The people writing "applications" for this type of machine are used to writing for message-passing cluster architectures. These are jobs that have a *lot* of inheirant parallelism already. Mostly math-heavy simulation type stuff, such as protein folding, weather simulation, chemical simulation, and the like. Things with large matrix multiplies, billions of independant simulation cells, or massive fourier transforms, all of which are *easy* to parallelize.
Given that these jobs used to be run on message-passing clusters, writing on a large NUMA shared-memory machine should be easy. At the worst, you simulate MPI and still get performance gains. At the best, you can use the tighter coupling allowed to improve performance and/or run less inheirantly parallel jobs.
35 TFLOPS is the peak performance number sitewide. Cobalt itself should be able to clear between 6 and 7, making it a much more modest 25ish place. There are rumours that a bigger cluster-style machine is in the works, once the issues with Tungten (NCSA's biggest and #5 in the world) are ironed out.
Does this mean that the applications running on the "old" clusters, presumably using some flavor of MPI to communicate between nodes, will have to be ported somehow to become multithreaded applications ?
NCSA still has plenty of "old" style clusters around. Two of the more aging clusters, Platinum and Titan are being retired, to make room for newer systems like Cobalt. Indeed, the official notice was made just recently--they're going down tommorrow. However, as the retirement notice points out, we still have Tungsten, Copper, and Mercury (Terragrid). Indeed, Tungsten is number 5 on the Top 500, so it should provide more than enough cycles for any message-passing jobs people require.
So, anyone has any insights as to why/how this matters for the programmers ?
What it means is that programming big jobs is easier. You no longer need to learn MPI, or figure out how to structure your job so that individual nodes are relatively loosely-coupled. Also, jobs that have more tightly-coupled parallelism are now possible. The older clusters used high-speed interconnects like Myrinet or Infiniband (NCSA doesn't own any Infiniband AFAIK, but we're looking at it for the next cluster supercomputer). Although they provided really good latency and bandwidth, they aren't as high-performing as shared memory. Also, Myrinet's ability to scale to huge numbers of nodes isn't all that great--Tugsten may have 1280 compute nodes, but a job that uses all 1280 nodes isn't practical. Indeed, untill recently the Myrinet didn't work at all, even after partitioning the cluster into smaller subclusters.
This new shared-memory machine will be more powerful, more convienient, and easier to maintain than the cluster-style supercomputers. Hopefully it will allow better scheduling algorithms than on the clusters too--an appaling number of cycles get thrown away because cluster scheduling is non-preemptive.
I'd also like to point out some errors in the Computerworld article. NCSA is *currently* storing 940 TB in near-line storage (Legato DiskXtender running on an obscenely big tape library), and growing at 2TB a week. The DiskXtender is licenced for up to 2 petabytes--we're coming close to half of that now. The article therefore vastly understates our storage capacity. On the other hand, I'd like to know where we're hiding all those teraflops of compute--35 TFLOPS after getting 6 TFLOPS from Cobalt sounds more than just a little high. That number smells of the most optimistic peak performance values of all currently connected compute nodes. I.e. - how many single-precision operations could the nodes do if they didn't have to communicate, everything was in L1 cache, we managed to schedule something on all of them, and they were all actually functioning. Realistically, I'd guess that we can clear maybe a quarter of that figure, given machines being down, jobs being non-ideal, etc. etc. etc.
As a disclaimer, I do work at NCSA, but in Security Research, not High-Performance Computing.
Really, that's a little unfair. I mean, not patching has been relatively consequence free for quite a while now. Suddenly dooming them right away is a bit harsh.
Rather, I'd create a small family of malwares, and have each one leave behind some indication of that it had been there. Do it in some way that the virus scanners may have a hard time cleaning it up. Also, notify the users that they've been hit. Tell them this is their last chance to repent. Give them pointers to resources to help them repent. Remember, the end is near...
After people have been duly warned by worms and viruses that have had their chance to spread and die out, I'd release the punisher malware. Something nasty, a blended threat type thing using all the latest techniques to spread. Now, this would be the one with the payload. It would look for evidence that you'd been hit with one of the previous malwares, and assign points based on how they got in, and if they were cleaned up. Say 1 point for getting infected by a zero-day exploit in Opera that you've since cleaned up after, but 100 points for getting hit by an email worm which proudly announces "I am a virus. Do not click on the pretty linkey and run me!" or somesuch. Points can also be assigned based on what software is on the system. A machine loaded with spyware, or infected with other viruses gets more points. One with a firewall or behind a NAT box gets a deduction.
Now what is the purpose of the points, you ask? A person collecting many points gets their machine fubar'ed. A person with no points gets a notice of how our nasty nasty worm got in, and help cleaning it up. People in between perhaps get all their valuable MP3s scrambled, or something. You get the picture.
Now, one really can't say that everyone hasn't recieved warning, now. Everyone who got infected previously got notice, and help to change their wicked ways. Of course, some people are rather dense. So, after a few months, we'd have to start over again. Sure, it'll be harder, since many people will have wised up (I hope so), but many probably won't. This time around, however, the stick should come closer in time to the bad behaviour. Reinforcement learning works better that way. Additionally, I think the standards for who gets the smackdown should be lowered. Frag'ing their systems for incrementially smaller violations every cycle will hopefully get people to shape up. After a while, it won't matter--an unpatched system won't last long in such a hostile environment. And if the software is just inheirently insecure *cough*IEandActiveX*cough* then the best thing to do is just not run it.
Sadly, this won't teach anyone. Especially those runninng "alternative" software--be it application or OS. Apple's customers are rather poor at running anti-virus software. One can't blame them--there isn't much reason to. I myself don't worry about email viruses at all--that's because I read my mail with Pine. On a Sparc. Somebody else's Sparc, where I don't keep any valuable data. Now, I'm sure Pine has some sort of exploitable bugs in it. With all of the MS PC's either patched or nuked to oblivion, what's a poor virus writer to do?
"I don't know what language engineers will be using in 25 years, but they will call it FORTRAN".
I mean, a modern FORTRAN has all sorts of fancy features such as recursion and heap-based allocation. Today's FORTRAN programmers have an unprecedented level of abstraction available to them in this dynamic, high-level language.
Finally, note that on modern drives, you can seek all the way across the disk in only about 30-50% more time than you can seek a short seek. Thus keeping your blocks close to the current cylinder but not in it has very limited value. Note that this is not the case on optical (CD/DVD) disks.
Bzzt. Wrong. Thanks for playing.
Seagate Barracuda 7200.7 Plus product manual. This is a modern drive. Look at page 17 of the PDF, section 2.7, "Seek Time". Track to track seek is listed as sub 1ms, while average seek is 8.5ms (for read). Latency (the rotational aspect), which is separate, is an average of 4.16 (as it should be for a 7200 RPM drive). So a short seek is 8 times faster than an average seek, much less a whole-drive seek. So keeping your blocks close to the current cylinder but not in it still has high value.
Additionally, if you can keep your data in the same track, you don't have any seek time, just rotational latency. And the size of track groups has been growing as densities have been growing. So there are lots and lots of blocks in the same track that aren't within readahead range.
And fragmentation is not a bugaboo. It's a fact. When you have random allocation on a volume, it will get fragmented.
Now you misuse the term "random allocation". When talking about disk, random allocation means that you randomly choose your next block--which certainly will cause fragmentation. I think what you are looking for is random file creation. However, under FFS, if you maintain sufficient free space, it is very unlikely that real fragmentation will occur even with random file creation/deletion. Yes, you won't be able to store all of your files contiguously. But as the grandparent points out, the old FFS block allocator finds "nearly optimal" blocks. From the original paper on FFS, you don't get serious fragmentation-related performance issues until you reach 90% disk utilization (there's an '86 FFS paper I can't find an electronic copy of which does a better analysis, but even the '84 paper has the 90% figure). At 90% utilization, nearly every file system ever starts getting severe performance problems due to fragmentation.
You can go back after the fact and unfragment it, but doing so in any serious fashion when writing files actually degrades performance due to the extra effort required.
You should read about log-structured file systems sometime. Like Sprite.
The base idea of a log-structured file system
(LFS) is that you don't try to keep your blocks the same. You write a log of changes, and stream that out continuously to make maximum use of your available write bandwidth. This has the severe downside of causing horrific freespace fragmentation, since every change to a file means the affected blocks are reallocated, and the old blocks are now garbage. So you have continuous freespace compaction (called segment cleaning in the paper). LFSs didn't catch on because at the time, they needed a large amount of free space compared to more traditional "overwrite" file systems to maintain performance. However, recent work with log file systems shows that the changing performance characteristics of disks (much higher bandwidth but same seek/rotational delays) have tipped the balance to log structuring. Regardless, the cleaning process is precisely going back and unfragmenting your data, and is necessary for a LFS.
HFS+ does nothing to prevent fragmentation except for use super-clusters.
As pointed out by others, Apple's later implementations do defragmentation of smaller files when they are accessed.
NTFS could do this, but I believe they do not. However, NTFS on servers has an allocation block size of 8K
The high conviction rate comes primarily from the way job performance and opportunities for advancement happen in the Japanese law enforcement/legal system.
At the level of the individual officer, your job performance is how many "good" arrests you make (those that lead to convictions), and how many "bad" arrests you make (those that do not lead to a conviction). Making a bad arrest is essentially a career-killer. So, they are very careful to only arrest those against whom they have a watertight case. Somewhat more negatively, anyone with enough political influence that they *might* be able to get off won't be arrested at all. Organized crime figures are essentially immune to investigation from lower-level law enforcement--they can't get a conviction, so they don't even bother. Most negatively is what happens when a truly innocent party is arrested. There is intense pressure to build a case regardless of actual guilt or innocence. What percentage of innocent suspects are railroaded by the police is unknown, although probably not too high.
At the next level is the prosecution. Again, convictions are good, and failed attempts at prosecution==dead career. Additionally, as the parent post notes, there is a limited budget to bring about a prosecution. So, the prosecution ends up cherry-picking cases. Compounding the success rate is that (as elsewhere in the world) the police generally won't arrest anyone with the prosecuting attorney's OK. Again, this has the negative effect that "hard" cases aren't even considered, and there is the occasional attempt to railroad an innocent suspect.
At the last level is the judiciary. At this point, potential cases have been through two very stringent filters. The trial is almost unnecessary at this point, but judges do occassionally acquit. As the parent post notes, judges who acquit end up with worse careers. The analysis says this isn't based on acquitting the innocent, but to acquitting on "reasons of statutory or constitutional interpretation, often in politically charged cases." However, it seems that nearly any case with an acquittal is "polically charged" since the prosecution will use politics to their advantage in a weaker case.
Overall, the Japanese criminal legal system has some serious problems. Most of the problems come from the inability or lack of desire to attempt any "hard" or "weak" cases, but some comes from aggressively pushing a case which shouldn't have been brought in the first place. On the other hand, it mosly works. Overall crime and incarceration rates are low, and success is the hardes metric against which to argue. Still doesn't mean that the Winny author has a snowballs chance of getting off.
but I don't see what role zynot can fill in the linux landscape.
If it does nothing else, it consumes resources that would have otherwise been used by Gentoo. Forks can be a good thing, but they can also be a bad thing. The OpenBSD fork ended up being a good thing--OpenBSD fills an entirely different role from NetBSD. The GCC/EGCS fork also ended up being good--the new totally replaced the old.
In a way, I agree with you--the Zynot fork probably is hurting more than helping. There was a window of time to get moving, and that window has passed. Now all the fork is doing is diverting resources. On the other had, that a fork was created can have a positive impact just by existing. It highlights problems with the original project, and creates a drive to fix those problems. And if the problems aren't fixed, the secondary project is waiting in the wings to take over.
My point in noting the fork was this: you don't create a fork unless there is a problem. Generally a big problem, where a lot of effort has been put in to try and resolve, but nothing has come of it. Regardless of whether Zynot amounts to anything, there was a fork, and that is signifigent in and of itself.
First, I am not commenting on the Why of drobbins' departure. I saw many posts complementing Mr. Robbins on his work (complements which he does deserve), and making statements to the effect of "this will be bad for gentoo". I felt the need to respond to the second half of these posts.
By disconnected, I don't mean unresponsive to day-to-day issues. Two examples of disconnect that come most readily to mind are the xfree86 license snafu and the legendary tightness by which core *BSD source trees are controlled. The first is ignoring the needs/wants of your support base, the second is ignoring the partipation of your support base.
Supposing you are right, and that drobbins is walking away because his is burnt out, then I applaud that. The trouble comes in that the "pissing people off" part started some time ago. It hasn't cause major damage to the project (like xfree86's self-immolation), but there was a non-negligable negative effect. His departure would have been better 9 months ago (prior to the Zynot fork). On the other hand, hindsight is 20/20--it is quite likely that staying on was the best decision given the information available at the time.
Really, I'm just trying to put a different light on the end effect of the departure. Whatever the reasons, drobbins is leaving, and that will have an effect. As an administrator, drobbins had his faults, and they were starting to wear on the project. While a committee is not as focused as an individual, committees scale better. It isn't the end of the line for gentoo, just another step.
While he certainly was the driving force behind Gentoo, it wasn't all good.
To put it bluntly, Mr. Robbins is an utter PITA to work with. Many of the same things that have been said about Theo de Raadt (true or not) could be said about drobbins. This has alread caused a fork in Gentoo. My very limited exposure to dealing with Mr. Robbins left me with such a bad taste in my mouth I just decided to drop it rather than put up with it.
Gentoo is a wonderful distribution, which I still use. But like several other open source projects (*cough*xfree86*cough*thebsds*cough*), those running the project were operating disconnected from their user/developer base. It isn't that central control is bad (Linus keeps control of final say) but that a central control that doesn't play nice with others is going to spell trouble.
Hmm, the length of this post got a little out of control....
yet you don't seem to realize that if you have a method whereby you can always acquire a product for free, you completely undermine the basis for the system
You seen not to understand the full nuances of price theory. Here is a decent overview of the concept of marginal cost.
For the lazy amoung you, the point is that if it is possible to get a product for free, then that is going to be its price. More formally, in a non-distorted market the long run price of any good will approach the marginal cost or production--that is, the cost to produce the last additional unit. Prices above the marginal cost are, in the long run, unsustainable in a fair, undistorted market, since any competitors can increase their total profits by increasing production and reducing price, and new firms will find it profitable to move into the industry and begin producing and selling nearer the marginal cost.
It is important to note that fixed costs do not come into the picture at all--they increase the average cost, and cut into profits, but in a fair market they have little effect on actual prices. Fixed costs do effect whether firms decide to produce at all, though. So if it isn't possible to recover your fixed costs by selling at your marginal cost, then you simply leave the industry (AKA go out of business) or neglect to enter the industry at all. And we can note that high fixed costs do not equate high prices--look at the semiconductor industry. Fabs are **expensive**, but most parts sell for close to marginal cost, collusion and short-term fluctuations notwithstanding.
The most glaring example of prices approaching marginal cost in action can be seen in farm goods. This is an open market, where the individual firms (farms, in this case) have been unable to collude, that has been around for long enough for long-run forces to take full effect. The price of corn, milk, eggs, flour, etc. _is_ the marginal cost of the least efficient producer still in business. Well, there could a few firms which are selling at below their costs, and they will shortly go out of business, or it is possible that demand has driven up prices short term and new firms haven't yet entered the market, but those are short-term boundary effects. Unless you have better land, equipment, or intrinsic skill as a farmer than the marginal producer, there is no money to be had in farming--as farmers have been bemoaning for years.
True, DVD movies are not nearly as fungable as corn or rice. But to a large extent, any one action film is a good substitute for another. So it should be possible for a new movie studio to start up shop and start producing action movies, charging movie theatres less to show them and selling the DVDs at $10 each rather than $20. And they should therefore by competitive pressure force the other studios either to reduce their prices or to go out of business by loss of market share.
We can observe that since the cost of a DVD movie is obviously waaaay more than the marginal cost, the market for movies must be distorted. Some of the distortion can be explained by the effects of copyright law--a necessary distortion. But $20 for something which has a marginal cost of closer to $5 cannot be explained away by copyright's distortionary effects alone. There must therefore be some other form of market distortion--either outright collusion or some other unfair trade practice.
More evidence can be seen in the cost/gross numbers of most movies. Take "Tommorrow Never Dies". While it can sell on the strength of the 007/James Bond brand, it is mostly a standard action film. An expensive action film to be sure--$110 million to produce. But it has grossed $210 million--a $100 million dollar net. Lets assume (for the sake of argument) that the money arrived in $21 million dollar increments over 10 years. These assumptions heavily unde
It is important to remember, that for most purposes all of the codecs they tests except Quicktime/Sorenson are actually MPEG4. I don't know if WMP9 is standards-compliant, but it is based on MPEG4 technology. DivX *is* MPEG4, with many of the optional features implemented. And of course Quicktime/MPEG4 is too.
So except for Sorenson, and the differences in packing technology (WMP, DivX, and Quicktime use different container formats), this is a comparison of 3 different MPEG4 implementations. This isn't really a codec shootout--it is an implementation shootout. Not that they included the best MPEG4 implementation available, XVid, but what do you expect from Ziff-Davis.
Really, this just proves that MPEG4 is a really good standard. With its impressive features (double indirect frames, global motion compensation, quarter-pel interpolation, improved motion vector compression, and object based decomposition), there really isn't much more to do without making major changes in how things are done. Just about the only refinement I can think of would be to move from DCT-based entropy coding to a wavelet-based entropy coding, like the move from JPEG to JPEG2000. That would help the codec scale to higher resolutions and lower bitrates more gracefully, but in the common case wouldn't gain much.
Another thing, which is better--to tell people "Go away, I don't want to talk to you", or to wear headphones and let them figure it out on their own. If I'm going to be in a situation where I know people are going to be pestering me, I'll take my MP3 player along. I don't get hassled, and the hasslers don't get a punch in the mouth. Everybody is better off.
I posted in more detail above, but the Laputa dub makes up additional dialogue. This isn't a creative translatoin, this is a minute and a half of dialogue in English where there is none in Japanese. A good long stretch of it right at the beginning. Check it out yourself if you don't believe me.
Especially after watching that, I don't think the Disney name detracts from the release at all.
While I can agree that many newer english dubs (WHR, Noir, and Haibane Renmei come to mind) are rather good, Disney's dubs are not amoung them. I'm sure others can generate a more complete list of Disney doing a terrible job, but I'll stick with one particularly glaring example: the opening of "Castle in the Sky".
Watch it with the translated subtitles on, but the English soundtrack. There's an awful lot of dialogue, but no subtitles to match? Are the subtitles broken or something? Try the beginning again with Japanese language. Nobody is saying anything. The dub has new dialouge spun out of whole cloth. This isn't a creative interpretation of what was being said, changing the word structure to get the lips to sync, or even a sloppy job. This is purposeful and deliberate mangling of Miyazaki's work.
Perhaps the flak Disney took for these changes (and others in Mononoke-hime) convinced them to remain more faithful to the original script in Spirited Away, but I don't trust them. I'll listen to a little bit of the English dub just to see the quality of the voice acting (which, is of higher quality than usual, but not as good as the newer stuff ADV or Bandai is putting out) and to see if I can match voices to famous actors, but that's it. Watching the entire thing when I can't be sure if what I'm hearing is genuine or mangled isn't worth it.
That, and Disney is just evil anyway. Well, Michael Eisner's Disney is at least.
Inkjet printers are hardly economical for light usage either. A sealed ink cartridge will last for a long time, but as soon as you open it up and print from it, its lifetime is strictly limited. So you are more or less guaranteed to be replacing your ink cartridges on a strict timetable (3-9 months depending on the manufacturer), unless you are a heavy user--in which case you will replace them faster.
It therefore makes sense even for light users to go laser. Yes, the printer itself is more expensive--but it should last a very long time. My previous printer was an HP LaserJet 6P I got used. It took 4 years heavy use (abuse, rather) in a hospital office, and then 4 years of light use from me. It finally died after I (stupidly) printed several hundred pages with inadequate space to let it cool, and some of the rear rollers melted. In 4 years, I only needed to replace the toner cartridge once, and that was starting from a partially used cartridge. I'm now using a used LaserJet 1200, and I expect to get at least 6 years use from it. Overall, very inexpensive printing.
Of course, a laser printer generally doesn't do color, unless you get a color laser. A color laser printer is way more expensive, but you really should ask yourself--how often do you need color? And how much is it costing you to print that color? A lot of people will find that it is cheaper to just take their color jobs to Kinko's. Their online price for a single-sided color laser print job on 28lb paper (nice and thick) is only $1.00--send it off and pick it up when it's done. The quality could hardly be better. And Kinko's is the high end of print shops--I'm sure one can find better prices elsewhere.
Very few people have an excuse to use a inkjet printer. Yes, lasers are more expensive up front, but it isn't 5x anymore, more like 3x to 2x. And even for light users, the consumables costs of inkjets add up very fast. Unless there's some insane deal ($10 printer and 5 free ink carts!), I will never again own an inkjet printer.
Saying that it isn't a strange move is an understatement--large corps regularly retain counsel simply to monopolize the availability of the best firms. On the up side, that every large corp is attempting this tends to fragment the law firms--if Ford has a firm on retainer, then somebody else can hire that firm to fight Toyota, and vise versa. So the little guy isn't completely screwed, unless you'd like to sue an entire industry. But if you have that many enemies, you're SOL anyway...
Slashdot Mirror
The people writing "applications" for this type of machine are used to writing for message-passing cluster architectures. These are jobs that have a *lot* of inheirant parallelism already. Mostly math-heavy simulation type stuff, such as protein folding, weather simulation, chemical simulation, and the like. Things with large matrix multiplies, billions of independant simulation cells, or massive fourier transforms, all of which are *easy* to parallelize.
Given that these jobs used to be run on message-passing clusters, writing on a large NUMA shared-memory machine should be easy. At the worst, you simulate MPI and still get performance gains. At the best, you can use the tighter coupling allowed to improve performance and/or run less inheirantly parallel jobs.
35 TFLOPS is the peak performance number sitewide. Cobalt itself should be able to clear between 6 and 7, making it a much more modest 25ish place. There are rumours that a bigger cluster-style machine is in the works, once the issues with Tungten (NCSA's biggest and #5 in the world) are ironed out.
No, you get the message
"Windows has detected a new sound card and is installing it"
followed by
"Windows needs to restart your computer. Click OK to restart".
in a cycle of 1024 reboots.
Does this mean that the applications running on the "old" clusters, presumably using some flavor of MPI to communicate between nodes, will have to be ported somehow to become multithreaded applications ?
NCSA still has plenty of "old" style clusters around. Two of the more aging clusters, Platinum and Titan are being retired, to make room for newer systems like Cobalt. Indeed, the official notice was made just recently--they're going down tommorrow. However, as the retirement notice points out, we still have Tungsten, Copper, and Mercury (Terragrid). Indeed, Tungsten is number 5 on the Top 500, so it should provide more than enough cycles for any message-passing jobs people require.
So, anyone has any insights as to why/how this matters for the programmers ?
What it means is that programming big jobs is easier. You no longer need to learn MPI, or figure out how to structure your job so that individual nodes are relatively loosely-coupled. Also, jobs that have more tightly-coupled parallelism are now possible. The older clusters used high-speed interconnects like Myrinet or Infiniband (NCSA doesn't own any Infiniband AFAIK, but we're looking at it for the next cluster supercomputer). Although they provided really good latency and bandwidth, they aren't as high-performing as shared memory. Also, Myrinet's ability to scale to huge numbers of nodes isn't all that great--Tugsten may have 1280 compute nodes, but a job that uses all 1280 nodes isn't practical. Indeed, untill recently the Myrinet didn't work at all, even after partitioning the cluster into smaller subclusters.
This new shared-memory machine will be more powerful, more convienient, and easier to maintain than the cluster-style supercomputers. Hopefully it will allow better scheduling algorithms than on the clusters too--an appaling number of cycles get thrown away because cluster scheduling is non-preemptive.
I'd also like to point out some errors in the Computerworld article. NCSA is *currently* storing 940 TB in near-line storage (Legato DiskXtender running on an obscenely big tape library), and growing at 2TB a week. The DiskXtender is licenced for up to 2 petabytes--we're coming close to half of that now. The article therefore vastly understates our storage capacity. On the other hand, I'd like to know where we're hiding all those teraflops of compute--35 TFLOPS after getting 6 TFLOPS from Cobalt sounds more than just a little high. That number smells of the most optimistic peak performance values of all currently connected compute nodes. I.e. - how many single-precision operations could the nodes do if they didn't have to communicate, everything was in L1 cache, we managed to schedule something on all of them, and they were all actually functioning. Realistically, I'd guess that we can clear maybe a quarter of that figure, given machines being down, jobs being non-ideal, etc. etc. etc.
As a disclaimer, I do work at NCSA, but in Security Research, not High-Performance Computing.
Really, that's a little unfair. I mean, not patching has been relatively consequence free for quite a while now. Suddenly dooming them right away is a bit harsh.
Rather, I'd create a small family of malwares, and have each one leave behind some indication of that it had been there. Do it in some way that the virus scanners may have a hard time cleaning it up. Also, notify the users that they've been hit. Tell them this is their last chance to repent. Give them pointers to resources to help them repent. Remember, the end is near...
After people have been duly warned by worms and viruses that have had their chance to spread and die out, I'd release the punisher malware. Something nasty, a blended threat type thing using all the latest techniques to spread. Now, this would be the one with the payload. It would look for evidence that you'd been hit with one of the previous malwares, and assign points based on how they got in, and if they were cleaned up. Say 1 point for getting infected by a zero-day exploit in Opera that you've since cleaned up after, but 100 points for getting hit by an email worm which proudly announces "I am a virus. Do not click on the pretty linkey and run me!" or somesuch. Points can also be assigned based on what software is on the system. A machine loaded with spyware, or infected with other viruses gets more points. One with a firewall or behind a NAT box gets a deduction.
Now what is the purpose of the points, you ask? A person collecting many points gets their machine fubar'ed. A person with no points gets a notice of how our nasty nasty worm got in, and help cleaning it up. People in between perhaps get all their valuable MP3s scrambled, or something. You get the picture.
Now, one really can't say that everyone hasn't recieved warning, now. Everyone who got infected previously got notice, and help to change their wicked ways. Of course, some people are rather dense. So, after a few months, we'd have to start over again. Sure, it'll be harder, since many people will have wised up (I hope so), but many probably won't. This time around, however, the stick should come closer in time to the bad behaviour. Reinforcement learning works better that way. Additionally, I think the standards for who gets the smackdown should be lowered. Frag'ing their systems for incrementially smaller violations every cycle will hopefully get people to shape up. After a while, it won't matter--an unpatched system won't last long in such a hostile environment. And if the software is just inheirently insecure *cough*IEandActiveX*cough* then the best thing to do is just not run it.
Sadly, this won't teach anyone. Especially those runninng "alternative" software--be it application or OS. Apple's customers are rather poor at running anti-virus software. One can't blame them--there isn't much reason to. I myself don't worry about email viruses at all--that's because I read my mail with Pine. On a Sparc. Somebody else's Sparc, where I don't keep any valuable data. Now, I'm sure Pine has some sort of exploitable bugs in it. With all of the MS PC's either patched or nuked to oblivion, what's a poor virus writer to do?
That's just my 2 cents
After they've been hit with one of these, I do believe that "can't shoot back" will describe their opponents.
Unfortunately, Lisp is predated by FORTRAN.
"I don't know what language engineers will be using in 25 years, but they will call it FORTRAN".
I mean, a modern FORTRAN has all sorts of fancy features such as recursion and heap-based allocation. Today's FORTRAN programmers have an unprecedented level of abstraction available to them in this dynamic, high-level language.
Plus, it's reeaallyy fast.
Finally, note that on modern drives, you can seek all the way across the disk in only about 30-50% more time than you can seek a short seek. Thus keeping your blocks close to the current cylinder but not in it has very limited value. Note that this is not the case on optical (CD/DVD) disks.
Bzzt. Wrong. Thanks for playing.
Seagate Barracuda 7200.7 Plus product manual. This is a modern drive. Look at page 17 of the PDF, section 2.7, "Seek Time". Track to track seek is listed as sub 1ms, while average seek is 8.5ms (for read). Latency (the rotational aspect), which is separate, is an average of 4.16 (as it should be for a 7200 RPM drive). So a short seek is 8 times faster than an average seek, much less a whole-drive seek. So keeping your blocks close to the current cylinder but not in it still has high value.
Additionally, if you can keep your data in the same track, you don't have any seek time, just rotational latency. And the size of track groups has been growing as densities have been growing. So there are lots and lots of blocks in the same track that aren't within readahead range.
And fragmentation is not a bugaboo. It's a fact. When you have random allocation on a volume, it will get fragmented.
Now you misuse the term "random allocation". When talking about disk, random allocation means that you randomly choose your next block--which certainly will cause fragmentation. I think what you are looking for is random file creation. However, under FFS, if you maintain sufficient free space, it is very unlikely that real fragmentation will occur even with random file creation/deletion. Yes, you won't be able to store all of your files contiguously. But as the grandparent points out, the old FFS block allocator finds "nearly optimal" blocks. From the original paper on FFS, you don't get serious fragmentation-related performance issues until you reach 90% disk utilization (there's an '86 FFS paper I can't find an electronic copy of which does a better analysis, but even the '84 paper has the 90% figure). At 90% utilization, nearly every file system ever starts getting severe performance problems due to fragmentation.
You can go back after the fact and unfragment it, but doing so in any serious fashion when writing files actually degrades performance due to the extra effort required.
You should read about log-structured file systems sometime. Like Sprite. The base idea of a log-structured file system (LFS) is that you don't try to keep your blocks the same. You write a log of changes, and stream that out continuously to make maximum use of your available write bandwidth. This has the severe downside of causing horrific freespace fragmentation, since every change to a file means the affected blocks are reallocated, and the old blocks are now garbage. So you have continuous freespace compaction (called segment cleaning in the paper). LFSs didn't catch on because at the time, they needed a large amount of free space compared to more traditional "overwrite" file systems to maintain performance. However, recent work with log file systems shows that the changing performance characteristics of disks (much higher bandwidth but same seek/rotational delays) have tipped the balance to log structuring. Regardless, the cleaning process is precisely going back and unfragmenting your data, and is necessary for a LFS.
HFS+ does nothing to prevent fragmentation except for use super-clusters.
As pointed out by others, Apple's later implementations do defragmentation of smaller files when they are accessed.
NTFS could do this, but I believe they do not. However, NTFS on servers has an allocation block size of 8K
Your sig is wrong. It ought be libellous bastards.
I would frankly nail the libellous sons-of-bitches to the wall, profit or no.
I thing you meant libellous bastards.
The high conviction rate comes primarily from the way job performance and opportunities for advancement happen in the Japanese law enforcement/legal system.
At the level of the individual officer, your job performance is how many "good" arrests you make (those that lead to convictions), and how many "bad" arrests you make (those that do not lead to a conviction). Making a bad arrest is essentially a career-killer. So, they are very careful to only arrest those against whom they have a watertight case. Somewhat more negatively, anyone with enough political influence that they *might* be able to get off won't be arrested at all. Organized crime figures are essentially immune to investigation from lower-level law enforcement--they can't get a conviction, so they don't even bother. Most negatively is what happens when a truly innocent party is arrested. There is intense pressure to build a case regardless of actual guilt or innocence. What percentage of innocent suspects are railroaded by the police is unknown, although probably not too high.
At the next level is the prosecution. Again, convictions are good, and failed attempts at prosecution==dead career. Additionally, as the parent post notes, there is a limited budget to bring about a prosecution. So, the prosecution ends up cherry-picking cases. Compounding the success rate is that (as elsewhere in the world) the police generally won't arrest anyone with the prosecuting attorney's OK. Again, this has the negative effect that "hard" cases aren't even considered, and there is the occasional attempt to railroad an innocent suspect.
At the last level is the judiciary. At this point, potential cases have been through two very stringent filters. The trial is almost unnecessary at this point, but judges do occassionally acquit. As the parent post notes, judges who acquit end up with worse careers. The analysis says this isn't based on acquitting the innocent, but to acquitting on "reasons of statutory or constitutional interpretation, often in politically charged cases." However, it seems that nearly any case with an acquittal is "polically charged" since the prosecution will use politics to their advantage in a weaker case.
Overall, the Japanese criminal legal system has some serious problems. Most of the problems come from the inability or lack of desire to attempt any "hard" or "weak" cases, but some comes from aggressively pushing a case which shouldn't have been brought in the first place. On the other hand, it mosly works. Overall crime and incarceration rates are low, and success is the hardes metric against which to argue. Still doesn't mean that the Winny author has a snowballs chance of getting off.
but I don't see what role zynot can fill in the linux landscape.
If it does nothing else, it consumes resources that would have otherwise been used by Gentoo. Forks can be a good thing, but they can also be a bad thing. The OpenBSD fork ended up being a good thing--OpenBSD fills an entirely different role from NetBSD. The GCC/EGCS fork also ended up being good--the new totally replaced the old.
In a way, I agree with you--the Zynot fork probably is hurting more than helping. There was a window of time to get moving, and that window has passed. Now all the fork is doing is diverting resources. On the other had, that a fork was created can have a positive impact just by existing. It highlights problems with the original project, and creates a drive to fix those problems. And if the problems aren't fixed, the secondary project is waiting in the wings to take over.
My point in noting the fork was this: you don't create a fork unless there is a problem. Generally a big problem, where a lot of effort has been put in to try and resolve, but nothing has come of it. Regardless of whether Zynot amounts to anything, there was a fork, and that is signifigent in and of itself.
First, I am not commenting on the Why of drobbins' departure. I saw many posts complementing Mr. Robbins on his work (complements which he does deserve), and making statements to the effect of "this will be bad for gentoo". I felt the need to respond to the second half of these posts.
By disconnected, I don't mean unresponsive to day-to-day issues. Two examples of disconnect that come most readily to mind are the xfree86 license snafu and the legendary tightness by which core *BSD source trees are controlled. The first is ignoring the needs/wants of your support base, the second is ignoring the partipation of your support base.
Supposing you are right, and that drobbins is walking away because his is burnt out, then I applaud that. The trouble comes in that the "pissing people off" part started some time ago. It hasn't cause major damage to the project (like xfree86's self-immolation), but there was a non-negligable negative effect. His departure would have been better 9 months ago (prior to the Zynot fork). On the other hand, hindsight is 20/20--it is quite likely that staying on was the best decision given the information available at the time.
Really, I'm just trying to put a different light on the end effect of the departure. Whatever the reasons, drobbins is leaving, and that will have an effect. As an administrator, drobbins had his faults, and they were starting to wear on the project. While a committee is not as focused as an individual, committees scale better. It isn't the end of the line for gentoo, just another step.
While he certainly was the driving force behind Gentoo, it wasn't all good.
To put it bluntly, Mr. Robbins is an utter PITA to work with. Many of the same things that have been said about Theo de Raadt (true or not) could be said about drobbins. This has alread caused a fork in Gentoo. My very limited exposure to dealing with Mr. Robbins left me with such a bad taste in my mouth I just decided to drop it rather than put up with it.
Gentoo is a wonderful distribution, which I still use. But like several other open source projects (*cough*xfree86*cough*thebsds*cough*), those running the project were operating disconnected from their user/developer base. It isn't that central control is bad (Linus keeps control of final say) but that a central control that doesn't play nice with others is going to spell trouble.
I hope that Gentoo will end up better for it.
If you can't figure out how to stretch the work you do have out an additional 2 hours, I have no pity for you.
Hmm, the length of this post got a little out of control....
yet you don't seem to realize that if you have a method whereby you can always acquire a product for free, you completely undermine the basis for the system
You seen not to understand the full nuances of price theory. Here is a decent overview of the concept of marginal cost.
For the lazy amoung you, the point is that if it is possible to get a product for free, then that is going to be its price. More formally, in a non-distorted market the long run price of any good will approach the marginal cost or production--that is, the cost to produce the last additional unit. Prices above the marginal cost are, in the long run, unsustainable in a fair, undistorted market, since any competitors can increase their total profits by increasing production and reducing price, and new firms will find it profitable to move into the industry and begin producing and selling nearer the marginal cost.
It is important to note that fixed costs do not come into the picture at all--they increase the average cost, and cut into profits, but in a fair market they have little effect on actual prices. Fixed costs do effect whether firms decide to produce at all, though. So if it isn't possible to recover your fixed costs by selling at your marginal cost, then you simply leave the industry (AKA go out of business) or neglect to enter the industry at all. And we can note that high fixed costs do not equate high prices--look at the semiconductor industry. Fabs are **expensive**, but most parts sell for close to marginal cost, collusion and short-term fluctuations notwithstanding.
The most glaring example of prices approaching marginal cost in action can be seen in farm goods. This is an open market, where the individual firms (farms, in this case) have been unable to collude, that has been around for long enough for long-run forces to take full effect. The price of corn, milk, eggs, flour, etc. _is_ the marginal cost of the least efficient producer still in business. Well, there could a few firms which are selling at below their costs, and they will shortly go out of business, or it is possible that demand has driven up prices short term and new firms haven't yet entered the market, but those are short-term boundary effects. Unless you have better land, equipment, or intrinsic skill as a farmer than the marginal producer, there is no money to be had in farming--as farmers have been bemoaning for years.
True, DVD movies are not nearly as fungable as corn or rice. But to a large extent, any one action film is a good substitute for another. So it should be possible for a new movie studio to start up shop and start producing action movies, charging movie theatres less to show them and selling the DVDs at $10 each rather than $20. And they should therefore by competitive pressure force the other studios either to reduce their prices or to go out of business by loss of market share.
We can observe that since the cost of a DVD movie is obviously waaaay more than the marginal cost, the market for movies must be distorted. Some of the distortion can be explained by the effects of copyright law--a necessary distortion. But $20 for something which has a marginal cost of closer to $5 cannot be explained away by copyright's distortionary effects alone. There must therefore be some other form of market distortion--either outright collusion or some other unfair trade practice.
More evidence can be seen in the cost/gross numbers of most movies. Take "Tommorrow Never Dies". While it can sell on the strength of the 007/James Bond brand, it is mostly a standard action film. An expensive action film to be sure--$110 million to produce. But it has grossed $210 million--a $100 million dollar net. Lets assume (for the sake of argument) that the money arrived in $21 million dollar increments over 10 years. These assumptions heavily unde
3 words: minority shareholder lawsuit.
Gates and co. may own a controling block, but if they are clearly making bad decisions, things can get sticky for them.
It is important to remember, that for most purposes all of the codecs they tests except Quicktime/Sorenson are actually MPEG4. I don't know if WMP9 is standards-compliant, but it is based on MPEG4 technology. DivX *is* MPEG4, with many of the optional features implemented. And of course Quicktime/MPEG4 is too.
So except for Sorenson, and the differences in packing technology (WMP, DivX, and Quicktime use different container formats), this is a comparison of 3 different MPEG4 implementations. This isn't really a codec shootout--it is an implementation shootout. Not that they included the best MPEG4 implementation available, XVid, but what do you expect from Ziff-Davis.
Really, this just proves that MPEG4 is a really good standard. With its impressive features (double indirect frames, global motion compensation, quarter-pel interpolation, improved motion vector compression, and object based decomposition), there really isn't much more to do without making major changes in how things are done. Just about the only refinement I can think of would be to move from DCT-based entropy coding to a wavelet-based entropy coding, like the move from JPEG to JPEG2000. That would help the codec scale to higher resolutions and lower bitrates more gracefully, but in the common case wouldn't gain much.
Another thing, which is better--to tell people "Go away, I don't want to talk to you", or to wear headphones and let them figure it out on their own. If I'm going to be in a situation where I know people are going to be pestering me, I'll take my MP3 player along. I don't get hassled, and the hasslers don't get a punch in the mouth. Everybody is better off.
I posted in more detail above, but the Laputa dub makes up additional dialogue. This isn't a creative translatoin, this is a minute and a half of dialogue in English where there is none in Japanese. A good long stretch of it right at the beginning. Check it out yourself if you don't believe me.
Especially after watching that, I don't think the Disney name detracts from the release at all.
While I can agree that many newer english dubs (WHR, Noir, and Haibane Renmei come to mind) are rather good, Disney's dubs are not amoung them. I'm sure others can generate a more complete list of Disney doing a terrible job, but I'll stick with one particularly glaring example: the opening of "Castle in the Sky".
Watch it with the translated subtitles on, but the English soundtrack. There's an awful lot of dialogue, but no subtitles to match? Are the subtitles broken or something? Try the beginning again with Japanese language. Nobody is saying anything. The dub has new dialouge spun out of whole cloth. This isn't a creative interpretation of what was being said, changing the word structure to get the lips to sync, or even a sloppy job. This is purposeful and deliberate mangling of Miyazaki's work.
Perhaps the flak Disney took for these changes (and others in Mononoke-hime) convinced them to remain more faithful to the original script in Spirited Away, but I don't trust them. I'll listen to a little bit of the English dub just to see the quality of the voice acting (which, is of higher quality than usual, but not as good as the newer stuff ADV or Bandai is putting out) and to see if I can match voices to famous actors, but that's it. Watching the entire thing when I can't be sure if what I'm hearing is genuine or mangled isn't worth it.
That, and Disney is just evil anyway. Well, Michael Eisner's Disney is at least.
But what about the bootloader?
Inkjet printers are hardly economical for light usage either. A sealed ink cartridge will last for a long time, but as soon as you open it up and print from it, its lifetime is strictly limited. So you are more or less guaranteed to be replacing your ink cartridges on a strict timetable (3-9 months depending on the manufacturer), unless you are a heavy user--in which case you will replace them faster.
It therefore makes sense even for light users to go laser. Yes, the printer itself is more expensive--but it should last a very long time. My previous printer was an HP LaserJet 6P I got used. It took 4 years heavy use (abuse, rather) in a hospital office, and then 4 years of light use from me. It finally died after I (stupidly) printed several hundred pages with inadequate space to let it cool, and some of the rear rollers melted. In 4 years, I only needed to replace the toner cartridge once, and that was starting from a partially used cartridge. I'm now using a used LaserJet 1200, and I expect to get at least 6 years use from it. Overall, very inexpensive printing.
Of course, a laser printer generally doesn't do color, unless you get a color laser. A color laser printer is way more expensive, but you really should ask yourself--how often do you need color? And how much is it costing you to print that color? A lot of people will find that it is cheaper to just take their color jobs to Kinko's. Their online price for a single-sided color laser print job on 28lb paper (nice and thick) is only $1.00--send it off and pick it up when it's done. The quality could hardly be better. And Kinko's is the high end of print shops--I'm sure one can find better prices elsewhere.
Very few people have an excuse to use a inkjet printer. Yes, lasers are more expensive up front, but it isn't 5x anymore, more like 3x to 2x. And even for light users, the consumables costs of inkjets add up very fast. Unless there's some insane deal ($10 printer and 5 free ink carts!), I will never again own an inkjet printer.
"Plutonium may give you grief for thousands of years, but arsenic is forever"
Good Omens, Neil Gaiman and Terry Pratchett
Saying that it isn't a strange move is an understatement--large corps regularly retain counsel simply to monopolize the availability of the best firms. On the up side, that every large corp is attempting this tends to fragment the law firms--if Ford has a firm on retainer, then somebody else can hire that firm to fight Toyota, and vise versa. So the little guy isn't completely screwed, unless you'd like to sue an entire industry. But if you have that many enemies, you're SOL anyway...