous amounts of sysadmin intervention.
Thats what these groups are going for. RAID doesn't come close. Not even the same topic.
2. How do you index this thing? Centralised or distributed? Who controls it if central?
Distriubted with a series of "central" directory servers would probably be the best bet.
That is not really a great problem - all it needs is a few hours (days) of good thinking,
and a little testing and a good system would be worked out quickly.
Whoa there. Have you tried to work through this before ?:)
People have been trying to do this for _Years_. I spent more than a few days doing a thesis on just this problem - distributed algorithms are not trivial. Especially when they need to be reliable and arbitrarily scalable. This system seems to have good reliability hooks, but as you eluded to, they chose a compromise for decentralization... their update policy has fuzzy semantics and requires centralized control for strict atomicity for single-client updates. They say "it should scale to a few million servers". That would be pretty cool, but they might be wrong, and that means it will wear out before its even fully realized:)
They have a pretty kick ass solution to the "delete" bit though. In short, they dont seem to do it. Also, they keep around old _versions_ of files. For once, you can actually get persistant storage. Updates to a file increment its version number, but they seem to indicate that old versions will stick around and can even be requested, (this is not stated: Similar to CVS/RCS). So you can conceivably make something analagous to
Other filesystems (VMS did this ?) had versioning right in the file-system. I think its a good idea. FS support for solutions to things like DLL hell ? Count me in!
Notice in section 4.2 where they talk about data write policies. This resembles the IRC network quite a bit for writing. I envision a scenario in which ACLs and permissions can be hijacked, _especially_ if they used a distributed consensus algorithm (they mention these algorithms in several places).
Making fully distributed untrusted systems is hard. Making hybrid distributed systems is harder.
Reading section 4.4.3 makes this apparent - they dont have a fully distributed system. If you want the most hardcore possible file-system semantics, a small number of servers with huge bandwidth vote on what to do. So its sort of centralized some of the time. later in 4.4.3 they say if you dont need such strong semantics you dont have to use these tier 1 machines to do the update ordering.
That said, We(tm) need a good anonymous ubiquotous data store. Incidentallly, for all you.NET haters, this is exactly the sort of thing MS has in mind to store your data on. Notice in the conclusion section they compare their work most closely to the "Farsite" project, except theirs was designed with WANs in mind. Farsite is a MS Resaerch project:)
Many of you may not be familiar with TERA, the seattle based super computer company that bought CRAY from SGI and then renamed themselves CRAY for marketing reasons.
Tera's home-brew supercomputers used what they called the TERA MTA - Multi-threaded architecture processors. You could get a 4 proc MTA machine that would significantly outperform much larger super computers.
Essentially the MTA cpu has knowledge of 128 virtual threads of execution inside of it. AFAICT, the point of the MTA design, and apparently of this one, is to minimize the penalty for branches, context switches, etc, wherever possible by putting fine grained execution knowledge in the CPU itself.
Given that superscaling has reached its limit and superpipelining is getting nastier and nastier, this might be a good way to go. Apparently Tera gets great numbers with their MTA stuff.
Carmack has already done a.plan update where he explains the need for 64bit color.
It's not because people see banding with 24bits of discrete color. It's because graphics accelerators accumulate too many range-clamping errors once you start doing multiple passes per pixel.
His proposal, iirc, was 4 16bit floating point values for RGBA, for a total of 64 bits. Going from fixed to floating point solves the range problem but introduces its own problems. First off, you have to start doing ordered adds for for low-order/low-value accuracy. For instance, if i have a few floating point numbers like this:
.00000001
.00000002
24^2342
if i add them in the above order, the "result" will be more correctly represented than if i add them like this:
24^2342
.0000001
.0000002
this is becasued floating points are expressed as mantissa (fixed # of bits) and an exponent (fixed number of bits). with thousands of numbers to add, order of operation actually makes a difference because you lose the accuracy if your mantissa is already 30 digits away from the decimal point.... if you dont beleive me, figure out the smallest representable floating point value (i mean absolute value, not the "most negative"). write a loop that adds it to itself a few billion times, then add a significantly larger number. now do things the other way around and see if the results are the same:)
Also, you may or may not be able to correctly represent a given floating point value with a given number of bits. You get less sort of "compression" artifacts on the color space with floating point colors, but you get floating point errors:)
alot of the people that are badmouthing.net say one of two things:
1) It's MS only
this is blatantly false. obviously if people are deploying applications _today_ using SOAP, and the future medium of all communication is going to be XML over HTTP, it is amazingly interoperable. Microsoft isn't stupid. You may not know this, but microsoft has already said that they realize the operating system is becoming a commodity product and in the near future you wont be able to have a dominant role in the industry if all you ship is an OS. With that in mind they're designing future products to be best of breed solutions in an interoperable environement. Either you'll like their products/tools or you wont.
2) I can do everything VS.NET promises today with open source tools.
Well, you can, sort of. Afterall, the.NET platform is all about interoperability. THe question is, what will the developer experience be like? How easy will it be for you to the equivlanet of
Nice to see someone not spouting crap about VS.NET. Yes, VS.NET Beta 1 shipped over a month ago, in 3 different languages no less.
Hardly vapourware.
As far as when the final product will be released: as soon as possible. The people the see Visual Studio.NET nearly universally agree: they want it _now_.
Alot of people with strong anti-MS bias usually say the same two things
I guess i don't see the necessity of a killer app, or how email or the web qualify.
Email: like sending a letter with no paper, and instantly delivered.
Neat. Sounds like what the internet seems to do to many types of real-world things. Ebooks for instance.
Web: Large distributed encyclopedia/newspaper/product catalog all combined into more or less the same kinda of rectangles on your screen. Anyone can publish here relatively cheaply.
Those are both neat things, but now they've become such a regular part of life that i dont really think of them as "apps". I already think of internet connectivity like i do electrical service. There's no killer app for household current - its just something thats used by nearly everything I enjoy.
So it will be with internet connectivity. I'd really like to be able to read email or post my silly slashdot messages from the couch, on a tv, but conveniently. I can do that now but its a bit of work to setup.
Things like TiVo are actually some of the most exciting penetrations of "smart" devices into the home. TiVo should _really_ use your internet connection to get program schedules, not to mention using your inernet connection to allow you to download movies etc that arent being carried by your cable provider (which will go away soon enough). It couldn't possibly be that much work to add a DiVX player to a box thats already as capable as a tivo machine.
Anyway, i could go on and on. More and more things will use a ubiquotous data network and as that happens i'll be able to spend more and more of my life relaxing and doing whatever the hell i want, be it reading about the real-time snow conditions on a mountain i want to ski down (a co-worker does this daily) or getting little reminders in my car that know that one of the cheapest gas stations carrying the brand and grade of gas I prefer will be en route to where I happen to be going today, so i should stop by and tank up as a matter of convenience (and cost savings).
In-car technology is already doing neat stuff - in this months issue of the Roundel (BMW club magazine) one of the writers talks about the benefits of having integrated hyper-accurate road maps and GPS.. she was able to get effectively a track map of a backroad she had never been on before.. allowing her to extend her senses a little bit further around corners than she might be able to otherwise. It's not so much of a leap to imagine a system that builds on this and lets drivers know that there is stopped traffic or even more importantly road construction or an accident just around the next corner..
The "killer app", if you must, of the future will be the convergance of all the data sources and custom application logic that goes with them into an interoperable framework that lets me do frankly whatever the hell i want to without worrying too much about it. Take a good look at SOAP and it doesn't look like it's so far off.
A strange thing happened to me the other nite. My fiance and I were trying to decide where to go eat. We finally thought of a place, but wanted to call ahead to see if it was still open. I went to yp.yahoo.com as per usual to look up the number.
Problem is, I had forgotten that my DSL had been down for about 20 minutes, and was still down.
What is strange about this is that i felt helpless. I didn't even know where in our apartment the phonebook was. Not to mention the huge distaste that entered my mind in having to guess which part of the stupid fucking book that the place would be listed.
I can usually have the phone number to a place in under 10 seconds via yp.yahoo.com. That includes the time to hit ctrl-N, manually type in yp.yahoo.com, and then manually set focus to the input box for what i want to look up.
Comparatively, using a phonebook seems about as appealing as with a sandblaster.
Last evening I totally forgot that the us presidential candidates were giving their victory/concession speeches. At like 3 am when i remembered, i simply went to cnn.com and was able to read the full text of both speeches. At no point did it even enter my mind to go check C-span or one of the tv news channels to see if i could find out what was said.
I've only lived in the pacific northwest for a few months, so I still don't know where all the different sections of town are geographically, or where some of the best places to go are. The streets here are great and easy to navigate on the east side of the sound, but in seattle proper there are quite a few streets that make no sense at all.
Yet Maps.yahoo.com is accurate about 95% of the time out here. I religiously check it before I decide to go to a new specific place. Even when I take my weekend drives, I get a large overview map of the area to sort of guage what direction I should go in for the kind of mood im in (twisty roads, mountain views, etc).
I hate to sound like a yahoo commercial:), but this access to information has become a routine and important part of my life. I really like the idea of "everyting in one place". Problem is, that place is at home. I can only store a very small cache of what I want to know in my palm IIIe, and i have to manage that manually and ahead of time. Already I want to get it upgraded to 8 megs so i can use the king county map (1.3mb for low res version) from http://www.mapopolis.com
When i have ubiquotous access to everything I want to know from anywhere in the country, we'll be making good progress.
Next, when the interfaces I use all behave in a consistant and intuitive way, no matter where I am or what device I am using - we'll be making good progress.
When ATMs and payphones are a thing of the past, and instead i can walk into something like an imap/cybercash/whatever booth and be confident that i am accessing and manipulating my data securely, that will be a good start.
Once I dont need to use these booths because we have a pervasive wireless cloud that is even more convenient, then we'll be making good progress.
Networks will make all of these things happen, and probably sooner than we all think. They wont be cable TV networks, and they wont be telco networks. The most adaptible and pervasive network in history is the logical choice to build all of these sorts of things on.
The voltages on a phone line in different electrical-states are well documented. The peak voltage is over 100 though (at ring time). There is actually something called "black box" iirc which simulates the voltage condition of an idle line while keeping the connection physically usable for voice transmission - the phone company doesn't bill you for the true length of the call, just the base fee + the amount of time before you switch on the box.
Fwiw, i was working on a handset and was holding the circuit board while the phone rang. It was _quite_ a shock. Furthermore, i was working on smallish box (20 lines) outdoors standing on very wet ground in my boots. Even so, i would occasionally have to take breaks because i'd feel the current starting to freak out my hands.
Slashdot Mirror
Alpina B10 Bi-Turbo (modified BMW E34 5 series)
http://www.bmwm5.com
right...
:)
and once the limits of perception are reached (and probably before they're reached, honestly), perception will just improve
So maybe you can render enough polys to make something on a monitor look like a photograph, but can you compute enough video, sound, and AI, to make something thats as real as "the matrix" when you are bypassing your eyes altogether for sensory input ? Think of the staggering computation required to put you in a beleivable 3d world by stimulating the brain directly.
you have got to be kidding.
:)
:)
driveline losses for 2wd systems are typically 10-20%, most people just say 15%. if you dont beleive, me, go have your car dyno'd.
furthermore, the transmission/clutch assembly is not where all the losses occur, putting 30hp at the front of the driveshaft would still subject you to the losses of the driveshaft, coupllings, and more importantly the differential. Why do you think performance vehicles have temperature senders for differential fluids ? Because they turn some of your hard earned power into heat (and tiny metal filings
even in a perfect system, 30hp to the tires themselves would be a useless vehicle. even motorcycles deliver more power to the road.
this car is interesting because they're actually getting respectable performance out of it. im examining this site to see where they're "cheating". if they could make a 3500 pound electric vehicle that ran under 8 seconds for 0-60, seated 4, and had a 100mile range, then that'd be something. that'd be your average lame sedan. afaik, No one has made that car yet. once you start going to electrics, you need small, light, cheap, flimsy vehicles.
I'll never drive anything that looks/handles/performs like a geo metro just because it runs on batteries. They can't even make laptops run a long time on batteries
Thats the _1000_, not the 100.
:)
Different machines. The 1000 has been out "forever". Seems _you_ needed to do some proofreading
that _Exact_ url wasn't there as recently as 10 hours ago, when i first started seeing stories about it.
There were definitely news sources reporting this one before suns site had hte info available.
Humanity will always have conflict. People will always have agression.
Any book that talks about giving people drugs to treat their "Aggressive emotions or tendancies" had better do so in the style of a satirical vonnegut novel. I like my feelings just the way they are, thank you.
Fwiw, I'd much rather juice someone into a puff of red mist in a video game and laugh about how nasty it looked than do it in real life and spend the rest of my days grabbing the bars while i get porked from the rear.
While we're at it, as long as people are at home playing video games, they're not out raping someone i care about. Thats fine by me.
Because as evidenced over and over, judges are woefully stupid w.r.t. technology cases.
I think video conferencing and all that business is silly for court cases, especially if its only for technology cases. But I'm all for pickning some handful of judges and having them take a bunch of CS courses or _something_ so that they are at least qualified to talk intelligently about computing and technology with someone of 8th grade comprehension. once thats happened, i think it should be reasonable to allow any technology/computer crime cases to be moved to those districts.
IANAL, and so on..
Sega OUT of the console business, and instead doing what they do best. Making _software_.
:) will have some exposure to an interesting fighting game :)
Everyone knows that Sega has some of the best franchises ever, and that they're developed in house. Having sega as a developer on your platform means you're guaranteed at least a few good games - none of this playstation2 launch bullshit.
I love my dreamcast (first console i've owned since an 8bit nes) and while on one hand its disappointing to see the DC going away, its good that sega will have more focus on making their titles. Who cares what color the machine is as long as Sonic the hedgehog is still blue ? Plus, with the VF series on playstation hardware, maybe all those button mashing idiots (read: tekken fans
As an avid console gamer, i think you are full of shit. I've been playing video games all my life and when i read shit like this i just want to kill the dumb
:)
Nevermind.
How did this get +4 insightful ?
.NET service is irrelevant.
.NET
Java is a language. It happenes to be compiled to a VM which has many implementations on many platforms. It also has a huge honkin set of class libraries that make it so you dont have to invent the wheel for each app you write.
.NET is about distributed computing for the masses. The language you choose is irrelevant, the OS providing a
Do you not see the fundamental advantage of being able to make _function calls_ across HTTP ?
When VS.NET ships suddenly (apart from the time it takes to get used to the changes from VB6->VB7) your average VB developer can write distributed reusable services. If shes not so hot at writing some peice, her friend the mighty C++ guru can cook her up a few components, which she can use pretty easily. Infact, she can _inherit_ from them if she likes and extend them in new ways.
So when the majority of future MS products are designed against this distributed computing architecture, "things should be pretty cool".
So, can you do distributed computing with java ? Certainly.
Is it easy ? Not so sure...
Is it designed from the ground up for massively distributed scalable services using the public internet as the connecting infrastructure ?
No.
Enter
actually it was "fsn" on an IRIX box
renders your file-system as 3d columns, where the ehight of the column has something to do with the size of the file/directory. You fly around in the fs and can click on things.
It is clearly not the case that all whites naturally assume they are better employees than blacks, just like it is clearly not the case that all whites have access to good education (have you ever been to the south ?), just like it is _clearly_ not the case that all members of your "Racial group" even give a shit about the educational progress or financial stratification of their race.
There is obviously racism in our field, because our field is made of PEOPLE, some of whom happen to be racist, or even have subconscious prejudicial tendancies.
The problem with this lawsuit is that it states that microsoft as a _company_ has a corporate policy of discrimination against blacks. This lawsuit isn't saying "there are some bad apples at MS" or "all the people we had to deal with were racist" - it says that "microsoft as a company is run by a secret good old boy network of black-hating people that turn otherwise unbiased managers into agents of the white devil".
How likely do you think that is ?
As someone who has been through Microsoft interviewer training, at what point in the MS hiring process do i get told about the secret "discriminate against blacks" rule ? It sure wasn't in my training packet. It wasn't anywhere listed in the whole section of stuff about actively but tastefully avoiding any questions or issues regarding ethnicity/religion/gender/preference/whatever. As interviewers there are whole lists of things we're not allowed to talk about and anything that could even be used to inferr financial history or caste or racial background or whatever is strictly taboo.
I've worked at a number of tech firms - there is more racial diversity at MS than anywhere else i've worked, both in general and in management.
Although I don't beleive in the government mandating AA programs and quotas, given a hiring choice where all things were equal, if one candidate clearly had an advantage growing up because of racial and financial history, and the other probably had to work _really_ hard against lots of financial and social obsticales, i'd want to try and hire the less priviledged of the two - the one that had to work harder to cover the same distance. But I like to root for the underdog. Unfortuneately, thanks to all the AA and PC stuff thats been forced on institutions, my understanding is that i am not allowed to take into consideration such issues when making those decisions.
Shame really - the very basis to judge who's had the harder climb, a method to really show minorities and those who are disenfranchised with "playing by the rules" that they _can_ win - pulled right out of the whole hiring process.
And for what its worth, I salute you for sticking with it to get to where you are now - I got glimpses of how the black kids at my HS who were really working hard to make the most of their education were treated by others - sometimes whites - but more often than not, the blacks from their very own neighborhoods that seemed to have given up on the system. It was disheartening all around, to say the least.
It's hard to say wether AA type programs or even optional preference given to minorities will do anything to help rekindle the spirit of the kids that see a tunnel with no light at the end. Part of me thinks that people see through AA and minority preference as a cop out, enforcing the idea that minorities _cant_ get ahead without help.
Whats the right answer ? The government says im not allowed to have a preference one way or the other. What do you say ? What should the hiring practice in the tech industry be ?
Let me put it bluntly: _WRONG_
Beowulf is a shared-nothing cluster.
The E10k can be configured as a single-image 64 proc SMP machine, or 16 4 proc SMP machines, or several variations inbetween.
Linux is crap for single-image computing with lots of processors. Even Win2k is better (and win2k datacenter runs on the Unisys ES7000, 32 proc Win2k machine!)
Incidentally, the E10000 is a crap architecture for SMP computing - its basically a big ass backplane for 2-or-4-cpu boards and lcoal ram to plug into. But at the heart of things, its a bus connecting crossbar node cards. Not very scalable, because eventually you just have too much bus contention.. A 64-way crossbar would be practically impossible, and having 64 procs on the same cpu/memory bus doesn't work at all. So the E10k is a hybrid : 2 or 4 proc node cards with local ram using essentially the standard sun4u crossbar. Then each of these boards plugs into the "gigaplane" backplane. Non-local memory requests go out over the backplane.
The SGI O2k and 3800 are done right, comparatively. Thats why SGI can ship a 512proc single-image machine. No one else comes close.
There are problems with the SGI approach though, NUMA can be tricky to tune and some argue that if you've got to tune at all you might as well go straight MPP via MPI or PVM - i.e. beowulf, and get more or less infinite scalability, but little/no support from the OS for shared resources.
The cost of data storage isn't the physical capacity - its the management.
You should know that if you've heard of EMC. EMCs are practically self contained black-boxes of "poop data here and dont worry about it, ever". Some (all?) EMC systems phone home when they think there will be a problem. It is not uncommon for the first sign of disk failure in an emc to be the new disk arriving in the mail on the sysadmins desk!
Its not tricky to slap a bunch of drives together and get an assload of capacity. It is tricky to figure out how to keep 23523 18gb disks running if you just have an excel spreadsheet telling you which cabinet each disk sits in.
Ditto with IBM. The coolest thing i've ever seen are the multiple-arm tape storage libraries with the ADSM interfaces infront of them that make data archival and retreival pretty painless.
The key is managing data distribution when you've got an assload of data. This is one project that addresses that, among other things.
The point is scalable storage. RAID isn't scalable. It doesn't scale beyond the number of chips and cables you can stuff into one box. You can get really big boxes. You can get fibre chanel. You can get SANs. But eventually, it all runs out, and you need clusters with a distributed file-system.
This, and many other previous projects (and even a few products) are attempts at that holy grail: the file-system for loosely coupled clusters that gives people useful and familiar semantics, and scales well.
Personally i know i'd love to just be able to add capacity to my entire enterprise by plugging in a few more boxes with 30 gig ide disks in them, and have everything just figure itself out. Bam, i get more cpu power, more disk space, and more app serving capacity. Oh, and better fault tolerance.
Oh, and it all manages itself without ridiculous amounts of sysadmin intervention.
Thats what these groups are going for. RAID doesn't come close. Not even the same topic.
2. How do you index this thing? Centralised or distributed? Who controls it if central?
Distriubted with a series of "central" directory servers would probably be the best bet. That is not really a great problem - all it needs is a few hours (days) of good thinking, and a little testing and a good system would be worked out quickly.
Whoa there. Have you tried to work through this before ? :)
People have been trying to do this for _Years_. I spent more than a few days doing a thesis on just this problem - distributed algorithms are not trivial. Especially when they need to be reliable and arbitrarily scalable. This system seems to have good reliability hooks, but as you eluded to, they chose a compromise for decentralization... their update policy has fuzzy semantics and requires centralized control for strict atomicity for single-client updates. They say "it should scale to a few million servers". That would be pretty cool, but they might be wrong, and that means it will wear out before its even fully realized :)
They have a pretty kick ass solution to the "delete" bit though. In short, they dont seem to do it. Also, they keep around old _versions_ of files. For once, you can actually get persistant storage. Updates to a file increment its version number, but they seem to indicate that old versions will stick around and can even be requested, (this is not stated: Similar to CVS/RCS). So you can conceivably make something analagous to
http://site.made.in.1987.com/index.html;version=1. 1
Other filesystems (VMS did this ?) had versioning right in the file-system. I think its a good idea. FS support for solutions to things like DLL hell ? Count me in!
Notice in section 4.2 where they talk about data write policies. This resembles the IRC network quite a bit for writing. I envision a scenario in which ACLs and permissions can be hijacked, _especially_ if they used a distributed consensus algorithm (they mention these algorithms in several places).
.NET haters, this is exactly the sort of thing MS has in mind to store your data on. Notice in the conclusion section they compare their work most closely to the "Farsite" project, except theirs was designed with WANs in mind. Farsite is a MS Resaerch project :)
Making fully distributed untrusted systems is hard. Making hybrid distributed systems is harder.
Reading section 4.4.3 makes this apparent - they dont have a fully distributed system. If you want the most hardcore possible file-system semantics, a small number of servers with huge bandwidth vote on what to do. So its sort of centralized some of the time. later in 4.4.3 they say if you dont need such strong semantics you dont have to use these tier 1 machines to do the update ordering.
That said, We(tm) need a good anonymous ubiquotous data store. Incidentallly, for all you
...For all we know Win2k could be a wrapper around the Linux kernel...
-23235: Flamebait ?
Oh but wait.. heres a step I secretly captured from a Win2k build log!! Lets see what is revealed!!
gcc -DUSE_REAL_SMP \
-DWORKING_MULTI_THREADING \
-DACTUALLY_SCALE_TO_MULTIPLE_CPUS \
-DBASIC_NON_BRAINDEAD_ACLS \
-DPOST_1975_SECURITY_ARCHITECTURE \
src/arch/x86/hal.cpp \
-o '\\w2kbuild20\dailydrops\hal.o'
Man... i wonder if i can get any of these options to work in the normal linux builds ?
Many of you may not be familiar with TERA, the seattle based super computer company that bought CRAY from SGI and then renamed themselves CRAY for marketing reasons.
Tera's home-brew supercomputers used what they called the TERA MTA - Multi-threaded architecture processors. You could get a 4 proc MTA machine that would significantly outperform much larger super computers.
Essentially the MTA cpu has knowledge of 128 virtual threads of execution inside of it. AFAICT, the point of the MTA design, and apparently of this one, is to minimize the penalty for branches, context switches, etc, wherever possible by putting fine grained execution knowledge in the CPU itself.
Given that superscaling has reached its limit and superpipelining is getting nastier and nastier, this might be a good way to go. Apparently Tera gets great numbers with their MTA stuff.
Carmack has already done a .plan update where he explains the need for 64bit color.
:)
:)
It's not because people see banding with 24bits of discrete color. It's because graphics accelerators accumulate too many range-clamping errors once you start doing multiple passes per pixel.
His proposal, iirc, was 4 16bit floating point values for RGBA, for a total of 64 bits. Going from fixed to floating point solves the range problem but introduces its own problems. First off, you have to start doing ordered adds for for low-order/low-value accuracy. For instance, if i have a few floating point numbers like this:
.00000001
.00000002
24^2342
if i add them in the above order, the "result" will be more correctly represented than if i add them like this:
24^2342
.0000001
.0000002
this is becasued floating points are expressed as mantissa (fixed # of bits) and an exponent (fixed number of bits). with thousands of numbers to add, order of operation actually makes a difference because you lose the accuracy if your mantissa is already 30 digits away from the decimal point.... if you dont beleive me, figure out the smallest representable floating point value (i mean absolute value, not the "most negative"). write a loop that adds it to itself a few billion times, then add a significantly larger number. now do things the other way around and see if the results are the same
Also, you may or may not be able to correctly represent a given floating point value with a given number of bits. You get less sort of "compression" artifacts on the color space with floating point colors, but you get floating point errors
Fuck. hitting tab + enter is not a good idea :)
.net say one of two things:
.NET platform is all about interoperability. THe question is, what will the developer experience be like? How easy will it be for you to the equivlanet of
l ")
:)
To continue where i left off,
alot of the people that are badmouthing
1) It's MS only
this is blatantly false. obviously if people are deploying applications _today_ using SOAP, and the future medium of all communication is going to be XML over HTTP, it is amazingly interoperable. Microsoft isn't stupid. You may not know this, but microsoft has already said that they realize the operating system is becoming a commodity product and in the near future you wont be able to have a dominant role in the industry if all you ship is an OS. With that in mind they're designing future products to be best of breed solutions in an interoperable environement. Either you'll like their products/tools or you wont.
2) I can do everything VS.NET promises today with open source tools.
Well, you can, sort of. Afterall, the
dlopen("http://somesitehere.com/greatLibrary.xm
and have it "just work" ?
Don't make any statements about what VS.NET can and can't offer with complete confidence until you've sat down and used it.
Nice to see someone not spouting crap about VS.NET. Yes, VS.NET Beta 1 shipped over a month ago, in 3 different languages no less.
Hardly vapourware.
As far as when the final product will be released: as soon as possible. The people the see Visual Studio.NET nearly universally agree: they want it _now_.
Alot of people with strong anti-MS bias usually say the same two things
I guess i don't see the necessity of a killer app, or how email or the web qualify.
Email: like sending a letter with no paper, and instantly delivered.
Neat. Sounds like what the internet seems to do to many types of real-world things. Ebooks for instance.
Web: Large distributed encyclopedia/newspaper/product catalog all combined into more or less the same kinda of rectangles on your screen. Anyone can publish here relatively cheaply.
Those are both neat things, but now they've become such a regular part of life that i dont really think of them as "apps". I already think of internet connectivity like i do electrical service. There's no killer app for household current - its just something thats used by nearly everything I enjoy.
So it will be with internet connectivity. I'd really like to be able to read email or post my silly slashdot messages from the couch, on a tv, but conveniently. I can do that now but its a bit of work to setup.
Things like TiVo are actually some of the most exciting penetrations of "smart" devices into the home. TiVo should _really_ use your internet connection to get program schedules, not to mention using your inernet connection to allow you to download movies etc that arent being carried by your cable provider (which will go away soon enough). It couldn't possibly be that much work to add a DiVX player to a box thats already as capable as a tivo machine.
Anyway, i could go on and on. More and more things will use a ubiquotous data network and as that happens i'll be able to spend more and more of my life relaxing and doing whatever the hell i want, be it reading about the real-time snow conditions on a mountain i want to ski down (a co-worker does this daily) or getting little reminders in my car that know that one of the cheapest gas stations carrying the brand and grade of gas I prefer will be en route to where I happen to be going today, so i should stop by and tank up as a matter of convenience (and cost savings).
In-car technology is already doing neat stuff - in this months issue of the Roundel (BMW club magazine) one of the writers talks about the benefits of having integrated hyper-accurate road maps and GPS.. she was able to get effectively a track map of a backroad she had never been on before.. allowing her to extend her senses a little bit further around corners than she might be able to otherwise. It's not so much of a leap to imagine a system that builds on this and lets drivers know that there is stopped traffic or even more importantly road construction or an accident just around the next corner..
The "killer app", if you must, of the future will be the convergance of all the data sources and custom application logic that goes with them into an interoperable framework that lets me do frankly whatever the hell i want to without worrying too much about it. Take a good look at SOAP and it doesn't look like it's so far off.
A strange thing happened to me the other nite. My fiance and I were trying to decide where to go eat. We finally thought of a place, but wanted to call ahead to see if it was still open. I went to yp.yahoo.com as per usual to look up the number.
:), but this access to information has become a routine and important part of my life. I really like the idea of "everyting in one place". Problem is, that place is at home. I can only store a very small cache of what I want to know in my palm IIIe, and i have to manage that manually and ahead of time. Already I want to get it upgraded to 8 megs so i can use the king county map (1.3mb for low res version) from http://www.mapopolis.com
;)
Problem is, I had forgotten that my DSL had been down for about 20 minutes, and was still down.
What is strange about this is that i felt helpless. I didn't even know where in our apartment the phonebook was. Not to mention the huge distaste that entered my mind in having to guess which part of the stupid fucking book that the place would be listed.
I can usually have the phone number to a place in under 10 seconds via yp.yahoo.com. That includes the time to hit ctrl-N, manually type in yp.yahoo.com, and then manually set focus to the input box for what i want to look up.
Comparatively, using a phonebook seems about as appealing as with a sandblaster.
Last evening I totally forgot that the us presidential candidates were giving their victory/concession speeches. At like 3 am when i remembered, i simply went to cnn.com and was able to read the full text of both speeches. At no point did it even enter my mind to go check C-span or one of the tv news channels to see if i could find out what was said.
I've only lived in the pacific northwest for a few months, so I still don't know where all the different sections of town are geographically, or where some of the best places to go are. The streets here are great and easy to navigate on the east side of the sound, but in seattle proper there are quite a few streets that make no sense at all.
Yet Maps.yahoo.com is accurate about 95% of the time out here. I religiously check it before I decide to go to a new specific place. Even when I take my weekend drives, I get a large overview map of the area to sort of guage what direction I should go in for the kind of mood im in (twisty roads, mountain views, etc).
I hate to sound like a yahoo commercial
When i have ubiquotous access to everything I want to know from anywhere in the country, we'll be making good progress.
Next, when the interfaces I use all behave in a consistant and intuitive way, no matter where I am or what device I am using - we'll be making good progress.
When ATMs and payphones are a thing of the past, and instead i can walk into something like an imap/cybercash/whatever booth and be confident that i am accessing and manipulating my data securely, that will be a good start.
Once I dont need to use these booths because we have a pervasive wireless cloud that is even more convenient, then we'll be making good progress.
Networks will make all of these things happen, and probably sooner than we all think. They wont be cable TV networks, and they wont be telco networks. The most adaptible and pervasive network in history is the logical choice to build all of these sorts of things on.
There's no way the internet has peaked
The voltages on a phone line in different electrical-states are well documented. The peak voltage is over 100 though (at ring time). There is actually something called "black box" iirc which simulates the voltage condition of an idle line while keeping the connection physically usable for voice transmission - the phone company doesn't bill you for the true length of the call, just the base fee + the amount of time before you switch on the box.
Fwiw, i was working on a handset and was holding the circuit board while the phone rang. It was _quite_ a shock. Furthermore, i was working on smallish box (20 lines) outdoors standing on very wet ground in my boots. Even so, i would occasionally have to take breaks because i'd feel the current starting to freak out my hands.