and found that they got better mileage with the "unnecessarily complex dual drive train" than they did with the charging-only engine.

1) It's not about how much you eat, but how much your body converts to fat. I went on Atkins and lost quite a bit of weight while still eating lots of calories.

2) If you severely cut down on calories, your body can become more efficient at using the calories it does eat.

So the trick is to convince your body that it's not hungry, doesn't need to be efficient, and doesn't need to store fat against future needs.

They're mostly empty calories.

The Mellanox driver code I saw was pretty messy, especially the SR-IOV support. And their device model is sort of weird for anyone used to "traditional" ethernet hardware. But they're really the only game in town for 40gig.

In contrast, the Intel 10gig drivers are reasonably clean, follow the same device model as their other ethernet hardware, and their datasheets are available for anyone to download.

For exactly the reason that you describe, a kernel driver that is intended to be built on many different kernel versions really shouldn't be making assumptions based on the version number, it should test for the actual desired functionality.

The simplest solution is:

1) ISPs apply traffic shaping to each subscriber separately, without looking at packet type, source, destination, etc. The only criteria are which subscriber the packet belongs to, and what level of subscription they've paid for.

2) As an optional step (opt-in or opt-out) the ISP can do QoS within the packets belonging to a particular subscriber. This would only affect that subscriber, nobody else. Ideally this would be under the control of the end-user in some way, via ToS packets, classification rules, etc.

To me, Net Neutrality means that all traffic (regardless of far end *or* type) should be treated equally.

The only fair way to allocate resources on a subscriber network is by doing traffic shaping based on the subscriber plan, *without looking at traffic type*.

Suppose we've both paid for an identical subscription. I use my entire bandwidth for streaming video and torrenting, you use your entire bandwidth for videoconferencing. Traditional QoS would give your packets priority over mine. Since we're paying the same, that makes no sense!

The ISP should shape both our streams based on our subscriber plans. As an optional step they could apply QoS to the traffic belonging to each individual subscriber, but that would only affect the traffic for that specific subscriber.

Why should your videoconferencing packets get priority over my netflix stream?

If we've both paid for equivalent plans, then we should have equal use of the network.

The only truly fair option is for ISPs to weight traffic between subscribers based on their plans, without looking at traffic type. Then within the traffic belonging to a single subscriber they could (if approved by the subscriber) do QoS based on traffic type.

The ISPs should be legally limited to two types of traffic shaping:
1) Based purely on subscriber plan, without looking at traffic type. If you've paid for a better plan then me, your traffic gets weighted more heavily.
2) Optionally (if the subscriber requests it) they could shape based on traffic type, but only within the packets belonging to a specific subscriber.

That way, your web pages would get exactly the same priority as their video stream.

The ISPs should be legally limited to two types of traffic shaping:
1) Based purely on subscriber plan, without looking at traffic type. If you've paid for a better plan then me, your traffic gets weighted more heavily.
2) Optionally (if the subscriber requests it) they could shape based on traffic type, but only within the packets belonging to a specific subscriber.

That way, if we have equivalent plans then your torrent packets and my VoIP packets get exactly the same treatment, but my VoIP packets get priority over my torrent packets.

If you own .0000004%, and the company owes amount X and doesn't have enough money to cover it, then I see nothing wrong with you as an owner being sued for .0000004% of the outstanding debt if the company goes bankrupt.

Suppose you stepped on the brake and the car messed up and triggered the accelerator. I think the natural tendency would be to think that you had accidentally stepped on the accelerator, lift up on your foot, realize it was still accelerating, then try to brake--by which time you might have hit something.

There was an interesting article a while back about designing for robustness in vehicle ECUs. Things like putting variables at the bottom of memory so that a stack trampler would be less likely to overwrite them. Can't remember where it was now, but it was a good read.

And lets be real in our comparisons. The 35W Haswell is going to destroy your P8700.

The i5-4202Y still beats the P8700 handily in cpu benchmarks and has a max TDP of 11.5W, including memory controller and graphics.

If you start with the assumption that we need 5th gen aircraft then the F35 is the only option. But I don't think it's valid to start with that assumption--we should start with the missions that need to be fulfilled.

I think an argument could be made that it would make more sense to buy fewer F35s and buy a larger fleet of dedicated aircraft for close ground support, long-range recon, etc.

The Canadian gov't is trying to use the F35 for many different roles. It may be a fantastic air superiority fighter, but I'm not convinced it'd be better than something like an A-10 for ground support.