eck

I once found this bug, where a destructor was taking forever:

    hash_set<char*> hash;
    // <fill up/use hash>
    while(!hash.empty()) {
      free(*hash.begin());
      hash.erase(hash.begin());
    }

hash tables are not meant to be used like that.

> the risk of an additional failure in say a RAID5 or RAID6 configuration is just too high.

Well there's one issue that another whole drive will fail and you're screwed.

The other issue is that modern disks have an unrecoverable read error rate compared to their size such that a total cover-to-cover read -- necessary on every remaining disk to rebuild a RAID5 -- is kinda unreliable, even with a supposedly healthy disk.

> Don't model rockets routinely go much higher than the 400 feet the FAA has traditionally limited hobby aircraft to?

Even if they do, they're mostly cardboard, whereas model airplanes have things like small engine blocks. I bet a 737 engine could ingest a model rocket without anyone noticing.

> I want to send you XX mbps and you want to receive it. You pay Comcast. Am I now entitled to a free connection?

Yes, provided you deliver it to the designated IX for my location and pay whatever applicable meet-me-room fees. Realistically only CDNs are going to do that for the O(100s) of locations in the country.

The alternative is to "negotiate" with Comcast. Since your BATNA is losing access to a huge fraction of Internet subscribers, Comcast is going to calculate your business's profit per megabyte and demand 90% of that. They know what Netflix costs. They know how much bandwidth the average Netflix customer uses.

A peer connection in the same city as the subscriber is literally the cheapest traffic for Comcast's network to handle. If their rates don't cover bandwidth to that, they should raise their rates.

My point is that they ban explicit prioritization yet ISPs can achieve the exact same result through network topology. It's a huge loophole.

Put another way, if I pay Comcast for XX mbps, it should be possible for some company to send me data at XX mbps without paying Comcast. Otherwise what exactly am I paying for?

One solution would be to force last-mile ISPs to designate a location for each customer (for example "all Manhattan subscribers => 60 Hudson St") where Netflix etc could peer without settlement only with those customers. Comcast doesn't have to carry Netflix's traffic across the country, just from the nearest IX to the customer. If the customer monthly fees don't cover that, what exactly are they paying for? Any VOD offerings from Comcast would be required to routed through the same points, without routing preference over 3rd party traffic like Netflix.

In some sense, that would be a light version of forced cable leasing, since instead of movies, a service could offer a vpn/tunnel to general internet access with better transit provisioning than the last-mile ISP.

> 30. But this Order does not apply the open Internet rules to interconnection.

TLDR: Relax, Comcast. You can still intentionally let all your transit ports saturate every evening so Netflix will blow unless they pay you to peer. Which is what you really wanted to do.

> out of compliance with their peering rules

I'm not a networking expert so someone please correct me if I'm wrong here, but those peering terms also specify shortest-exit routing, in which case the 1.8:1 traffic ratio is necessary to ensure that one peer does not do dramatically more "work" in terms of bit-miles.

It seems like Netflix and a consumer ISP are never going to be able to peer with traffic ratio requirements. If the consumer ISP offered best-exit routed settlement-free peering options, though, Netflix could easily comply.

It is my impression that a consumer ISPs refusing to offer settlement-free peering with a best-exit routing policy is apparently charging for access to its customers, not charging for access to its network.

Sounds more expensive than the ransom.

What I wondered about that data set is, if two people living/working at two locations, consistently take taxis to meet at various other locations at the same times, could that pattern be identified in the data?

That is, are there locations A and B such that there are matching trips to locations M1,M2,... at times T1,T2,... i.e. (A,M1,T1),(B,M1,T1),(A,M2,T2),(B,M2,T2) and perhaps reverse trips (M1,A,T1+x) etc?

Further classification of M* -- hotels, for example -- could classify the nature of the meetings. You might be able to identify the addresses of people having affairs, or other deliberately secret rendezvous.

It needs a whitelist of giant corporations like Netflix that acquiesce to Comcast's fees. Images should only load slowly on poor websites.

I'm intrigued by the combination of last-mile ISP and neighborhood social networks. It seems that if you were organizing a Google Fiber "fiberhood", or municipal ISP, or some sort of cooperative ISP, these sorts of local social networks are the ideal organizing tool. My Facebook friends all whine about neutrality but are scattered all over the world. My Nextdoor neighbors are largely within a box of CAT5e from each other.

The bit about selling her soul seemed a bit contrived. They don't have to be the applicants' stories, but they don't have to be the ghostwriter's story either -- once it's no longer a true story for the applicant, whether it's your story or total fiction seems irrelevant.

The part about interviewing the applicants made me wonder whether there is any fact checking of essays, or whether there will be in the future. The story about being poor and having your clothes stolen from a laundromat is a real tearjerker, but if the admissions officer knows that you're the child of China's 99th richest business magnate, I doubt it would help.

> not primarily focused on just delivering books or groceries but being put to the use for AID

That is just their plan for getting regulatory approval: if you are against UAVs, you are against delivering life-saving medicine to disaster victims.

As a customer of both Netflix and an ISP that participates in Open Connect, I'm slightly annoyed that Netflix didn't turn around and immediately pass this charge on to their users that connect from Comcast. Why should I have to subsidize that? I wonder if their agreement forbids that.

> Panels commonly run at 12V (or some low multiple thereof),

That is absolutely not true. Generally speaking nontrivial sized photovoltaic systems are designed with panels in series such that the voltage stays just barely within the 600v rating on the wire.

For example, grid tie inverter, MPPT rated 195-550v:

http://pdf.wholesalesolar.com/inverter%20pdf%20folder/Schnei...

Could Aereo pivot to selling a physical device that one would plug into a home router and do roughly the same thing?

> Why would Netflix pay more for single network peering than full transit?

Because the interconnection between your transit provider and Comcast is frequently saturated and Comcast is not going to fix it since they would prefer to be paid.

It is kind of amusing how similar this is to the network neutrality debate -- i.e. comcast's dispute with netflix affecting comcast/netflix customer content delivery -- yet with decidedly old-fashioned media.

Incompetence was perhaps too harsh. The outcome was unforeseeable.

> Like when a burglar hits a small countryside town, and suddenly everyone has to start locking their doors.

I think your analogy just proves my point. They caught the burglar, but he turned out to be the local kid that everyone kind of liked. He was punished as everyone agreed the law required, but nobody was happy about it. And it would have all never happened had they just locked their doors to begin with.

The small countryside town with no crime and the open MIT environment were always illusions. It was inevitable that sooner or later someone would try to steal. You can either prevent it from happening or punish it after it does. [edit] But you can't choose not to prevent it then claim you didn't want to punish it.

The Swartz/MIT debacle reminds me of a chapter in Starship Troopers. A mobile infantry recruit punches his drill instructor. The recruit is publicly lashed and discharged.

Later in private, the commander reprimands the drill instructor: there were standing orders to never let a recruit hit you. They obviously have to punish recruits who strike instructors, but they do not want to. The drill instructors are supposed to be elite soldiers, and this one let a recruit get the drop on him, and that mistake cost the infantry a recruit.

MIT had sufficient opportunity to stop Swartz with technical countermeasures -- but they failed to do so. They allowed the situation to escalate until official punishment became unavoidable. MIT may claim that they did not want to punish Swartz; few people will argue that they did. It was not their intent but rather their incompetence that lead to his punishment.

The ADXL78 series [1] which is recommended for vehicle collision sensing draws 1.3 mA. If it were the only load, it would take on the order of ten years to discharge a car battery, which is well below the self-discharge limitations. I don't know if lower power sensors are available, or if some rule like "enabled for 24 hours after the car is run" would be appropriate, but lack of power for the sensor does not seem like a good reason for the airbags to be disabled.

Bluetooth devices actively searching for devices to pair with could plausibly draw significantly more power than that.

[1] http://www.analog.com/static/imported-files/data_sheets/ADXL...

This concept is often referred to as "homes with tails". The idea is that if everyone controlled a line from their house to a central facility (kind of like a small meet-me room) then at those facilities there could be significant competition that would drive down prices.

http://googlepublicpolicy.blogspot.com/2008/11/homes-with-ta...

> It's interesting that a program which the NSA's director claimed to congress didn't even exist, which was created by the executive branch requires the legislative branch to terminate it.

IIRC, the Guantanamo terrorist prison program was also created without congressional action, yet its closure was successfully prevented by Congress -- or at least, congressional opposition to closing it was sufficient for Obama to escape blame for failing to close it as he had promised during the election.

So, this could just be a plan to transfer blame for the NSA onto Congress. If he proposes a bill to reign in the NSA that he knows Congress will not pass, it maintains the status quo but he can say that he wanted to end it.

What are the rules for new tile placement? Is it deterministic, random, or adversarial?

I worry that the rules will forbid ISPs from having explicit whitelists/blacklists, but will do nothing to prevent them from having undersized and saturated transit links then demand payment for peering with content providers like Netflix.

What would happen if consumer ISPs were required, for each customer, to choose a single point (at the metro level, probably) where they are required to peer openly? That allow all Netflix-like companies access to customers, but since it's a point local to the customer, would not impose any long-distance bandwidth requirement on the ISP.

> However, it's hard to imagine that we wouldn't be more secure hiring educated behaviorists to have a quick chat with passengers before they board a plane.

"quick chat"?

Would this would be a "quick chat" where they can ask anything they want, lying is a felony, and choosing not to answer would get you locked in a back room for "secondary screening" until ten minutes after your plane leaves?

So, right now, if someone detonates a bomb with a cell phone, Verizon/AT&T is basically a co-conspirator? You'd think they'd want to be a common carrier, at least if it weren't for the US's general lack of enthusiasm for holding corporations responsible for anything.

He's talking about desktop/server CPUs where people care about performance. If you don't care so much about performance, you can increase the transistor sizes, reduce the clock speed, and achieve totally insane MTBF... as space-rated hardware tends to do. Kind of like how server CPUs are underclocked to increase MTBF, but more so.

No, you don't. In fact, you couldn't even tell an app that it doesn't have permission, because then the stupid flashlight app will do something like, "let me see your location, otherwise i can only be 50% bright!" Obviously the writers of the app want that info.

Instead, it thinks it has a GPS, but you've decided to tell it that, for example, its "GPS" will always be e.g. times square or perhaps something a bit more realistic, but not real.

If someone is willing to divulge inside knowledge of his last employer you have to assume that in the future he will be willing to divulge your inside knowledge.