Forgot your password?
typodupeerror

Comment Re:Just get used to it (Score 4, Informative) 53

We had an article here about power grids pulsing due they suspect to large numbers UPS switching in and other systems leaving sleep states on the hour running cron jobs and the like all synced very to very precise time via the network.

Most software testing has focused on 'correctness' and but 'emergent behavior' and specifically how small seemingly innocuous changes to software will manifest not request volume / batch size scale but 1000s separate servers scale. Microsoft leaned about this last week with patches that got it wrong on the single system scale worked well for handfuls of CPU cores not so well on systems with 10s or 100s of concurrent thread support, and at "hyper-scale" where fail-over and back-off behavior lead things to cascade.

However how do you test that, even for Microsoft the economics having a few extra data centers fully kitted out with their latest hardware platforms for 'testing' is tough to swallow and even if they had that how do they simulate the traffic? You can do experiments like 'well if we really light up this load balancer, we should see it shed requests to the others...' but that does not tell you how all those client workloads are going to react.

Nor can you lab this out easily either because you don't have as much real concurrency, even if you had VM representing every physical machine you own.. you're going to be serializing things in a way that don't happen in the real world where there are truly independent systems running.

For all the testing you can do, at that scale humans or maybe AI can really only pontificate about what might happen what the odd failure modes could be..We just can't know. The thing to remember though is even if you build you own separate systems you are going to have failures and downtime, for most of us more failures and more downtime than "the cloud" but you get a known worse case, that being completely rebuilding should be no worse than the initial build. There is probably some serious soul search to be done about just how tightly coupled and where information systems really belong. Now none of that really happen of course because that might cost real money..

Comment WAT? (Score 4, Insightful) 53

I have not idea what he is even say.

Sure if there is only capital and no labor you don't have an economy. I guess that much is true capital lacks agency it just sits there unless labor, even if the laborer is its owner decides to make it do something. - fine

But all that stuff about the model can be anything ... sounds like he is on cocaine, every bit as wireheaded as anyone at anthropic.

Comment Re:Should have... (Score 1) 65

This, I don't think anyone should feel to bad for him right now. He is the one did this to himself, even if others might have intentionally dangled temptation in front of him.

He is already rich enough that neither he nor his dependents will ever want for anything, even if he never finds another job anywhere. Boohoo he'll have to find a hobby.

Comment Re:This was a hit job to kill opensource models (Score 1) 65

Yeah, well they did not make him cheat on his wife, and they could not have made him shtup his subordinates.

He did decided to be stupid and blew up his life. The lesson in this the Billy Graham Rule, always and with no exceptions. Once you are married and/or have achieved a positive net-worth or any degree of social importance, whichever comes first - Never ever be alone with a member of the opposite sex unless they are your wife, parent or possibly a sibling and in the later two cases use your judgement as to if that is even safe in your specific situation.

Comment Re:Economics of Starlink (Score 1) 104

The cost per bandwidth for satellite is declining exponentially. Fiber optic deployments cost real money and take real time, the costs not improving. At the moment satellites are outpacing fiber, cable and wifi deployments by a radical amount - the entire planet was covered with no dead zones in just a few years. At some point the cost of providing satellite Internet is less than the cost of maintaining the wires in the ground. Along the way bandwidth increases everywhere, consumer costs go down.

Comment Re:Economics of Starlink (Score 5, Informative) 104

EBITDA margins on Starlink are 63% already. Huge moneymaker, which is becoming a problem for NASA since the revenues already exceed NASA's entire budget and escalating exponentially. NASA is becoming small potatoes to their only functional launch provider. The 26 satellites launched this time each carry 10x the bandwidth of previous versions so equivalent to 260 of the old birds and 10 Falcon 9 launches. The new bigger sats weigh 1/2 ton and required the bigger rocket. A fully recovered Starship launch costs about $4M for fuel, an indeterminate amount for refurbishment but nowhere near 10x the internal cost of Falcon 9 launches of $15M. The satellites themselves are mass produced by SpaceX in a factory and all-in only about $500K. The economics of Starlink are sound.

Comment Re:I just don't see the point (Score 1) 43

Honestly I think that is best part of this; you can play with the bits that actually interest you and let AI filling the gaps and/or do the parts that are not your bag.

If you are into retro gaming or something and you want to run win95 on contemporary VM platform with a display driver that is better than basic SVGA/Vesa; have AI whip you up a driver. If you are into legacy Windows internals sure it is playing with assembled legos but most people are not, understating legacy Windows internals isn't very applicable skill, I don't know why someone wouldnt skip that that part if what they are really after is playing some game from 1996.

On the other hand if you like systems programing and enjoy a challenge you can go on archive.org and get yourself on old WindowsSDK and see what you can make happen, then you could have AI write some a test suite so you can find out just how correct you really got it all; rather than firing up Quake and going 'guess it works'.

Comment The booster fell short. (Score 5, Interesting) 92

It uses two BE-4 engines from Blue Origin. The same engine that on Blue's New Glenn rocket destroyed the rocket and LC36 launch facility in May. Max payload to LEO for Vulcan is 23 tons. For New Glenn, 50 tons. Vulcan is designed fully expendable. New Glenn first stage reusable.

Things aren't going well for either launch provider. They can't launch at all now for months and then they have to prove their rockets can make orbit before they take paying cargo. Several proof of flight trips will be required for national security launches. Amazon doesn't have a route into orbit for their LEO Internet constellation before the FCC starts getting pushy. SpaceX will probably be forced to take some national security launches on their half-reusable Falcon rocket, but they want to retire it and have quit accepting commercial flights on it.

Meanwhile SpaceX landed their 52m fully reusable space freighter back on the tower it launched from, just the day before, having lifted 26 of their latest satellites into a nominal orbit. 54 metric tons of payload, and the rocket back. SpaceX did ditch the booster in the gulf but that was an easily correctable flight profile issue (low fuel) not a vehicle functionality problem. Once orbital refuelling is proven Starship will be able to target most of the Solar system.

If Blue gets a working rocket that would be amazing for US spaceflight. But neither company has any hope of being competitive on price or cadence with their partially or fully expendable rocket.

Slashdot Top Deals

This file will self-destruct in five minutes.

Working...