SMP is definetely worth it. I bought my dual P450 at summer, and it is still fast:). My uptimes have been around 50 days, but the shutdowns have not been crashes. (Well, except one incident with hdparm..)
I've had my X crash sometimes with GL-applications, but I'm not convinced it is an smp-problem.
And yes, if an application is not thread-safe (and uses threads), it is more likely that it crashes on smp than on a single processor. This is because considering that tasks are switched 100 times per second (if they're busy), a code fragment can be easily executed in 1/100 seconds - but on smp, if the locks aren't there, the violating code segments are ran simultaenously, hilighting the problems with poor locking.
I'm sorry to see this issue was clearly (?) not taken care from the very beginning (or atleast from the current reincarnation's beginning), as it might be hard if not impossible to fix all the race conditions as an afterthought.. Perhaps adding global locks will cure, but then the performance won't be as good as it could be.
Uh, are you sure? Ethernet mac space is 281 billion numbers in size, and has no such artificial partitioning as IP does. (Except for the two numbers in front to identify manufacturer; there can be many numbers/manufacturer though.)
First, Linux doesn't even need 4MB if you strip stuff you don't need. (ext2fs, isofs, ipx, you get the pictures.) I must say that I don't know what the situation is with the 2.2 (or 2.3) series, but I've seen low-resource distributions based on 1.1-series that need only as little as 2MB, and still have the stuff needed for running your computer;).
What is the idea with having a real OS in it, then? The application base! Easy to develope! Anyone could just get a cross-compiling gcc and port their favourite hello-world application over. And if this thing ever gets a keyboard, I can see emacs ported over;).
Of course, there are small os'es that provide POSIX interface (isn't QNX one of these?), but still, it's less hassle to port from exactly similar environment.
About usability (desktop) - hey, I don't care, I could write my own. (And somebody will almost certainly write a palm3-clone for this thingy.)
Actually it does run some sort of web server, as one of the shots show an url (partially visible) like alhost/foo/bar, where alhost would apparently be localhost - dunno if it really connects to itself via tcp/ip, though.
Because this way you already have a large base of software you can run in it.. Plus developing software by yourself would be really easy. No need to download such sdk's that you'd need to boot into the other os..
You haven't actually tried for example the demos in latest JDK?
You really -don't- need a web browser to run them, or even the applet viewer. And you're not missing the graphics. I've done some programs that access sql in Java, no web browser involved there!
Yes, I've done some sql-stuff in perl too, but it's nice to have a concept of 'types' (instead of just 'arrays, scalars and hashes'). This way you can see already in the compile time if something is broken before you run it.. And perhaps that's the reason I like C++ the best of these three: you don't need to subclass everything from 'Object' or somesuch and do dynamic typecasting all the time (templates save your ass). You could say that the types in C++ are more static.. (Which has its downsides too, though.)
Java's strength is that it has many library functions/classes with it, and you know that anyone else going to run your program have them too. On every platform. If they don't have the packages you use (for example I don't expect my friends to have postgresql.jar), you can just deliver the package with the program and it will work without difficult installing (just make sure CLASSPATH has "." too).
I think a machine running java as native code would be kinda cool. Btw, there's already a code generator for gcc to output java.. (No, I haven't tried it.)
Nice. People at kernel.org must be really happy now when you've shown THAT link instead of the mirrors-link. Of course, it might not be mirrored everywhere yet..
What sucks with airplane hijackings compared to home accidents, that with airplane accidents someone does the thing on purpose, and can do it too. As a though it really pisses me off: someone just decides to hijack a plane, and then you find that your plane has been hijacked. You can't blame the Great Existance (or whatever) for it.
Often accidents at home are your own fault, or then a machine breaks down and there's not much you could've (known to) done about it..
Perhaps this would work: - Allow web-servers to receive and reply via udp too - Define a protocol for queries (could be something as simple as querying for/cgi-bin/search?foo, but it should be the same everywhere). Perhaps with an extension that client could request a reply always or only in cases when something of interest was actually found. - The server could forward the query to other servers it sees fit - thus companies with many web-servers could make their main server to ask every server in the company -> one query would sweep the whole house. - Make a list of web-servers that are capable of replying to these requests. I guess the list could be either 'strobed' or done more conventionally with spiders and companies telling about them to the server.
Now, when someone wants to search something, he does a query to a server that sends the query to all the servers, which could send the reply directly to the client. I guess this would involve a java-client so everyone could easily use it - of course, native clients would be nice too.
I see the problem with the query-server choking its tube with sending udp-requests to many sites, which is why they should use the already mentioned subquerying on hosts all around the world (which all would send the replies directly to the client). The traffic would be basically only sending with the search server. Unless, of course, the traditional search engine was also used - this would remove burden of udping gazillion little sites.
Of course, nothing would prevent anyone from sending a query to a server. This approach also makes life difficult for firewalled people. Plus you could get MANY answers, as no server knows how 'good' answers you've already received. But hey, it'd be nice to see in action, worry about these problems later:).
Google does exactly this. I don't actually use any other search engines nowadays anymore - with most queries there's no chance in finding it with for example Altavista, it returns so much crap. Google has a linux-search too.
Also, according to this, google uses linux, and on many machines too:).
I think the problem with a distributed irc/icq-network is that there needs to be a server you connect to first.
However, I've been thinking it might be possible to make all clients not only file servers but also 'access servers' - build the network hierarchy on fly. All you'd need to know would be one server in the network, and you'd be on. A good guess would be the local subnet's broadcast address. If the server feels you're consuming too much bandwidth, it could tell you to move to another server - after all, once you're in the network, you could know all that.
If you're being really paranoid about letting information about yourself out to the net, it could be arranged so that each server not only 'routes' the queries/chat, but also masquerades it, thus hiding the original IP. Of course, the actual transmission of files would still need ip-addresses to be exhanged..
There could also be a trust ring, so that if I trust someone and you trust me, you also trust that someone. I don't mean the whole network should be like this, but perhaps some people could have need for having their own rings of trust.
False results can he handled easily: just submit the packet to two different places, or to 1.5 places in average, and if they disagree, the system checks the packet by itself (or hands over to third machine.) Yes, It'll slow down, but I can't see any other viable alternative..
Data theft.. Isn't the idea that the data is already there, but it needs to be processed? No idea in data theft then. Also the system could look after domains or ip-address spaces that keep eating and eating the data space faster than anyone else and blackhole them.. Or sue them:).
What if someone took a picture of your house, from safe distance? Would that be wrong? How about if they took pictures of every house in the world, but they wouldn't write down the addresses of these houses.. Would that be illegal?
Re:What's wrong with not being anonymous?
on
License to Surf
·
· Score: 1
Well, you still need to tell someone who you are. Not so in the net. Also, if I ever saw you in the street (not likely), I wouldn't know it till I asked. And even if I did ask, I might not remember the connection - in the net computers can do all this automatically.
Basically, this wouldn't work. People could always change ISP's, and I personally wouldn't like my ISP blocking any of my traffic anywhere. Perhaps the other way around though, blackholing spammers in their routers is ok for me.
Do you think there aren't real reasons to send 100+ mails a day? Perhaps you've backlogged after being on a vacation. Perhaps you run a mailing list. What ever the reason might be, I don't want my ISP to read my email in any case!
Fining for spam is ok for me, but hey, why would I choose such an ISP if I were a spammer - propably all ISP's still wouldn't have such demands. Of course, if big upstreams demanded that from their clients, we'd be getting somewhere..
Securing SMTP I take for granted. Debian 2.2 uses exim as the default mailer, and the configuration asks which subdomains it allows the mail to be relayed from.
Slashdot Mirror
I've had my X crash sometimes with GL-applications, but I'm not convinced it is an smp-problem.
And yes, if an application is not thread-safe (and uses threads), it is more likely that it crashes on smp than on a single processor. This is because considering that tasks are switched 100 times per second (if they're busy), a code fragment can be easily executed in 1/100 seconds - but on smp, if the locks aren't there, the violating code segments are ran simultaenously, hilighting the problems with poor locking.
I'm sorry to see this issue was clearly (?) not taken care from the very beginning (or atleast from the current reincarnation's beginning), as it might be hard if not impossible to fix all the race conditions as an afterthought.. Perhaps adding global locks will cure, but then the performance won't be as good as it could be.
Uh, are you sure? Ethernet mac space is 281 billion numbers in size, and has no such artificial partitioning as IP does. (Except for the two numbers in front to identify manufacturer; there can be many numbers/manufacturer though.)
What is the idea with having a real OS in it, then? The application base! Easy to develope! Anyone could just get a cross-compiling gcc and port their favourite hello-world application over. And if this thing ever gets a keyboard, I can see emacs ported over ;).
Of course, there are small os'es that provide POSIX interface (isn't QNX one of these?), but still, it's less hassle to port from exactly similar environment.
About usability (desktop) - hey, I don't care, I could write my own. (And somebody will almost certainly write a palm3-clone for this thingy.)
Actually it does run some sort of web server, as one of the shots show an url (partially visible) like alhost/foo/bar, where alhost would apparently be localhost - dunno if it really connects to itself via tcp/ip, though.
Because this way you already have a large base of software you can run in it.. Plus developing software by yourself would be really easy. No need to download such sdk's that you'd need to boot into the other os..
So, Crusoe is just vaporware too, then?
You really -don't- need a web browser to run them, or even the applet viewer. And you're not missing the graphics. I've done some programs that access sql in Java, no web browser involved there!
Yes, I've done some sql-stuff in perl too, but it's nice to have a concept of 'types' (instead of just 'arrays, scalars and hashes'). This way you can see already in the compile time if something is broken before you run it.. And perhaps that's the reason I like C++ the best of these three: you don't need to subclass everything from 'Object' or somesuch and do dynamic typecasting all the time (templates save your ass). You could say that the types in C++ are more static.. (Which has its downsides too, though.)
Java's strength is that it has many library functions/classes with it, and you know that anyone else going to run your program have them too. On every platform. If they don't have the packages you use (for example I don't expect my friends to have postgresql.jar), you can just deliver the package with the program and it will work without difficult installing (just make sure CLASSPATH has "." too).
I think a machine running java as native code would be kinda cool. Btw, there's already a code generator for gcc to output java.. (No, I haven't tried it.)
Of course, this was done already long before DeCSS ever existed. So DeCSS isn't really necessary for vcd-piratism either.
Nice. People at kernel.org must be really happy now when you've shown THAT link instead of the mirrors-link. Of course, it might not be mirrored everywhere yet..
Often accidents at home are your own fault, or then a machine breaks down and there's not much you could've (known to) done about it..
- Allow web-servers to receive and reply via udp too
- Define a protocol for queries (could be something as simple as querying for
- The server could forward the query to other servers it sees fit - thus companies with many web-servers could make their main server to ask every server in the company -> one query would sweep the whole house.
- Make a list of web-servers that are capable of replying to these requests. I guess the list could be either 'strobed' or done more conventionally with spiders and companies telling about them to the server.
Now, when someone wants to search something, he does a query to a server that sends the query to all the servers, which could send the reply directly to the client. I guess this would involve a java-client so everyone could easily use it - of course, native clients would be nice too.
I see the problem with the query-server choking its tube with sending udp-requests to many sites, which is why they should use the already mentioned subquerying on hosts all around the world (which all would send the replies directly to the client). The traffic would be basically only sending with the search server. Unless, of course, the traditional search engine was also used - this would remove burden of udping gazillion little sites.
Of course, nothing would prevent anyone from sending a query to a server. This approach also makes life difficult for firewalled people. Plus you could get MANY answers, as no server knows how 'good' answers you've already received. But hey, it'd be nice to see in action, worry about these problems later :).
Also, according to this, google uses linux, and on many machines too :).
However, I've been thinking it might be possible to make all clients not only file servers but also 'access servers' - build the network hierarchy on fly. All you'd need to know would be one server in the network, and you'd be on. A good guess would be the local subnet's broadcast address. If the server feels you're consuming too much bandwidth, it could tell you to move to another server - after all, once you're in the network, you could know all that.
If you're being really paranoid about letting information about yourself out to the net, it could be arranged so that each server not only 'routes' the queries/chat, but also masquerades it, thus hiding the original IP. Of course, the actual transmission of files would still need ip-addresses to be exhanged..
There could also be a trust ring, so that if I trust someone and you trust me, you also trust that someone. I don't mean the whole network should be like this, but perhaps some people could have need for having their own rings of trust.
False results can he handled easily: just submit the packet to two different places, or to 1.5 places in average, and if they disagree, the system checks the packet by itself (or hands over to third machine.) Yes, It'll slow down, but I can't see any other viable alternative..
:).
Data theft.. Isn't the idea that the data is already there, but it needs to be processed? No idea in data theft then. Also the system could look after domains or ip-address spaces that keep eating and eating the data space faster than anyone else and blackhole them.. Or sue them
What if someone took a picture of your house, from safe distance? Would that be wrong? How about if they took pictures of every house in the world, but they wouldn't write down the addresses of these houses.. Would that be illegal?
Well, you still need to tell someone who you are. Not so in the net. Also, if I ever saw you in the street (not likely), I wouldn't know it till I asked. And even if I did ask, I might not remember the connection - in the net computers can do all this automatically.
Am I missing something here? Why isn't 1.1.1999 an 'odd day'?
Basically, this wouldn't work. People could always change ISP's, and I personally wouldn't like my ISP blocking any of my traffic anywhere. Perhaps the other way around though, blackholing spammers in their routers is ok for me.
Do you think there aren't real reasons to send 100+ mails a day? Perhaps you've backlogged after being on a vacation. Perhaps you run a mailing list. What ever the reason might be, I don't want my ISP to read my email in any case!
Fining for spam is ok for me, but hey, why would I choose such an ISP if I were a spammer - propably all ISP's still wouldn't have such demands. Of course, if big upstreams demanded that from their clients, we'd be getting somewhere..
Securing SMTP I take for granted. Debian 2.2 uses exim as the default mailer, and the configuration asks which subdomains it allows the mail to be relayed from.