Mozilla wants me to update from 2.0.0.12 to 2.0.0.13. Is there any reason I shouldn't just go to 3.0 Beta 5? I'm assuming it either fixes that security bug or replaces it with some new ones.
Are the critical extensions available? For me, that's Adblock, NoScript, and Flashblock.
Lots of organizations put out canned news stories for TV, often with a "reporter" interviewing someone. Occasionally the government gets caught doing this kind of deception, but it's also used for commercial propaganda as well.
The big difference is that on the Internet, everything you read is true.
The other difference is on the Internet, nobody can tell that the government is a dog.
While personally I'd be more interested in seeing the Oscillation Overthruster, the fnords, and the white hole that'll be created after they fire the thing up, I'd say than an article by somebody who's actually seen the LHC really does deserve some Mod Informative points.
Sure, most of us haven't done much with fine-grained parallelism, unless we've been programming for graphics processors or the SSE / Altivec / etc. SIMD processing capabilities on PCs and Macs. On the other hand, most of us have used multi-user web servers, and many of us have used a fairly radical technology called "time-sharing" that has some relationship to the ancestors of Linux, from back when computers cost enough that more than one person might use them.... It's been a few years since I stuck my fingers into the guts of an operating system kernel, but Windows is happily (well, grudgingly) running about 30 processes which could get spread out across multiple cores, and generally the only thing that burns CPU on my computer is runaway Firefox processes, or brief bursts of computing when my VPN starts up. I suspect the graphics system is doing a lot more in the GPU, but I don't see that, and it's all SIMD stuff so it's not like finding parallelism for it to exploit is particularly difficult.
So how do you find workloads that can exploit many cores effectively? You can do servers, which can do just fine by sticking each client process/thread onto a single core, or reinvent database parallelism, or run virtual machines since nobody writes their applications securely enough to use Unix-style multiple users any more. Or you can find applications like graphics that parallelize well enough and use a bit of glue to hand of chunks of SIMD work to different cores. There are getting to be fewer and fewer CPU-bound problems that don't fit onto a single processor well. So maybe it's time to start moving more kinds of applications into the same chip - move the graphics off the GPU and back into the main processor, or have different types of cores for different applications.
If we can't find ways to do things like that, it's going to be annoying to chip-makers to have to find ways to make money as Moore's law makes high-speed CPUs cheaper and cheaper. There's a certain amount of Microsoft Bloatware(tm) and gamer framerate-lust that can keep them going, but basically desktop CPUs are already faster than most of us know what to do with. VmWare means that the fast multicore CPUs are getting exploited by people who use them to cut down on the number of servers they're buying by factors of 5-10, though the electricity savings means that you can sell a lot of server cores into the replacement market.
While you might not want to replace your entire disk infrastructure with RAM, there's an intermediate level which is to use Flash. The densities and price/GB are a good bit better than RAM, it doesn't need to stay powered to keep data, and you don't have millisecond-long delays from rotation or seeking, though the throughputs aren't as fast as RAM. The limited numbers of writes that old flash had have been improved a lot, and wear-leveling technology takes care of most of the rest. Operating systems aren't always really set up to use it (Vista supposedly has some capabilities), and you may end up just using it as swap space, but it's still theoretically a big win for some uses.
I was at NCR in the early 90s, and we were one of several companies selling parallel computers in the 4-16 processor class, running Unix (variants on System V Release 4.x), mostly for the general server market. The CPUs were Intel, either 486 or 586 depending on the year. Some of the companies were specialists, like Sequent, some were generalists like HP. I'd been using some of that technology in the late 80s as well. Also in the late 80s, at AT&T Bell Labs, we had some graphics machines with up to 64 or 128 DSPs on them, back when DSPs were already a lot faster than general-purpose chips. It let us get rendering speeds that you still couldn't achieve today without spending at least $29 for a video card (:-), and up to about a GigaFLOPS of horsepower if you could keep the floating-point pipelines full.
And we also had Teradata, which had specialized database machines, with up to about 500 disks, with a processor per disk and a master processor to split up workloads, typically spreading out an SQL query across a bunch of disks.
Parallel's always going to be with us, because it's an obvious technique to take whatever hardware's available that generation and build bigger machines by clustering a bunch of them together. The only real difference today is that the parallelism is increasingly moving onto the chips, because it's getting tougher to shrink chip features.
Sure, I was partly trolling (:-), but they are conservative, in the sense of being Establishment Radio. They're not right-wing ranters like Limbaugh (who was actually fun to listen to during the Clinton years), but they're generally supporting the Government, especially the Civil Service side of it, and while they're not out actively shilling for the Bush Administration, they pretty much let government-fed stories through with the default assumption that they're relatively accurate and framed honestly. They're happy to carry stories from the Democrat side of the Establishment (which is where most of the NPR folks come from) but they're too busy being "objective" to challenge obviously bogus propaganda for what it is.
Radio gives me several different things when I'm driving - music (I could listen to CDs instead), entertaining talk (podcasts could do that, but they don't replicate interactive talk radio well), news (I read news online at home, so radio news is mainly headlines as entertainment), sports scores (I don't care, but they tell me traffic is on next:-), and traffic. My car radio alternates between traffic-radio, conservative news/culture (NPR), and leftist news/music (KPFA Pacifica broadcasting), and it's usually on traffic radio if I need to know or CD player if I don't.
If I wanted to do something high-tech and expensive for traffic info, I could probably get some kind of traffic-integrated-GPS-thing; in a few cities there's also a low-tech cheap traffic widget that has a fixed LCD map and gets traffic by subscription for $5-10/month on some kind of radio channel. It's not as detailed about individual events as the every-10-minutes traffic radio, but it covers more of the highways and you don't have to wait for reports.
While they're downloading and rendering, I've got no problem with Firefox needing to use the CPU. But five minutes later, or an hour later, they shouldn't be using CPU time to maintain that. And especially after you've closed the tabs, they shouldn't be burning CPU time.
The date of Easter is approximately "The Sunday in Passover", because (unlike Christmas, for which there's no recorded time of year for the original event, and therefore the holiday was set to rip off Roman pagan holidays, though some modern Yuletide customs were adopted from northern Europeans) the events being celebrated at Easter happened in conjunction with the Passover holiday, and there are symbolic and theological connections to Passover in addition to just the date. Since the Jews have a lunar calendar and the Romans used a solar calendar, it was somewhat difficult for the Romans to reconcile the two, and they weren't willing to use the obvious method ("ask some Jews when Passover is each year").
If you want to say that Passover's date is set at spring pagan holiday time, you'll need to argue with your rabbi or maybe Lehrhaus Judaica about whether your druids are at all the same kinds of pagans as Caananites were. And if you want to say that the name "Easter" and the bunnies and and eggs and marshmallow chickens are ripped off from Germanic spring fertility goddess stuff, you'll have a tough time getting anybody to argue the other side except maybe some atheists who'll say that the Germanic fertility goddess folks ripped that off from nature, which provided the bunnies and eggs, or from the chemical industry who brought us marshmallow peeps.
[U]nlike Java [emphasis added], you can also write Evil Javascript, and you can also write Broken CPU-sucking Javascript. That's different from Java... how? That's different from Java because Java has a security model to prevent this, and most implementations do a pretty good job of not breaking it. Occasionally there's been an implementation bug, which has gotten fixed, but AFAIK (as of a few years ago, when I last paid attention to Java), the security model itself was still solid. No surprise, because that was one of Gosling's objectives in writing it. Javascript has bits of security tacked on the side, but that's about it.
Under WinXP, open the Task Manager to show CPU and memory consumption, minimize it, take Firefox 2.x to fark.com, open all the news article links in tabs, then close all of them (but leave the main page open.) Most of the time you'll not only have well over 100MB of RAM still in use, but you'll have your CPU smoking as well.
Ok, we're talking about how you'd use do computing differently if you could have all the memory you wanted. That's been done before (:-). The Massive Memory Machine Project, at Princeton, was a 1980s version of the same approach. You can look up details about it on the web, though it doesn't appear to be in Wikipedia. Dr. Peter Honeyman was somehow involved with the project, so the stories I've heard about it have been from him. They did a lot of paper studies about what to do, and had a non-massive toy machine to do experiments on. It was a VAX with 128MB of RAM - it needed 10 cabinets to store it all, and had a 450MB Fujitsu Eagle drive just for swap. (Back then, my VAX was pretty large with 4MB, and we would have been happy to have an Eagle drive instead of some of the slower 256MBMB RM05 removables we had.) A lot of the research was looking at things like MMAP, persistent objects, and that sort of thing.
Now, of course, it's hard to get current memory as small as 128MB, and the $20 flash drives at my local pharmacy are bigger than the disk drives on my VAX were.
This solution isn't at all useful for your laptop - burns way too much power. What your laptop needs for operating systems is a flash drive, since that doesn't need power to maintain stored data, and it'll use a lot less power than mechanical disks as well as not having rotational or seek delays. Modern flash has gotten past the problem of limited write cycles, and wear levelling takes care of the limited extent that it still has that, and for OSs, read-mostly data, or data that you're not changing a lot (like that copy of the movie) it doesn't happen much anyway. Also, even if you wanted to pick a worst-case application, like swap, a $20 4GB ram stick that lasts a year is still a great choice.
Vista supposedly knows how to make more intelligent use of flash as a middle-speed storage tier; good thing, given that it's apparently a memory hog. I don't know if Linux has explicit plans for it, but you could probably hack it to do some thing useful, but even just using one drive for the OS and another for swap could be a big help.
A few years ago I put 640MB of RAM on one of my lab servers just because I could. It didn't need it at the time, but it seemed appropriate:-)
My current notebook, about 2 years old, only has 512MB, because our corporate IT droids didn't think it needed more. Of course, IE didn't have tabs back then, and they weren't running Firefox as their browser. Now that RAM comes free with breakfast cereal (at least if you eat breakfast at Fry's) I suppose I should just go upgrade it myself, but I assume that if I do that they'll upgrade our machines before I've used up $50 worth of Firefox going faster.
I've really disliked Javascript since Netscape 2.0 or thereabouts. The problem isn't that you can't write perfectly good safe Javascript; it apparently works quite well for some things. The problem is that unlike Java, you can also write Evil Javascript, and you can also write Broken CPU-sucking Javascript. So if I leave Javascript turned on, because there are sites that require me to use their Javascript to do the things I want to do there, my browser's open to Evil Javascript on other sites, plus there's enough Bad Javascript out there that after I've done enough browsing, Firefox is burning most of my CPU on leftover broken cruft that I have to kill it.
Yes, I'm aware of NoScript and similar add-ons, and I'm happily using them. That helps, but there's still too much bad and ugly stuff out there to be happy about anything good that JS can do.
Handing bogus traffic to other people is rude at best, even if it hadn't occurred to you that somebody would register donotreply.com. And any traffic they're getting is either bogus traffic (because people didn't read the message that said to click the web link, not to reply) or autoreplies from robots.
Handing mail to example.com is more or less fine - originally there wasn't anything there, though the fine people at ICANN decided to put an explanatory web page there; AFAICT, telnet example.com 25 times out. And "invalid"'s even better, since it NXDOMAINs, and you can use addresses like donotreply@really.donotreply.invalid.
But you can also manage it yourself - use a subdomain like donotreply.mydomain.com, with some appropriate treatment like NXDOMAIN or a stub email server that replies "554 we told you donotreply, please use the URL in our email" or points to 127.0.0.86 or whatever. That way it's obvious who;s managing it.
Of course, if you're using donotreply.com because you're a spammer, none of these explanations matter to you, because you're a rude nyeculturny thug who doesn't mind bothering people. And some fraction of the people who reply to those will be including their credit card numbers, mother's maiden name, and postal address, so that they can collect the Microsoft Lottery or order their Nigerian Herbal Fake Viagra, and well, more power to the folks at donotreply.com for offering to educate those poor suckers:-)
It doesn't look as bad as the Cogent - Level3 de-peering incident a few years ago, but both sides have recovered from that one. Cogent's always been an interesting player, though some years they've looked kind of marginal. I first ran into them around 2001, when they were selling 100 Mbps Ethernet connections for about the same price other carriers charged for 1-2 T1s. They could afford to do this in part because they were selling to large multi-tenant buildings, so they could drop a fiber into the basement and connect to multiple customers with simple riser circuits, as opposed to having to run fiber individually to everybody. (Kind of like the Korean-apartment model, but for businesses.) Most of their customers that I talked to didn't think they could afford consistently deliver a full 100 Mbps for long periods, but they didn't care, since they could probably deliver at least 3 Mbps pretty much all the time, so anything above that was gravy. They're still in business, and seem to mostly sell to content providers.
The Internet *is* a co-op, as well as an idea for how everything should be connected together. There isn't any central backbone, and hasn't been for years; there are a bunch of large providers who connect together, a bunch of interconnection points, and a bunch of smaller providers and service providers who connect to either the big ISPs or the IXs or both. The big ISPs can make money (at least sometimes:-) because they're able to provide the value of connecting everybody together.
The question is whether they're willing to connect to each other for free, or whether they're going to charge each other money. In general, big carriers will interconnect for free (splitting the cost of the interconnection) if they're similar enough in size to hand each other relatively balanced traffic loads, or if they're playing in niche markets that complement each other. For instance, eyeball carriers like DSL and cable modem companies and big content providers like big hosting companies have an incentive to peer with each other, because the alternative is for both of them to pay a transit provider to interconnect them. But if they can't agree on terms, or can't make their connections work together, then they're not going to peer, and the fallback is that some of them may have to buy transit, either with the small ISP buying from the big one, or buying from some third ISP that connects to both of them.
Each ISP knows about the IP addresses of its own customers, and connects to other ISPs to exchange routing information. If two ISPs are peering, they're going to share address and route information for their own customers with the other ISP, so ISP A can reach ISP B's customers and vice versa. On the other hand, if ISP A is selling transit to ISP B, then A is going to tell B about all the addresses it knows how to reach, and how good the routes are, and B is going to send A packets for those addresses (and money.) In the general case, A knows how to reach every address on the Internet, either because the address is directly connected, or because A peers with that address's ISP, or because A pays for transit from some other ISP that knows how to reach it. (There are also exceptions and special cases, like national-monopoly ISPs.) And not everything's a pure case; one ISP might pay another to carry traffic for some routes but not others, or handle some traffic for free and pay for the rest.
I don't know quite what happened with Telia and Cogent here. Cogent mostly sells to content providers in the US and Europe; Telia's a more general ISP but I get the impression their customers tend to be end users and eyeball handlers. Cogent's side of the story seems to be that Telia's not maintaining their peering links correctly, so they de-peered with them and stopped exchanging traffic directly. If Telia's buying transit from some other ISP, that should let Telia's customers reach Cogent's customers. If Cogent's blocked that traffic now, that's weird; carriers don't usually do that on purpose.
This is sort of the opposite of the Cogent-Level3 fight of a few years ago. During that even, Level 3 decided that Cogent was no longer sufficiently useful to peer with, and dropped peering, which would force Cogent to either pay money to L3 to get the service, or else pay some third ISP for transit. This time it's Cogent dropping the other carrier.
It's been a while since I've lived back East. I don't remember highways in PA getting actually blocked, but there was always either construction going on which slowed you down, or roads that badly needed repair, so the potholes slowed you down. This is like having a barrier across the entire road and forcing you to drive down to Philly or up to Port Jervis.
According to Abdul Alhazred's books, we should really make sure not to fly them over that city with all the pyramids; there are Elder Things that not only was mankind not intended to know, but which get really annoyed if you try...
First of all, it reduces spam because almost nobody's actually bothering to encrypt their email at the client level, though there's a lot of encryption happening on the transport level (TLS on submission links or between SMTP servers, etc.) So spammers aren't going to bother when there are so many easier targets, but neither are your friends and customers.
But it also reduces spam because it takes CPU work to encrypt email, and spammers are generally not going to bother with that. 5-10 years ago, it was _enough_ CPU work to make spamming non-scalable, but with the advent of botnets, that's mostly changed; computers have gotten a lot faster and spammers are using other people's computers instead of having to burn their own CPUs.
How encryption increases spam is that it means that your mail server can no longer run content-based scanning on your incoming email - you'll have to decrypt it first, and then have your mail client run a filter on it. I'm not aware of any mail clients that do that, though at least some of them let you type in your passphrase once and apply it to all incoming messages.
There are systems like some of the corporate PGP stuff that do most of the encryption at the mail server level rather than the mail client, and maybe some of them can help with that.
Mask laws in the US were mostly originally written to deal with the KKK, though some may have been written about masked robbers and similar criminals.
But they have occasionally been used to harass Muslim women; I think the case I saw in the press a few years ago was in or near Detroit (not surprising, since that's a heavily Muslim-immigrant area.) Of course, police don't hassle people for wearing ski masks there in the winter time...
Zen's an appealing form of Buddhism - pure, simple, difficult, uncluttered; if I were a Buddhist that's probably the form I'd pursue. But Buddhism's much broader than that, picking up all sorts of local cultures and pre-Buddhist religions and random other stuff along the way. Tibetan Buddhism incorporates a lot of Tibetan Bon religion, with all kinds of scary demons, mountain spirits, prayer flags, and the like. The Pure Land Buddhists worship Amitbha Buddha, also called Amida, hoping to enter the Pure Land in the next life as a result of their devotions; you'll see Jodo and Hongwanji missions spreading that. In a rather opposite direction, there are the Nichiren Shoshu people who chant their Nam-Myoho-Renge-Kyo to get their wishes granted. I'm inclined to suspect that Zen is more austere than what the Buddha himself believed...
The Crash-1 only does 40 Gbps (OC768) as far as I can tell. This is Nortel gear; not sure what they're feeding it with, since at speeds like that you'd rather do switching, not routing.
There's been somewhat of a race in the industry between the people who think the next step after 10 Gbps should be 40 Gbps or 100 Gbps - 10 was a really convenient speed, because the Telecom/SONET part of the world does multiples of ~155 Gbps * 2**N, so OC192 is basically the same speed as 10GE and they can reuse many of the components and technology. 40 Gbps OC768 is fairly cutting-edge, but some carriers have been deploying it, while others have been waiting for 100GE, and it's easier to do a 4x generation technology change than 10x.
Also, several generations of Cisco gear have had names like that; I think the 12000 GSR was HFR?
Slashdot Mirror
Are the critical extensions available? For me, that's Adblock, NoScript, and Flashblock.
The big difference is that on the Internet, everything you read is true.
The other difference is on the Internet, nobody can tell that the government is a dog.
While personally I'd be more interested in seeing the Oscillation Overthruster, the fnords, and the white hole that'll be created after they fire the thing up, I'd say than an article by somebody who's actually seen the LHC really does deserve some Mod Informative points.
So how do you find workloads that can exploit many cores effectively? You can do servers, which can do just fine by sticking each client process/thread onto a single core, or reinvent database parallelism, or run virtual machines since nobody writes their applications securely enough to use Unix-style multiple users any more. Or you can find applications like graphics that parallelize well enough and use a bit of glue to hand of chunks of SIMD work to different cores. There are getting to be fewer and fewer CPU-bound problems that don't fit onto a single processor well. So maybe it's time to start moving more kinds of applications into the same chip - move the graphics off the GPU and back into the main processor, or have different types of cores for different applications.
If we can't find ways to do things like that, it's going to be annoying to chip-makers to have to find ways to make money as Moore's law makes high-speed CPUs cheaper and cheaper. There's a certain amount of Microsoft Bloatware(tm) and gamer framerate-lust that can keep them going, but basically desktop CPUs are already faster than most of us know what to do with. VmWare means that the fast multicore CPUs are getting exploited by people who use them to cut down on the number of servers they're buying by factors of 5-10, though the electricity savings means that you can sell a lot of server cores into the replacement market.
While you might not want to replace your entire disk infrastructure with RAM, there's an intermediate level which is to use Flash. The densities and price/GB are a good bit better than RAM, it doesn't need to stay powered to keep data, and you don't have millisecond-long delays from rotation or seeking, though the throughputs aren't as fast as RAM. The limited numbers of writes that old flash had have been improved a lot, and wear-leveling technology takes care of most of the rest. Operating systems aren't always really set up to use it (Vista supposedly has some capabilities), and you may end up just using it as swap space, but it's still theoretically a big win for some uses.
And we also had Teradata, which had specialized database machines, with up to about 500 disks, with a processor per disk and a master processor to split up workloads, typically spreading out an SQL query across a bunch of disks.
Parallel's always going to be with us, because it's an obvious technique to take whatever hardware's available that generation and build bigger machines by clustering a bunch of them together. The only real difference today is that the parallelism is increasingly moving onto the chips, because it's getting tougher to shrink chip features.
Sure, I was partly trolling (:-), but they are conservative, in the sense of being Establishment Radio. They're not right-wing ranters like Limbaugh (who was actually fun to listen to during the Clinton years), but they're generally supporting the Government, especially the Civil Service side of it, and while they're not out actively shilling for the Bush Administration, they pretty much let government-fed stories through with the default assumption that they're relatively accurate and framed honestly. They're happy to carry stories from the Democrat side of the Establishment (which is where most of the NPR folks come from) but they're too busy being "objective" to challenge obviously bogus propaganda for what it is.
If I wanted to do something high-tech and expensive for traffic info, I could probably get some kind of traffic-integrated-GPS-thing; in a few cities there's also a low-tech cheap traffic widget that has a fixed LCD map and gets traffic by subscription for $5-10/month on some kind of radio channel. It's not as detailed about individual events as the every-10-minutes traffic radio, but it covers more of the highways and you don't have to wait for reports.
While they're downloading and rendering, I've got no problem with Firefox needing to use the CPU. But five minutes later, or an hour later, they shouldn't be using CPU time to maintain that. And especially after you've closed the tabs, they shouldn't be burning CPU time.
If you want to say that Passover's date is set at spring pagan holiday time, you'll need to argue with your rabbi or maybe Lehrhaus Judaica about whether your druids are at all the same kinds of pagans as Caananites were. And if you want to say that the name "Easter" and the bunnies and and eggs and marshmallow chickens are ripped off from Germanic spring fertility goddess stuff, you'll have a tough time getting anybody to argue the other side except maybe some atheists who'll say that the Germanic fertility goddess folks ripped that off from nature, which provided the bunnies and eggs, or from the chemical industry who brought us marshmallow peeps.
That's different from Java... how? That's different from Java because Java has a security model to prevent this, and most implementations do a pretty good job of not breaking it. Occasionally there's been an implementation bug, which has gotten fixed, but AFAIK (as of a few years ago, when I last paid attention to Java), the security model itself was still solid. No surprise, because that was one of Gosling's objectives in writing it. Javascript has bits of security tacked on the side, but that's about it.
Under WinXP, open the Task Manager to show CPU and memory consumption, minimize it, take Firefox 2.x to fark.com, open all the news article links in tabs, then close all of them (but leave the main page open.) Most of the time you'll not only have well over 100MB of RAM still in use, but you'll have your CPU smoking as well.
Now, of course, it's hard to get current memory as small as 128MB, and the $20 flash drives at my local pharmacy are bigger than the disk drives on my VAX were.
Vista supposedly knows how to make more intelligent use of flash as a middle-speed storage tier; good thing, given that it's apparently a memory hog. I don't know if Linux has explicit plans for it, but you could probably hack it to do some thing useful, but even just using one drive for the OS and another for swap could be a big help.
My current notebook, about 2 years old, only has 512MB, because our corporate IT droids didn't think it needed more. Of course, IE didn't have tabs back then, and they weren't running Firefox as their browser. Now that RAM comes free with breakfast cereal (at least if you eat breakfast at Fry's) I suppose I should just go upgrade it myself, but I assume that if I do that they'll upgrade our machines before I've used up $50 worth of Firefox going faster.
Yes, I'm aware of NoScript and similar add-ons, and I'm happily using them. That helps, but there's still too much bad and ugly stuff out there to be happy about anything good that JS can do.
Handing mail to example.com is more or less fine - originally there wasn't anything there, though the fine people at ICANN decided to put an explanatory web page there; AFAICT, telnet example.com 25 times out. And "invalid"'s even better, since it NXDOMAINs, and you can use addresses like donotreply@really.donotreply.invalid.
But you can also manage it yourself - use a subdomain like donotreply.mydomain.com, with some appropriate treatment like NXDOMAIN or a stub email server that replies "554 we told you donotreply, please use the URL in our email" or points to 127.0.0.86 or whatever. That way it's obvious who;s managing it.
Of course, if you're using donotreply.com because you're a spammer, none of these explanations matter to you, because you're a rude nyeculturny thug who doesn't mind bothering people. And some fraction of the people who reply to those will be including their credit card numbers, mother's maiden name, and postal address, so that they can collect the Microsoft Lottery or order their Nigerian Herbal Fake Viagra, and well, more power to the folks at donotreply.com for offering to educate those poor suckers
It doesn't look as bad as the Cogent - Level3 de-peering incident a few years ago, but both sides have recovered from that one. Cogent's always been an interesting player, though some years they've looked kind of marginal. I first ran into them around 2001, when they were selling 100 Mbps Ethernet connections for about the same price other carriers charged for 1-2 T1s. They could afford to do this in part because they were selling to large multi-tenant buildings, so they could drop a fiber into the basement and connect to multiple customers with simple riser circuits, as opposed to having to run fiber individually to everybody. (Kind of like the Korean-apartment model, but for businesses.) Most of their customers that I talked to didn't think they could afford consistently deliver a full 100 Mbps for long periods, but they didn't care, since they could probably deliver at least 3 Mbps pretty much all the time, so anything above that was gravy. They're still in business, and seem to mostly sell to content providers.
The question is whether they're willing to connect to each other for free, or whether they're going to charge each other money. In general, big carriers will interconnect for free (splitting the cost of the interconnection) if they're similar enough in size to hand each other relatively balanced traffic loads, or if they're playing in niche markets that complement each other. For instance, eyeball carriers like DSL and cable modem companies and big content providers like big hosting companies have an incentive to peer with each other, because the alternative is for both of them to pay a transit provider to interconnect them. But if they can't agree on terms, or can't make their connections work together, then they're not going to peer, and the fallback is that some of them may have to buy transit, either with the small ISP buying from the big one, or buying from some third ISP that connects to both of them.
Each ISP knows about the IP addresses of its own customers, and connects to other ISPs to exchange routing information. If two ISPs are peering, they're going to share address and route information for their own customers with the other ISP, so ISP A can reach ISP B's customers and vice versa. On the other hand, if ISP A is selling transit to ISP B, then A is going to tell B about all the addresses it knows how to reach, and how good the routes are, and B is going to send A packets for those addresses (and money.) In the general case, A knows how to reach every address on the Internet, either because the address is directly connected, or because A peers with that address's ISP, or because A pays for transit from some other ISP that knows how to reach it. (There are also exceptions and special cases, like national-monopoly ISPs.) And not everything's a pure case; one ISP might pay another to carry traffic for some routes but not others, or handle some traffic for free and pay for the rest.
I don't know quite what happened with Telia and Cogent here. Cogent mostly sells to content providers in the US and Europe; Telia's a more general ISP but I get the impression their customers tend to be end users and eyeball handlers. Cogent's side of the story seems to be that Telia's not maintaining their peering links correctly, so they de-peered with them and stopped exchanging traffic directly. If Telia's buying transit from some other ISP, that should let Telia's customers reach Cogent's customers. If Cogent's blocked that traffic now, that's weird; carriers don't usually do that on purpose.
This is sort of the opposite of the Cogent-Level3 fight of a few years ago. During that even, Level 3 decided that Cogent was no longer sufficiently useful to peer with, and dropped peering, which would force Cogent to either pay money to L3 to get the service, or else pay some third ISP for transit. This time it's Cogent dropping the other carrier.
It's been a while since I've lived back East. I don't remember highways in PA getting actually blocked, but there was always either construction going on which slowed you down, or roads that badly needed repair, so the potholes slowed you down. This is like having a barrier across the entire road and forcing you to drive down to Philly or up to Port Jervis.
According to Abdul Alhazred's books, we should really make sure not to fly them over that city with all the pyramids; there are Elder Things that not only was mankind not intended to know, but which get really annoyed if you try...
But it also reduces spam because it takes CPU work to encrypt email, and spammers are generally not going to bother with that. 5-10 years ago, it was _enough_ CPU work to make spamming non-scalable, but with the advent of botnets, that's mostly changed; computers have gotten a lot faster and spammers are using other people's computers instead of having to burn their own CPUs.
How encryption increases spam is that it means that your mail server can no longer run content-based scanning on your incoming email - you'll have to decrypt it first, and then have your mail client run a filter on it. I'm not aware of any mail clients that do that, though at least some of them let you type in your passphrase once and apply it to all incoming messages.
There are systems like some of the corporate PGP stuff that do most of the encryption at the mail server level rather than the mail client, and maybe some of them can help with that.
But they have occasionally been used to harass Muslim women; I think the case I saw in the press a few years ago was in or near Detroit (not surprising, since that's a heavily Muslim-immigrant area.) Of course, police don't hassle people for wearing ski masks there in the winter time...
There's been somewhat of a race in the industry between the people who think the next step after 10 Gbps should be 40 Gbps or 100 Gbps - 10 was a really convenient speed, because the Telecom/SONET part of the world does multiples of ~155 Gbps * 2**N, so OC192 is basically the same speed as 10GE and they can reuse many of the components and technology. 40 Gbps OC768 is fairly cutting-edge, but some carriers have been deploying it, while others have been waiting for 100GE, and it's easier to do a 4x generation technology change than 10x.
Also, several generations of Cisco gear have had names like that; I think the 12000 GSR was HFR?