Q) Which came first, the multithreaded chicken or the multithreaded egg? A) They came at the same time, but the multithreaded chicken terminated first.
Q) Which came first, the multithreaded chicken or the multithreaded egg? A) Neither; mt egg could not acquire chicken-lock from mt chicken. mt chicken could not acquire egg-lock from mt egg.
Q) Which came first, the multithreaded chicken or the multithreaded egg? A) Multithreaded egg, but it overwrote its DNA while still in use and became mt turkey.
Theoretical underpinnings of a security model can also include things like having a Chinese wall of syscalls be the *only* way a process can interact with other processes and the kernel. It includes things like having shared libraries only share data between threads of a single process instead of between all threads of all processes. It includes things like that at some level which can't be bypassed having *one* way to perform some action rather than Way, WayEx, AnotherWayEx, ad nauseum that do not share a common implementation. It includes things like "one application cannot directly send events to another".
You might call these practical or philosophic underpinnings, but these are not a question of for instance 'how good the code is' but rather of 'why did they make it this way'. There are lots of specific syscalls that could be added that would improve performance, for example a "folder listing with stat" combined call. But these are not part of unix because in large part the theoretical security model says the fewer ways to do something the more secure it is.
The problem is that there are 'applications' and there are 'system tools' and in linux these are the same. People want to install or use dozens of applications, but there are thousands of system tools (netcat for instance) that might be on a system. In linux these are one and the same, but they are not the same to the user.
It wouldn't even take that much more space except that the linux toolchain is really, really bad at making anything stand-alone. Compile a stand-alone "hello world" with GNU libc and it's 700k+. The way libraries are stored in ELF makes it almost impossible to remove functions at link time that are not used; if you link to a library you may have to include the whole thing instead of only the 20% reachable from what your code uses. The build system is at fault too, since often compiling in a single step instead of lots of individual.o object files saves significant space.
Basically what I am saying is that there has been virtually no effort at all in linux toward reducing the size of libraries and binaries, so just because a stand-alone program is really large now doesn't mean it would have to be so large. Besides, if the user has 100 applications each taking 10 Mb more from being self-contained, that's only 1gb. That's not a lot of space these days.
Constitution: "by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries"
Patents should be held by the people that invented it and non-transferable. To be "exclusive", no contract should be enforceable placing restrictions on the inventors re: their patent -- even for 'works for hire'. A company could still enter into an agreement with the inventors to have use of the technology and even to sublicense it to others, but could not prevent the inventor from also licensing it to anybody else.
This gives the actual people who invent the technology the power to cap the price. They don't have an incentive to give it away for free, and at the same time a holding company cannot extort much money from people because the inventor can always choose to make a quick buck by licensing to a competitor.
Technology is never "invented" by a corporation. Corporations do not think or invent, people do. The people should be guaranteed the benefit from their inventions.
And like everything else Microsoft copied from Apple, everybody on the planet knows "Reality Distortion Field: Vista Edition" sucks in comparison except for them; they had it pointed backwards and accidentally convinced themselves instead of everybody else.
Pity the researchers weren't able to create Maxwell's lesser-known "Angel", a device that -- using no power at all -- sits on a barrier and sorts molecules based on their goodness. All matter composed of "charm", "up" and "top" quarks collect on one side, whereas matter composed of the hateful 'strange', 'down', and 'bottom' quarks collect on the other.
This would totally change the world in the short term by finally providing a means to mass-produce holy water, and eventually even purifying the entire world of 'evil' particles (ie collect all the hateful particles together, send them up on the 'space elevator to heaven' and launch into the void).
Well first of all, corporations are only accountable to their shareholders, not to anyone else.
Actually this isn't the problem with corporations today. The problem is that employees are not responsible for the companies actions. Instead, corporations are now "people" who are responsible for the actions that the employees take. There's an obvious disconnect there; the "person" responsible is not the person actually doing the crime.
Corporations typically have a hierarchical structure just so that somebody can be held accountable. If a corporation does something illegal then the person behind that decision needs to be tried for it, not the corporation. If the company hires somebody to do something illegal then the person authorizing that is responsible and the people who knew about it are accessories and the people who found out later and did nothing are accessories after the fact.
This is the real problem today. For example, Microsoft gets convicted of criminal restraint of trade and there are absolutely no personal consequences for the people authorizing it and perpetrating it. There are plenty of people in MS who knew of this and would not have allowed it to happen if their own butt was on the line.
At some point you have to say enough and accept the risk. After all if somebody really wanted to see your vote that badly surely they could just rig a tiny camera watching you vote. I guess the dutch test for these at each polling place? If not it sounds like overreacting to me. They could also use use terahertz frequencies to see through walls and actually watch your vote even on paper. Maybe voting booths should be submerged in water to prevent this??
It's not the paper voting that's the problem, it's the paper counting. In the last election (for senate) there were IIRC 9 different measures on my ballot. Sometimes there are only a couple like in a primary, other times there could be a dozen or more. Also, some are yes/no and some are multiple choice + write in.
Keeping a count of all these without some form of help is pretty annoying. I guess you could have stations, where each one counts only a particular ballot measure. That would probably be the most efficient and least error prone. It's still going to take say O(10x) longer than a Canadian election.
Open source voting system
on
Who won?
·
· Score: 4, Informative
A simple, safe, completely OSS voting system can be made with only say tens or hundreds of hours of work. The key is to make it completely secure by only requiring trust in the ballot box, which is not electronic -- everything else is directly observable by the poll workers, observers, or voter. This lets you leverage any technology out there.
Voting machine:
1. Setup linux distro with apache, tomcat, whatever 2. Install ballot web app 3. Setup CUPS printer 4. Setup firefox for kiosk mode, home page is voting app
Ballots print like this, one measure per line:
PRESIDENT: AL GORE
SENATE: JAMES WEBB
STEM-CELL: YES
During the election, voters take their printout and drop it into the ballot box. After the election these are counted individually at each polling place using a counting machine.
Counting machine:
1. Setup linux distro 2. Install ballot counter program 3. Run ballots through OCR software 4. Update counters (in realtime as scanned)
For the counting program, all it needs to do is keep a count of unique lines on the ballots as returned by the OCR. It should include a simple display showing the most frequent lines and their count (sorted by count) along with the last vote scanned. This way it doesn't need to know anything about the election in order to count it.
For the voting machine you can add fancy CSS styles, javascript to prevent accidental undervoting, screen readers, on-screen keyboard, etc. To polish the system you will want to have some specific printer hardware so the votes print on something smaller than a sheet per vote.
I would say the fact that a bridge engineer can give a realistic estimate on how long it will take whereas the software engineer cannot, or does not, indicates that the bridge is easy in comparison.
Just for example, ask that same bridge engineer how long it will take to make the entire bridge out of a single piece of plastic. Now suddenly it's a much more difficult problem. How do you support the form as it is molding? How much plastic do you need to for integrity? How do you cast something that large at once? The answer will start with "uhh...". Now hold him to the estimate and see what kind of bridge you get. This is more what software development is actually like.
What you are missing is that "control" here is only a shorthand for "independence". No person can really have control over a computer anyway because it is a magic box that operates on the atomic scale, but one can possess that magic box and operate it.
If you have your data on your computer then the only thing you are relying on is power and the continued operation of your computer. You are independent because, with a generator, you have in your possession the only things you need. On the other hand, if google has your data you are also relying on their infrastructure, your internet, the intervening internet, your credit, their continued operation, their goodwill, and so on. In a catastrophe or a terrorism you could easily be without your data for weeks or months if it's stored on the web.
Put it this way, if there is a world-society-ending event I'd much rather have my laptop with a local copy of Wikipedia on disk / hddvd where the only thing I need to find is power rather than have it "on the web" in some datacenter in FL where I need 2+ megawatts just to access it.
No body has pointed this out yet, that the difference is 64-bit. Like 2d graphics acceleration, 32-bit will be the backwater poorly-optimized part of new processors. Yeah, you will be able to run your pirate copy of XP on new computers but over time it will get less and less efficient to do so. This transition basically solves Microsoft's problem of XP being 'good enough', so there will actually be a faster transition to Vista than say to 95->98, 98->ME, 95->NT, NT->2k, 2k->2k3. Only 3.1->95 will have been a faster transition IMO.
Sometimes in software like Samba the best code is written when developers don't know the exact specifications. If you say "make it share files and handle these special cases in Win2k, these ones in WinXP, these ones in Vista" then you are likely to get a solution that can only handle those bugs. But if you say "make it share files and oh yeah there are some special cases here's an example of some" then developers make the system with more flexibility.
Programmers don't want to get stuck on some obscure file locking issue for instance and realize they have to rewrite 50k of code just to handle it. So the less they know about the gotchas the more error checking, hooks, and gestalt they put in. This gets really ugly though when there isn't a well-defined goal (ie 'filesharing with windows') because coders go crazy making it overly generic and flexible.
Yes it is. All the other captains had 'hard' sounds in their name... KIRk, PIcard, KAtherine. Those are names you can say with power befitting a captain.
Just imaging Khan saying "I've done far worse than kill you blah blah blah I shall leave you as you left me, as you left her: marooned for all eternity in the center of a dead planet, buried alive. Buried alive.". And Kirk replying "Alex! Alex!!! Alex!!!!!!!!" (cue pullback shot from outside asteroid). Sounds lame doesn't it. Chase would be even lamer, if that's possible and it is.
Re:Is it any faster for client-side apps?
on
Java SE 6 Released
·
· Score: 1
they are double-buffered by default, and as a consequence they redraw completely before displaying If you read what I wrote and use just a little imagination you should figure out that non-double-buffered apps are 'fast' because they don't redraw completely. When you opaque-resize an app many times the interior components in non-double-buffered systems are not even redrawn at all until the resize is done so only the top-level structural elements get drawn. This is pretty much impossible with double-buffering, but gives users the feeling that the redraw is much quicker than it actually is.
The only performance difference to unbuffered drawing comes from increased memory usage and the displaying of that "image". This is true only if the graphics operations can be accelerated equally well on the buffer image. Many of Java's primitives were implemented in software because the hardware did not accelerate them, and in general images in the computer's main memory are not accelerated by most graphics cards. Note that I mentioned the unnecessarily complex graphics in the original post.
The useless clearing you talk about can be resolved by setting component opaqueness to correct value. For a while doing that only removed *one* of the background clears (I implemented a look and feel btw). But who cares? The point is that unless you see a component redrawing you as a component developer are not likely to clue in that you are wasting resources drawing it. The fact of the background being cleared is but one small example to illustrate the point.
Probably a joke but shouldn't be. The kernel developers could make it possible to accelerate the JVM/CLR by giving faster access to the actual hardware pages. What I mean is, the JVM has a 'scratch' area where recently created objects are allocated and then after garbage collecting this area the leftover objects are moved out. They do this because the vast majority of objects last only a tiny amount of time. So to be able to do a GC *only* of this scratch area the JVM actually replaces reference assignment with code that ads to a big list of all objects that took references to objects in this scratch area. Even though this is obviously slow it means the JVM doesn't have to garbage collect the entire contents of memory to make sure nothing has a ref to a new object.
If the JVM could get access to the hardware's dirty page bit that says if a page has been modified since last checked then the JVM could do direct reference assignment and then when garbage collecting only search the modified pages for references into the 'scratch' area. I expect this would be many times faster than the pointer write barrier used now.
Maybe a system call could take a mmap region and return a bitmask of page dirty flags? I think in any case there are plenty of things the kernel developers could do to make software virtual machines better if they tried. I think they just don't care to since that world is alien to them.
Re:Is it any faster for client-side apps?
on
Java SE 6 Released
·
· Score: 5, Interesting
Java GUIs have traditionally been 'slow' because they are double-buffered by default, and as a consequence they redraw completely before displaying. Hardware simply wasn't fast enough to do this gracefully -- it's only recently that most gnome/kde applications and some xp ones are expected to be double-buffered.
The double-buffering also lead to lots of inefficient widget redrawing, like for a while each widget was cleared with the bg color before being redrawn even if it then say put an image say over its whole area.
The other major slowdown was because Java's graphics were much more advanced than necessary, for example lines of width != 1 with end and joint caps, antialiasing, clipping regions (instead of boxes), custom renderers, etc. This made it difficult to integrate with the simple hardware acceleration at the time. Native apps had jaggy lines and solid colors as the main features.
1. Setup linux distro with apache, tomcat, whatever 2. Install ballot web app 3. Install ballot CUPS printer filter 4. Setup firefox for kiosk mode
Counting machine:
1. Setup linux distro with ballot_counter.py 2. Attach scanner 3. Run ballots through OCR software 4. Update counters (in realtime as scanned)
Ballots print like this, one measure per line:
PRESIDENT: AL GORE
SENATE: JAMES WEBB
STEM-CELL: YES
Also the tally program just keeps track of unique lines returned from the OCR and puts this count next to the on-screen list (sorted by count). Thus it doesn't need to know anything about the election. The top N lines on the screen as votes are scanned (after polls close) will automatically show the votes from the ballot followed by the most popular write-ins (ie Mickey Mouse).
You can add fancy CSS styles, javascript to prevent accidental undervoting, screen readers, on-screen keyboard, etc. There... a complete, secure, reliable, OSS, and cheap voting system for a day's work.
Slashdot Mirror
Grandparent poster just called and he wants his joke back.
Q) Which came first, the multithreaded chicken or the multithreaded egg?
A) They came at the same time, but the multithreaded chicken terminated first.
Q) Which came first, the multithreaded chicken or the multithreaded egg?
A) Neither; mt egg could not acquire chicken-lock from mt chicken. mt chicken could not acquire egg-lock from mt egg.
Q) Which came first, the multithreaded chicken or the multithreaded egg?
A) Multithreaded egg, but it overwrote its DNA while still in use and became mt turkey.
Theoretical underpinnings of a security model can also include things like having a Chinese wall of syscalls be the *only* way a process can interact with other processes and the kernel. It includes things like having shared libraries only share data between threads of a single process instead of between all threads of all processes. It includes things like that at some level which can't be bypassed having *one* way to perform some action rather than Way, WayEx, AnotherWayEx, ad nauseum that do not share a common implementation. It includes things like "one application cannot directly send events to another".
You might call these practical or philosophic underpinnings, but these are not a question of for instance 'how good the code is' but rather of 'why did they make it this way'. There are lots of specific syscalls that could be added that would improve performance, for example a "folder listing with stat" combined call. But these are not part of unix because in large part the theoretical security model says the fewer ways to do something the more secure it is.
The problem is that there are 'applications' and there are 'system tools' and in linux these are the same. People want to install or use dozens of applications, but there are thousands of system tools (netcat for instance) that might be on a system. In linux these are one and the same, but they are not the same to the user.
.o object files saves significant space.
It wouldn't even take that much more space except that the linux toolchain is really, really bad at making anything stand-alone. Compile a stand-alone "hello world" with GNU libc and it's 700k+. The way libraries are stored in ELF makes it almost impossible to remove functions at link time that are not used; if you link to a library you may have to include the whole thing instead of only the 20% reachable from what your code uses. The build system is at fault too, since often compiling in a single step instead of lots of individual
Basically what I am saying is that there has been virtually no effort at all in linux toward reducing the size of libraries and binaries, so just because a stand-alone program is really large now doesn't mean it would have to be so large. Besides, if the user has 100 applications each taking 10 Mb more from being self-contained, that's only 1gb. That's not a lot of space these days.
Constitution: "by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries"
Patents should be held by the people that invented it and non-transferable. To be "exclusive", no contract should be enforceable placing restrictions on the inventors re: their patent -- even for 'works for hire'. A company could still enter into an agreement with the inventors to have use of the technology and even to sublicense it to others, but could not prevent the inventor from also licensing it to anybody else.
This gives the actual people who invent the technology the power to cap the price. They don't have an incentive to give it away for free, and at the same time a holding company cannot extort much money from people because the inventor can always choose to make a quick buck by licensing to a competitor.
Technology is never "invented" by a corporation. Corporations do not think or invent, people do. The people should be guaranteed the benefit from their inventions.
It's not hand drawn. They obviously used dot from graphviz. You can't mistake that layout once you've seen it.
And like everything else Microsoft copied from Apple, everybody on the planet knows "Reality Distortion Field: Vista Edition" sucks in comparison except for them; they had it pointed backwards and accidentally convinced themselves instead of everybody else.
Pity the researchers weren't able to create Maxwell's lesser-known "Angel", a device that -- using no power at all -- sits on a barrier and sorts molecules based on their goodness. All matter composed of "charm", "up" and "top" quarks collect on one side, whereas matter composed of the hateful 'strange', 'down', and 'bottom' quarks collect on the other.
This would totally change the world in the short term by finally providing a means to mass-produce holy water, and eventually even purifying the entire world of 'evil' particles (ie collect all the hateful particles together, send them up on the 'space elevator to heaven' and launch into the void).
Next is belief in a higher power, say Hakkar the Soulflayer or C'Thun.
Well first of all, corporations are only accountable to their shareholders, not to anyone else.
Actually this isn't the problem with corporations today. The problem is that employees are not responsible for the companies actions. Instead, corporations are now "people" who are responsible for the actions that the employees take. There's an obvious disconnect there; the "person" responsible is not the person actually doing the crime.
Corporations typically have a hierarchical structure just so that somebody can be held accountable. If a corporation does something illegal then the person behind that decision needs to be tried for it, not the corporation. If the company hires somebody to do something illegal then the person authorizing that is responsible and the people who knew about it are accessories and the people who found out later and did nothing are accessories after the fact.
This is the real problem today. For example, Microsoft gets convicted of criminal restraint of trade and there are absolutely no personal consequences for the people authorizing it and perpetrating it. There are plenty of people in MS who knew of this and would not have allowed it to happen if their own butt was on the line.
It's more like linear algebra... you have a series of equations based on the random-ish mirror patterns, so for a simple 2x2 grid of pixels (mirrors):
...
1*x0y0 + 0*x1y0 + 1*x0y1 + 0*x1y1 = sample1
1*x0y0 + 1*x1y0 + 0*x0y1 + 1*x1y1 = sample2
0*x0y0 + 1*x1y0 + 0*x0y1 + 0*x1y1 = sample3
Then you basically solve it for the pixels. So think interpolating the entire image at once as a single value.
At some point you have to say enough and accept the risk. After all if somebody really wanted to see your vote that badly surely they could just rig a tiny camera watching you vote. I guess the dutch test for these at each polling place? If not it sounds like overreacting to me. They could also use use terahertz frequencies to see through walls and actually watch your vote even on paper. Maybe voting booths should be submerged in water to prevent this??
It's not the paper voting that's the problem, it's the paper counting. In the last election (for senate) there were IIRC 9 different measures on my ballot. Sometimes there are only a couple like in a primary, other times there could be a dozen or more. Also, some are yes/no and some are multiple choice + write in.
Keeping a count of all these without some form of help is pretty annoying. I guess you could have stations, where each one counts only a particular ballot measure. That would probably be the most efficient and least error prone. It's still going to take say O(10x) longer than a Canadian election.
A simple, safe, completely OSS voting system can be made with only say tens or hundreds of hours of work. The key is to make it completely secure by only requiring trust in the ballot box, which is not electronic -- everything else is directly observable by the poll workers, observers, or voter. This lets you leverage any technology out there.
Voting machine:
1. Setup linux distro with apache, tomcat, whatever
2. Install ballot web app
3. Setup CUPS printer
4. Setup firefox for kiosk mode, home page is voting app
Ballots print like this, one measure per line:
PRESIDENT: AL GORE
SENATE: JAMES WEBB
STEM-CELL: YES
During the election, voters take their printout and drop it into the ballot box. After the election these are counted individually at each polling place using a counting machine.
Counting machine:
1. Setup linux distro
2. Install ballot counter program
3. Run ballots through OCR software
4. Update counters (in realtime as scanned)
For the counting program, all it needs to do is keep a count of unique lines on the ballots as returned by the OCR. It should include a simple display showing the most frequent lines and their count (sorted by count) along with the last vote scanned. This way it doesn't need to know anything about the election in order to count it.
For the voting machine you can add fancy CSS styles, javascript to prevent accidental undervoting, screen readers, on-screen keyboard, etc. To polish the system you will want to have some specific printer hardware so the votes print on something smaller than a sheet per vote.
I would say the fact that a bridge engineer can give a realistic estimate on how long it will take whereas the software engineer cannot, or does not, indicates that the bridge is easy in comparison.
Just for example, ask that same bridge engineer how long it will take to make the entire bridge out of a single piece of plastic. Now suddenly it's a much more difficult problem. How do you support the form as it is molding? How much plastic do you need to for integrity? How do you cast something that large at once? The answer will start with "uhh...". Now hold him to the estimate and see what kind of bridge you get. This is more what software development is actually like.
What you are missing is that "control" here is only a shorthand for "independence". No person can really have control over a computer anyway because it is a magic box that operates on the atomic scale, but one can possess that magic box and operate it.
If you have your data on your computer then the only thing you are relying on is power and the continued operation of your computer. You are independent because, with a generator, you have in your possession the only things you need. On the other hand, if google has your data you are also relying on their infrastructure, your internet, the intervening internet, your credit, their continued operation, their goodwill, and so on. In a catastrophe or a terrorism you could easily be without your data for weeks or months if it's stored on the web.
Put it this way, if there is a world-society-ending event I'd much rather have my laptop with a local copy of Wikipedia on disk / hddvd where the only thing I need to find is power rather than have it "on the web" in some datacenter in FL where I need 2+ megawatts just to access it.
No body has pointed this out yet, that the difference is 64-bit. Like 2d graphics acceleration, 32-bit will be the backwater poorly-optimized part of new processors. Yeah, you will be able to run your pirate copy of XP on new computers but over time it will get less and less efficient to do so. This transition basically solves Microsoft's problem of XP being 'good enough', so there will actually be a faster transition to Vista than say to 95->98, 98->ME, 95->NT, NT->2k, 2k->2k3. Only 3.1->95 will have been a faster transition IMO.
Sometimes in software like Samba the best code is written when developers don't know the exact specifications. If you say "make it share files and handle these special cases in Win2k, these ones in WinXP, these ones in Vista" then you are likely to get a solution that can only handle those bugs. But if you say "make it share files and oh yeah there are some special cases here's an example of some" then developers make the system with more flexibility.
Programmers don't want to get stuck on some obscure file locking issue for instance and realize they have to rewrite 50k of code just to handle it. So the less they know about the gotchas the more error checking, hooks, and gestalt they put in. This gets really ugly though when there isn't a well-defined goal (ie 'filesharing with windows') because coders go crazy making it overly generic and flexible.
WOULD YOU LIKE TO PLAY A GAME?
> GLOBAL WAR oN TERRORISM
BEGIN TORTURE INSURGENTS
INSURGENT RECRUITING INCREASE 180%
INCREASE TORTURE
INSURGENTS REACH CRITICAL MASS ACQUIRE NUKE
LAUNCH FINANCIAL EMBARGO ATTACK
INSURGENT RECRUITING INCREASE 160%
STARVE POPULATION
INSURGENTS REACH CRITICAL MASS ACQUIRE BIOWEAPON
ACQUIRE NUKE
ACQUIRE BIOWEAPON
ACQUIRE SARIN
ACQUIRE GREY GOO
STRANGE GAME. THE ONLY WINNING MOVE.
IS NOT TO HATE.
Yes it is. All the other captains had 'hard' sounds in their name... KIRk, PIcard, KAtherine. Those are names you can say with power befitting a captain. Just imaging Khan saying "I've done far worse than kill you blah blah blah I shall leave you as you left me, as you left her: marooned for all eternity in the center of a dead planet, buried alive. Buried alive.". And Kirk replying "Alex! Alex!!! Alex!!!!!!!!" (cue pullback shot from outside asteroid). Sounds lame doesn't it. Chase would be even lamer, if that's possible and it is.
Probably a joke but shouldn't be. The kernel developers could make it possible to accelerate the JVM/CLR by giving faster access to the actual hardware pages. What I mean is, the JVM has a 'scratch' area where recently created objects are allocated and then after garbage collecting this area the leftover objects are moved out. They do this because the vast majority of objects last only a tiny amount of time. So to be able to do a GC *only* of this scratch area the JVM actually replaces reference assignment with code that ads to a big list of all objects that took references to objects in this scratch area. Even though this is obviously slow it means the JVM doesn't have to garbage collect the entire contents of memory to make sure nothing has a ref to a new object.
If the JVM could get access to the hardware's dirty page bit that says if a page has been modified since last checked then the JVM could do direct reference assignment and then when garbage collecting only search the modified pages for references into the 'scratch' area. I expect this would be many times faster than the pointer write barrier used now.
Maybe a system call could take a mmap region and return a bitmask of page dirty flags? I think in any case there are plenty of things the kernel developers could do to make software virtual machines better if they tried. I think they just don't care to since that world is alien to them.
Java GUIs have traditionally been 'slow' because they are double-buffered by default, and as a consequence they redraw completely before displaying. Hardware simply wasn't fast enough to do this gracefully -- it's only recently that most gnome/kde applications and some xp ones are expected to be double-buffered.
The double-buffering also lead to lots of inefficient widget redrawing, like for a while each widget was cleared with the bg color before being redrawn even if it then say put an image say over its whole area.
The other major slowdown was because Java's graphics were much more advanced than necessary, for example lines of width != 1 with end and joint caps, antialiasing, clipping regions (instead of boxes), custom renderers, etc. This made it difficult to integrate with the simple hardware acceleration at the time. Native apps had jaggy lines and solid colors as the main features.
You can't make a computer secure so that's why you make your system not rely on the computer:
Voter ---> Computer --> PRINTOUT --> BOX --> Counting --> Final Tally
I could care less if the computer is rigged as long as the printout say what I actually thought I voted for and as long as I can observe the counting.
Voting machine:
1. Setup linux distro with apache, tomcat, whatever
2. Install ballot web app
3. Install ballot CUPS printer filter
4. Setup firefox for kiosk mode
Counting machine:
1. Setup linux distro with ballot_counter.py
2. Attach scanner
3. Run ballots through OCR software
4. Update counters (in realtime as scanned)
Ballots print like this, one measure per line:
PRESIDENT: AL GORE
SENATE: JAMES WEBB
STEM-CELL: YES
Also the tally program just keeps track of unique lines returned from the OCR and puts this count next to the on-screen list (sorted by count). Thus it doesn't need to know anything about the election. The top N lines on the screen as votes are scanned (after polls close) will automatically show the votes from the ballot followed by the most popular write-ins (ie Mickey Mouse).
You can add fancy CSS styles, javascript to prevent accidental undervoting, screen readers, on-screen keyboard, etc. There... a complete, secure, reliable, OSS, and cheap voting system for a day's work.