In IPv6, the MAC address is kept in the ethernet frame but also in the low 48 bits of the IP address. Thus, routers do not need to have an ARP lookup table to get the MAC address - they can simply copy-and-paste from the IP address in the packet (for the final step) or the IP address of the next router in the path (for all other steps).
This means the number of tables for lookups is reduced by 1 and there is no need to do reverse lookups (so there is no latency in such activity). It is also central to the way IPv6 handles mobility, as it means (a) you're guaranteed there is an IP address available for you in the network you join, (b) the host part of the IP address will remain the same, only the network component will change, and (c) because only the network component changes, routers will be capable of re-routing traffic upstream to the new destination with zero packet loss.
(Most mobile IP uses forwarders, but IPv6 was designed from the start to have mobility within the protocol as far as possible and not as a hack.)
ARP did not strictly go away, it is now simply merged in. Zeroconf is achieved by getting the address of the network from the router, and the local address of the node by getting th MAC address, then combining the two to build the full address.
Wrongo on the routing. The last 48 bits are reserved for the node's MAC address and the first 16 bits are reserved for the type of traffic. The rest is heirarchically defined by the router. (The router advertisment is done via RADV and the address is discovered by the machine from that advertisment).
The practical upshot is that if the traffic is for that LAN, you need only test the 48 bits for the MAC address. If it is for a node further downstream, there will be a non-zero value in the next byte after your router heirarchy addreess, provided that is NOT within the MAC address. If it is for a node reachable upstream, then one of the bytes within the router heirarchy address will be different (up to an absolute maximum of 8 bytes, which is 2 words on a 32-bit machine or 1 word on a 64-bit system).
IPv6 also does not support packet fragmentation - the network is interrogated to find the largest supported packet from end-to-end, so stateful routing will be unnecessary, reducing the CPU workload. Also, because there are no fragments, packets should be more reliable. In IPv4, if a fragment is lost, the whole packet is resent. This not only increases the opportunities of a packet loss, it also increases the network load on a retransmit, which means a greater chance of packets being lost on the retransmit.
It's interesting to figure out what legacy equiptment out there will prove bothersome. Layer 2 switches won't notice or care. Cisco routers have supported IPv6 for a decade now. Bay - long dead - was also an early adopter, so many of their routers should be IPv6-capable, with no need of any updates. Linux has had IPv6 patches since 2.0.20, and mainstream since 2.1.8. I think IPv6 was added in Solaris 2.5.1. There was an alternative Windows TCP/IP stack by TCP Software that supported IPv6 about 9 to 10 years ago.
All in all, if anyone's complaining about a lack of support, it's NOT because support has been lacking.
There are fuel economy races all over Europe, Silverstone motor racing circuit being used for the British race in the series. These used to require petrol (gasoline to Americans) engines only, but in the last few years, this could be supplemented by other energy sources (but not human power). When I was still in 6th form, the winner had managed 6,500 mpg. The last race prior to introducing alternative energy saw an amazing 9,998 mpg. The races after that, to me, aren't nearly as interesting as it is impossible to distinguish on the numbers alone an improvement in design (of car or engine) from a really good, sunny day.
For the UBC to be at a paltry third of the efficiency of European cars is not terribly impressive in itself, unless the burdens placed by the rules are substantially more severe.
On a side-note, it occured to me some time back that very often, students living in a University city need something a little more solid than a bicycle and a lot cheaper to maintain than a full car. These vehicles would sorta fit into this category. The idea I have is for nearly-disposable cars, where it has sufficient fuel and oil to last a year or more of typical student usage. The student rents it for an academic year for next to nothing, needs to perform zero maintenance for the whole time, and then returns it. This eliminates any fuel price issues, the risk of running out of fuel when going to lectures or dates, etc.
Minis filled this role OK, but they're a pain to maintain and are relatively expensive on fuel. The biggest drawbacks are that the fuel efficient cars are incapable of carrying any significant weight (so forget carrying the books for a day - those would weigh more than the car!) and that you can't exactly carpool with them. The lack of creash resistance is a non-issue, as minis have a habit of exploding on impact. I'd swear that the scriptwriters for the A-Team must have owned minis.
Tell that to the Gigapixl project.:) Seriously, professionals will have geeks amongst them, and geeks love anything that's new, sparkly and does cool stuff. Furthermore, "serious" professionals are subject to market forces, and the market is more likely to buy bigger, better, flashier photograps - even if they can't tell the difference - simply because the adverts look more impressive. Impressions, even when illusions, sell. And photographers know that.
Finally, there are markets for extreme detail - surverying for construction, surveying for archaeology, etc. The latter is perhaps the most interesting, as these photographs are aimed at finding uber-faint, large-scale (but very thin) features from overhead photographs. Standard photographs will work and do work, but they invariably miss a lot. These new cameras might reduce how much they miss (by a little) and might increase other information they have to work with (by a little).
To generalize and say that professionals won't be interested, therefore, is unlikely to hold up to scrutiny. What will be interesting is what WILL professionals do, when this is more widely available?
There might be some validity in that, if it weren't that Microsoft's certs were obtained from Verisign by phishers a few years back. Regardless of anything else, I would NOT be willing to pay a company for a cert that doesn't certify.
That was a bad chair pun you threw. It needed more back and maybe a swivel thing. As it stood, it had nolegs to stand on. (Well, after Ballmer snapped them off, anyway...) Talking of Ballmer, does he chair a meeting and throw a party, or....
I think they should have Mrs. T (Thatcher) in charge of Microsoft. I can hear it now... "This hard-drive is NOT for turning!" Besides, those handbags wielded far more destructive power than a mere chair ever could.
I'm involved in a number of Open Source projects (covering crypto, multicasting, reconstructions of genuinely ancient games, etc) and volunteer projects (processing archaeological data, mostly) where there is a desperate need for coders who are determined to get results, who can be given assignments & deadlines, etc. I don't know what sort of bounties I could pony up, but if there are coders out there who would like to be given a nice, encapsulated, well-defined project, I would be more than happy to write up a formal requirements doc, milestones, deadlines, etc.
I'm probably not the only person who can provide the structure. (There are a bazillion project sites out there, but sites != structure.) There are an amazing number of projects out there. The problem is that there are simply not enough people to go around, and the lines of communication between coders and projects has traditionally been poor. Proper requirements analysis and project specifications are rare to non-existant outside of the best-of-breed elite institutions, paid or otherwise. Most of this is because geeks are often poor communicators, so the projects that are interesting (ie: geek-run) are the ones people know least about, and the ones with the best PR (run by marketing) often have the least novel or interesting work involved in them. This makes it hard to find out what REAL work is out there.
Absurd! It's been an invaluable tool for interrogating terrorist suspects. Five minutes with clippy, and they'll talk for a week! Nonstop, in some cases.
Is there any chance you could talk him into forwarding some of the research his group did to me? I'd be fascinated to understand why the unicast outperformed the multicast, particularly for the large number of nodes - whether it was actually hardware, topological, the multicast implementation, etc.
Yes, a control node reduces to O(n), but you suffer from having a potential hot-spot and from the risks involved in having a single point of failure. Given the cost of supercomputer time (even on pile-of-pc clusters) these are not good characteristics to have. Again, I'd love to learn more about how they have solved/plan to solve such issues. They are exceptionally nasty.
I remember talking to some of the Los Alamos guys at SC|05 and don't recall hearing this stuff being mentioned, but IIRC most of those who had in-depth understanding were off wandering the tables when I came by. Really, supercomputing is an area that really needs a Slashdot section to itself - if not an entire collaboration site! - there is so much fascinating work going on where the fundamentally defining researcg simply isn't getting circulated much beyond the first few cubes.
(I fully respect an organization's right to privacy, and particularly understand the interesting IP complexities involved in Government labs - I've worked in enough! - where copyright limitations on the Government are routinely bypassed by either classifying the work or having a contractor do enough to have them claim the IP. Sometimes the really interesting stuff simply can't be circulated for actual reasons. However, I've also run into many situations where really interesting stuff isn't circulated simply because nobody thought to, and for absolutely no other reason at all, good, bad or indifferent.)
I suspect you are correct. The interesting work there now seems to be in the supercomputing arena. LAMPI is damn good, for example, and they're a key player in the OpenMPI consortium. My biggest problem is that their work seems... limited. Not limited as in value (it's enormously valuable), but limited in the sense that the scope of the field of enquiry is gigantic, but the work they're putting in is barely scratching the surface. They need to do more - a LOT more. We're talking ten to a hundred times the output, superior quality control, better documentation on what they are doing, better interaction with those Open Source communities they are working with, etc.
V9FS is great, etc. But it's unicast. So is Lustre (a ClusterFS product) and OpenMOSIX. How are you supposed to share files or resources in a scalable fashion using a method that worsens exponentially? It seems obvious to me that if you want REALLY big clusters, with heavy-duty node interaction, we need to be using better protocols than the cheap and easy ones.
(Multicasting is normally an "unreliable" protocol, but the NRL - Naval Research Labs - have produced a nice library called NORM which gives you highly scalable reliable multicast. It wouldn't be easy to move NORM into the kernel, but it's an obvious thing to do to maximize scalability. Once you can spawn a totally arbritary number of threads simultaneously over any number of nodes you care to name with ZERO added latency, and provide those nodes with ZERO overhead RDMA and DSM, then you'll have a supercomputer that won't care if you have a two node cluster or two million nodes. It'll work just fine.)
The Wen Ho Lee fiasco was stupid, bt (sadly) can be blamed on the Clinton administration. Ultimately, paranoia (regardless of which party is being paranoid) is going to produce stupid, if not insane, consequences.
Los Alamos did an excellent job with LAMPI (their high-performance, highly reliable MPI implementation) and are doing OK with OpenMPI (the multi-vendor replacement), but let's face it, MPI is hardly on the same level as other products they've worked on. I was fairly impressed by their demo of high-performance collective operations at SC|05, but again this is where the LOW-END of an organization like Los Alamos needs to be. The high-end should be solving stuff the rest of humanity hasn't even realized IS a problem.
...how many Slashdotians have noticed that the blog mentioned in an earlier story from a Microsoft engineer on the delays of Vista has been removed? It says it was without pressure, but I'd probably type that too if Steve Balmer were in the room, armed with heavy-duty chairs.
I'd find that funny if it weren't for the fact that I could see it happening. It's Microsoft's best hope of not being laughed off the face of the planet. (It would also explain the two redacted "black" USAF projects from a UK MoD report that was recently declassified...)
I've scoured the Internet to find even a single US carrier that sells T2 or T4 lines. They don't, although they do use these standards internally, and they sell the equiptment for supporting these standards to anyone who asks. My only guess is that there simply isn't enough demand for the speeds those lines provide to warrant running the lines for general use.
Having said that, since they use the standards internally, and since they are mandated under law to open up their COs (other than the optic fibre) to ISPs, and since the equiptment to run these lines is available on the open market, it would seem only a matter of time before companies needing lines faster than single T1 or T3 would simply create their own spin-off pseudo-ISP and run the T2/T4 lines themselves, where they're close enough to the CO to make this economic to do.
Actually, the T* standards are American.:) The Europeans use a similar system, but denote the lines as E1 through E4. The best information I have been able to obtain is that T2 and T4 are indeed implemented in the US but only internally within telecom companies. Why this is the case is unclear, as T2 and T4 devices exist on the open market - you can find them in many electronics catalogs that deal with high-speed networking. However, I assume that it is because T2 would yield lower profit margins than T3, and too few people would need T4 speeds to make it a profitable market.
You are correct on the speed ratings, although as I noted in my earlier post, telecom companies are extremely good at worming their way out of any agreements they violate (even when they are superb at getting the companies that buy the lines to cough up any overcharges).
I am more inclined to be sympathetic to this argument today than, say, a few years ago. A few years ago, the "leaker" of the original "Halloween Documents" could have expected to have been nailed upside-down to a cross, dipped in tar and used to light the way of a latter-day chariot race. Even today, I am deeply suspicious of what is allowed to be published - whether there is hidden censorship that allows only relatively benign criticism to give the illusion of freedom, or even whether the blogs are legit or disinformation. (Hell, the Pentagon admits bribing newspapers to publish fiction as though real, on occasion, so why assume Microsoft would do less? Microsoft is vastly superior at marketing techniques, after all.)
This is not to say the stories are fabrications - they may well be genuine. I hope they are, because it would indicate the company is starting to mature. My concern is that we have absolutely no way of knowing for sure, and their historic track record is NOT impressive. It would be a grave error to assume improved professional conduct merely because they're telling us that it is improved, just as much as it would be a grave error to assume the blogs are FUD merely because that is a common method of Microsoft's.
If Microsoft is brought to book on further antri-trust charges (regardless of which country), I hope that greater transparency is mandated. We need to know if they are sincerely trying to be honest, or if this is a campaign to delude and befuddle the IT industry further.
Slashdot Mirror
This means the number of tables for lookups is reduced by 1 and there is no need to do reverse lookups (so there is no latency in such activity). It is also central to the way IPv6 handles mobility, as it means (a) you're guaranteed there is an IP address available for you in the network you join, (b) the host part of the IP address will remain the same, only the network component will change, and (c) because only the network component changes, routers will be capable of re-routing traffic upstream to the new destination with zero packet loss.
(Most mobile IP uses forwarders, but IPv6 was designed from the start to have mobility within the protocol as far as possible and not as a hack.)
ARP did not strictly go away, it is now simply merged in. Zeroconf is achieved by getting the address of the network from the router, and the local address of the node by getting th MAC address, then combining the two to build the full address.
The practical upshot is that if the traffic is for that LAN, you need only test the 48 bits for the MAC address. If it is for a node further downstream, there will be a non-zero value in the next byte after your router heirarchy addreess, provided that is NOT within the MAC address. If it is for a node reachable upstream, then one of the bytes within the router heirarchy address will be different (up to an absolute maximum of 8 bytes, which is 2 words on a 32-bit machine or 1 word on a 64-bit system).
IPv6 also does not support packet fragmentation - the network is interrogated to find the largest supported packet from end-to-end, so stateful routing will be unnecessary, reducing the CPU workload. Also, because there are no fragments, packets should be more reliable. In IPv4, if a fragment is lost, the whole packet is resent. This not only increases the opportunities of a packet loss, it also increases the network load on a retransmit, which means a greater chance of packets being lost on the retransmit.
It's interesting to figure out what legacy equiptment out there will prove bothersome. Layer 2 switches won't notice or care. Cisco routers have supported IPv6 for a decade now. Bay - long dead - was also an early adopter, so many of their routers should be IPv6-capable, with no need of any updates. Linux has had IPv6 patches since 2.0.20, and mainstream since 2.1.8. I think IPv6 was added in Solaris 2.5.1. There was an alternative Windows TCP/IP stack by TCP Software that supported IPv6 about 9 to 10 years ago.
All in all, if anyone's complaining about a lack of support, it's NOT because support has been lacking.
...the're really meaning it'll go to the moon instead?
For the UBC to be at a paltry third of the efficiency of European cars is not terribly impressive in itself, unless the burdens placed by the rules are substantially more severe.
On a side-note, it occured to me some time back that very often, students living in a University city need something a little more solid than a bicycle and a lot cheaper to maintain than a full car. These vehicles would sorta fit into this category. The idea I have is for nearly-disposable cars, where it has sufficient fuel and oil to last a year or more of typical student usage. The student rents it for an academic year for next to nothing, needs to perform zero maintenance for the whole time, and then returns it. This eliminates any fuel price issues, the risk of running out of fuel when going to lectures or dates, etc.
Minis filled this role OK, but they're a pain to maintain and are relatively expensive on fuel. The biggest drawbacks are that the fuel efficient cars are incapable of carrying any significant weight (so forget carrying the books for a day - those would weigh more than the car!) and that you can't exactly carpool with them. The lack of creash resistance is a non-issue, as minis have a habit of exploding on impact. I'd swear that the scriptwriters for the A-Team must have owned minis.
Finally, there are markets for extreme detail - surverying for construction, surveying for archaeology, etc. The latter is perhaps the most interesting, as these photographs are aimed at finding uber-faint, large-scale (but very thin) features from overhead photographs. Standard photographs will work and do work, but they invariably miss a lot. These new cameras might reduce how much they miss (by a little) and might increase other information they have to work with (by a little).
To generalize and say that professionals won't be interested, therefore, is unlikely to hold up to scrutiny. What will be interesting is what WILL professionals do, when this is more widely available?
...it's not UUencoded or BOOd first... (Does anyone know if a BOO decoder still exists outside of the museum of neandethal technology?)
There might be some validity in that, if it weren't that Microsoft's certs were obtained from Verisign by phishers a few years back. Regardless of anything else, I would NOT be willing to pay a company for a cert that doesn't certify.
That was a bad chair pun you threw. It needed more back and maybe a swivel thing. As it stood, it had nolegs to stand on. (Well, after Ballmer snapped them off, anyway...) Talking of Ballmer, does he chair a meeting and throw a party, or....
I think they should have Mrs. T (Thatcher) in charge of Microsoft. I can hear it now... "This hard-drive is NOT for turning!" Besides, those handbags wielded far more destructive power than a mere chair ever could.
I'm probably not the only person who can provide the structure. (There are a bazillion project sites out there, but sites != structure.) There are an amazing number of projects out there. The problem is that there are simply not enough people to go around, and the lines of communication between coders and projects has traditionally been poor. Proper requirements analysis and project specifications are rare to non-existant outside of the best-of-breed elite institutions, paid or otherwise. Most of this is because geeks are often poor communicators, so the projects that are interesting (ie: geek-run) are the ones people know least about, and the ones with the best PR (run by marketing) often have the least novel or interesting work involved in them. This makes it hard to find out what REAL work is out there.
Absurd! It's been an invaluable tool for interrogating terrorist suspects. Five minutes with clippy, and they'll talk for a week! Nonstop, in some cases.
Yes, a control node reduces to O(n), but you suffer from having a potential hot-spot and from the risks involved in having a single point of failure. Given the cost of supercomputer time (even on pile-of-pc clusters) these are not good characteristics to have. Again, I'd love to learn more about how they have solved/plan to solve such issues. They are exceptionally nasty.
I remember talking to some of the Los Alamos guys at SC|05 and don't recall hearing this stuff being mentioned, but IIRC most of those who had in-depth understanding were off wandering the tables when I came by. Really, supercomputing is an area that really needs a Slashdot section to itself - if not an entire collaboration site! - there is so much fascinating work going on where the fundamentally defining researcg simply isn't getting circulated much beyond the first few cubes.
(I fully respect an organization's right to privacy, and particularly understand the interesting IP complexities involved in Government labs - I've worked in enough! - where copyright limitations on the Government are routinely bypassed by either classifying the work or having a contractor do enough to have them claim the IP. Sometimes the really interesting stuff simply can't be circulated for actual reasons. However, I've also run into many situations where really interesting stuff isn't circulated simply because nobody thought to, and for absolutely no other reason at all, good, bad or indifferent.)
V9FS is great, etc. But it's unicast. So is Lustre (a ClusterFS product) and OpenMOSIX. How are you supposed to share files or resources in a scalable fashion using a method that worsens exponentially? It seems obvious to me that if you want REALLY big clusters, with heavy-duty node interaction, we need to be using better protocols than the cheap and easy ones.
(Multicasting is normally an "unreliable" protocol, but the NRL - Naval Research Labs - have produced a nice library called NORM which gives you highly scalable reliable multicast. It wouldn't be easy to move NORM into the kernel, but it's an obvious thing to do to maximize scalability. Once you can spawn a totally arbritary number of threads simultaneously over any number of nodes you care to name with ZERO added latency, and provide those nodes with ZERO overhead RDMA and DSM, then you'll have a supercomputer that won't care if you have a two node cluster or two million nodes. It'll work just fine.)
Los Alamos did an excellent job with LAMPI (their high-performance, highly reliable MPI implementation) and are doing OK with OpenMPI (the multi-vendor replacement), but let's face it, MPI is hardly on the same level as other products they've worked on. I was fairly impressed by their demo of high-performance collective operations at SC|05, but again this is where the LOW-END of an organization like Los Alamos needs to be. The high-end should be solving stuff the rest of humanity hasn't even realized IS a problem.
Microsoft has implemented the Evil Bit using a flip-flop?
...how many Slashdotians have noticed that the blog mentioned in an earlier story from a Microsoft engineer on the delays of Vista has been removed? It says it was without pressure, but I'd probably type that too if Steve Balmer were in the room, armed with heavy-duty chairs.
sed s/for/to/
7. ???
8. Profit!!! (To the tune of a few hundred billion)
I'd find that funny if it weren't for the fact that I could see it happening. It's Microsoft's best hope of not being laughed off the face of the planet. (It would also explain the two redacted "black" USAF projects from a UK MoD report that was recently declassified...)
Having said that, since they use the standards internally, and since they are mandated under law to open up their COs (other than the optic fibre) to ISPs, and since the equiptment to run these lines is available on the open market, it would seem only a matter of time before companies needing lines faster than single T1 or T3 would simply create their own spin-off pseudo-ISP and run the T2/T4 lines themselves, where they're close enough to the CO to make this economic to do.
You are correct on the speed ratings, although as I noted in my earlier post, telecom companies are extremely good at worming their way out of any agreements they violate (even when they are superb at getting the companies that buy the lines to cough up any overcharges).
I'd be happy to swap idiots, but the swap device isn't working correctly.
This is not to say the stories are fabrications - they may well be genuine. I hope they are, because it would indicate the company is starting to mature. My concern is that we have absolutely no way of knowing for sure, and their historic track record is NOT impressive. It would be a grave error to assume improved professional conduct merely because they're telling us that it is improved, just as much as it would be a grave error to assume the blogs are FUD merely because that is a common method of Microsoft's.
If Microsoft is brought to book on further antri-trust charges (regardless of which country), I hope that greater transparency is mandated. We need to know if they are sincerely trying to be honest, or if this is a campaign to delude and befuddle the IT industry further.
Nonono. That's DOS developers. Windows developers go "What the SCANDISK.EXE"?