For all of its secrecy, "Fogbank" is probably just some sort of polyurethane foam (my guess is that the "extremely flammable and explosive" solvent is an ether; polyols made from ethers can be combined with isocyanates to form polyurethanes). The foam is probably just as an insulator and space-filling material for the warhead components, and there are any number of products which would probably work just as well in the Trident warhead, but only one has been actually tested in a detonation of the device.
Um, no. Plain low-Z foams are well documented in a number of nuclear weapons - are perfectly straighforwards to make, use, and test.
Fogbank's described properties pretty much only match an aerogel of some sort with suspended high-Z material, and extensive (non-public) analysis by a physicist with inertial confinement fusion experience indicates that Fogbank could be extremely critical to the operation of highly compact thermonuclear secondaries (the second, fusion implosion stage after the primary fission fires). Very thin radiation cases and energy buffered into the aerogel's high-Z constituent apparently allow you to effectively push the secondary without having a thick (heavy) radiation case to contain the primary's energy for much longer.
There are a number of weapons that didn't have that level of compact secondary still in use - B61s, B83s, W62s, etc.
However, a number of the very tight tolerances secondaries in use - W76, W87, W88, possibly W80 and the other B61 derivatives with stepped radiation cases, possibly W89 and RRW derivatives, probably don't work without the suspended high-Z aerogel material.
Could we redesign them with thicker radiation cases instead? Sure. Add... 20% perhaps to overall weight. Oh, and we'd have to withdraw from the nuclear test ban treaty and the threshold test ban treaty to test the redesigned weapons, because that redesign is NOT a minor issue with reliability, it's a fundamental physics/engineering change, even if the primary and secondary are the same. It's changing the dynamics of the energy capture from the primary and the timing and intensity of the energy pulse delivered to the secondary, in a radical manner. So, you need to test it.
Or we could go back to earlier, heavier designs, like the B61 and B83s. Except that all our current ICBM warheads appear to use Fogbank now. Oops. Error. Try again.
The Nvidia GPUs are large MIMD vector machines - look at the specs, and what they're doing with CUDA. That they're mostly actually used to draw texels of monsters and walls and bullets in flight doesn't mean that they're not a highly capable general purpose vector processor...
Many people are (almost certainly correctly) stating that Nvidia wouldn't do that if they didn't think they had to, or didn't think that this would make them more money / market penetration than not doing it. Suggesting that sticking a small x86 on the corner of their big graphics chips is somehow an architectural black hole is silly - they're not dumb, they're chasing their market. The CPU won't slow anything else down.
Better metrics to use (field-proven at a bunch of dot-coms you may have heard of) - these are rough numbers, but more precise ones require a detailed customer analysis and usually detailed log analysis after you start the actual business up live.
Assume "business day" of 10 hours - most businesses have a high point during the day that corresponds to this type of distribution.
Assume peak load in a given second will be 10x of that value.
So, if you expect 100,000 visitors a day (the story's corrected load), "business day average" will be 10,000 visitors per hour, about 2.7/sec. Assume 27/sec for true peak load.
Multiply by pages viewed per visitor per day - do they typically hit 2 pages, 8, 16? Depends on the content, usage model, etc. This gives you pages per second. A good conservative number is 10, which makes calculation easy. So that's peak page views of 270/sec.
Multiply by the number of hits per page based on number of images and other page components (navigation bars, graphics, frames, etc) - typical numbers on this range from 2.2 or so (straight HTML with a background image and a sitewide favicon) to 25 (complex multilayer navigation implimented with many graphic icons, etc). Again, assume 10 for a conservative rough order of magnitude estimate.
That gives you static hits per second. So something like 10x hits/page is 2,700 hits/sec peak rate.
If your pages are static, that's all and you're done. If they're dynamic, you have to consider the "app server" load (whether that's PHP or Tomcat or CGIs on the webserver, or back ending onto a different Tomcat or Websphere or Weblogic server or some such).
10 years ago, we built a large ecommerce website for a large bricks and mortar company from the ground up using that modeling. At that time, the right answer for the servers was Solaris/SPARC. At that time, the performance we got (on 400ish MHz CPUs) was 230 dynamic page generations per second per CPU using Weblogic as the app server, and around 2,000 static pages per second per CPU using the C2 Apache Stronghold and HTTPS. Our build spec was significantly more than 100,000 users - we used a Sun E10K cluster with separate domains on each system for web and app servers.
Obviously, faster modern CPUs do better, and the amount of memory and CPU power available in even cheap desktop or rackmount servers is plenty for the stated user load requirements on these servers.
You can screw that up by using the lousiest PHP or CGI scripts in the world, using excessively large pages or numbers of navigation images, etc. But even a site which is an order of magnitude slower than it should be will support 100,000 users on a handful of systems, if not on a handful of CPU cores with 2 or 3 boxes for redundancy.
So - Figure out what your static and dynamic content balance is, figure out a pair of "load balancers" out in front, put at least 2 servers at each level (either web, or web+app servers separately - if you track user accounts etc, then add a redundant database layer too). The cheapest servers you can get will be more powerful than you need for pure load considerations but you need to think about degraded mode performance when one falls down.
If you're hosting this at a hosting facility get the cheapest 1U rackmount servers you can. If you're hosting it in an office then small tower cases are marginally cheaper, but computer room space may be at a premium pushing you to racks anyways.
Set up a master server with the master copy and backup / older copies of the content and configurations. Distribute those out (rsync, cfengine, whatever) to the "live" servers. Put one server with large hard drives out in a completely different location and back up the files there nightly or more often, for disaster recovery.
In terms of clustering - you want to load balance the web and app servers. Probably don't want to cluster the database - easier to set up a slave DB server than true clustering and true DB clustering leads to many failure modes that non-experts often fall into, leaving them less reliable than master/slave plus manual failover options.
The general concept doesn't totally suck - one could build an offshore hotel on an oil platform, and some of the winning design's components look reasonable to me.
Rough seas and high winds at times, and the occasional hurricane, aren't impossible to work around at a resort (it's not like parts of Florida, the Caribbean, Mexico etc don't have those problems on land). It's going to need more evacuations and will have an operations impact, but you just have to plan and budget for that. You give people refunds / free future stays / etc if their visit is cancelled due to a storm, etc.
But you can't build something with a blatant blaring flaw, and letting something with a blatant blaring flaw win a design competition even if it was a neat concept doesn't make sense to me.
Whoever the designer of this was, has no experience with offshore platforms.
You NEVER put the helicopter platform on the side. It's never in clear air then, you can't safely approach the pad and land. And it was too close to hotel rooms anyways, where they have it, if the pilot goofs he flies the rotor into the rooms.
The China Airlines 747 was severely damaged and nearly had to be scrapped. Not due to supersonic flight loads, but due to damage from the high-G pullout required to recover from the out of control power dive towards the ocean.
Among other things the landing gear locks pulled out of their fuselage mounts and the gear extended partly during the dive pullout, damaging the gear and gear doors.
The pullout encountered 5.1 and 4.8 G peaks, which exceed the normal structural limits, and the aircraft's wings were permanently bent upwards 2-3 inches.
This is nothing new
on
Less Is Moore
·
· Score: 4, Interesting
Some of you may remember the 1980s and early 1990s, where PCs started out costing $5,000 and declined slowly to around $2,500 for name brand models.
Around 1995, CPUs exceeded the GUI requirements of all the apps then popular (this is pre-modern gaming, of course). Around 1996 and into 1997 the prices of PCs fell off a cliff, down to $1,000.
Actually, that's completely incorrect. (And shows how little you've studied or thought about the topic.) A nuclear deterrent works if and only if you have a deliverable weapon - there's a reason why every nuclear weapons state and wannabe weapons state has developed or is also trying to develop IRBMs at a minimum, ICBMs if at all possible.
Of course, the scary thing from that perspective is the W-33 atomic artillery projectile. 8 inch (20 cm) diameter, 32 inches long, 241 pounds or so. And, like Little Boy, a gun-type weapon, though a much much smarter design than Little Boy. W33 (nuclear weapon) at Wikipedia
Fits in a missile reentry vehicle just fine, thank you...
Somehow typical...
on
Zoe's Tale
·
· Score: 2, Funny
John posts on his blog that he's busy all day with real life and a sick kid, and you go and slashdot him...
Not enough solar energy to warm a blimp that way, but radioactive heat sources do nicely, and yes people have studied hot hydrogen balloons / blimps on Jupiter, Saturn, etc. They seem to work ok, if you stay out of the regions with high wind shear (flying a blimp into a hurricane is a bad life path choice...)
Reactors are better, but little radioactive heater units will work in a pinch.
> Are you kidding? People have been black-holed for decades on the internet for stuff like this.
Citation needed.
Canter and Siegel were kicked off their ISPs in decently short order 14 years ago (1994) after starting to spam. See: https://secure.wikimedia.org/wikipedia/en/wiki/Canter_and_siegel
Anyone familiar with the history of spamfighting will be able to point to numerous examples every year since then, of escalating size and complexity.
Look, the solution here is laws not vigilantism... Because the simple truth is no matter how good you are sooner or later you're going to fuck it up. The law ensures that when this happens, there's recourse. A vigilante will just disappear into the night with the words "I'm sorry" on his/her lips. And not only that, but the entire tone of your response rather underscores the need to get emotion out of this situation and the justice system is far better suited to this than your "Let's get a posse together and ride" solution.
Vigilantism is acting extrajudicially AND illegally as a community group to right a wrong or combat a criminal. It's an inappropriate model here - the response was entirely legal. It was done by people who, contrary to your assertion, were openly identified and stood and stand by their information.
If people were assassinating botnet operators or burning McColo datacenters down, THAT would be vigilantism. This is just community response.
Are you aware that reprocessed Plutonium isn't bomb grade?*
One can produce nuclear weapons using reactor grade plutonium, the APS article notwithstanding.
Yes, the weapon predetonates. One can overcome predetonation with deuterium-tritium gas boosting - the "minimum yield" listed on the APS article corresponds to 0.027 of "nominal yield". Nominal yield is going to be 10 to 20 kilotons depending on basic fission configuration and fissile mass used. Boosting effects work at yields of at least roughly 250 tons TNT.... which is 0.025 of a nominal 10-kiloton design. Once the boost fusion reactor ignites, it proceeds much faster than the fission reaction, and boosting up to 10 kt nominal yield is only a gram or two or tritium (and equivalent deuterium).
The US fired a reactor-grade plutonium bomb to test that, and it worked just fine.
One can also use at least all of gas centrifuge, vortex, or electromagnetic separation on Plutonium to produce weapons grade plutonium from reactor grade. The facility is intensely radioactive and hard to work in... but very small, because you only need 6 kg of output per weapon or less, and you get on the order of 25% of the mass out as useful weapons material (as opposed to Uranium enrichment, where you get around 0.3% out, about two orders of magnitude more efficient, and you need 3-4 times more material for Uranium bombs, making it about 300-400 times the effective material processing capacity).
I don't oppose reprocessing - I think it can be done safely and with little proliferation risk if treated seriously. But this particular argument, that it's not a proliferation risk, is bull.
He was engaging in issue advocacy (as were Mr Byrne and Mr Bagley).
Mr Byrne's interest in the topic is an open conflict of interest - if disclosed, others can monitor to make sure that one doesn't insert improper bias. Byrne disclosed where he was coming from, and in that regard wasn't causing a problem. Weiss' edits, as a journalist and columnist covering the issue and editorializing on the issue, were helping to try and establish his personal opinion on the topic as the neutral, standard opinion represented in the Wikipedia article.
A reporter who does not engage in writing opinion columns or editorials would have little conflict of interest, as long as they didn't use their own writing as a "reliable independent source" without disclosing that, or use the Wikipedia article as a source for their outside writing. Weiss was editorializing and writing opinion columns and advocating on the topic. Doing so secretly, using a cover identity and intentionally lying to people about the identity, forms a conflict of interest.
I have seen emails and heard recordings of voicemail and phone calls left by Mr Byrne and Mr Bagley. I am familiar with California stalking and terroristic threats laws due to an unrelated incident. Had victims in California filed police reports, I believe both Mr Byrne and Mr Bagley would have had felony warrants outstanding.
I've seen multiple independent reports of web tracking bugs much more sophisticated than mere tracking images, including viruses and malware, and at least one report of a keylogger virus uploaded from their website. Tracking images are merely slimy. The other activity appears to have broken federal computer crime laws.
I am discounting in this any claims made by the persons we now reasonably believe were Weiss. This is reports from other, completely unrelated people. Even discounting pseudonymous user Slimvirgin's experience, there were many many victims in this, and many instances that rise to the level of computer crime or stalking.
Evidence was collected that pretty unambiguously identifies Gary Weiss as the person behind two Wikipedia accounts which were pushing a point of view on the "Naked Short Selling" article, including a shift in editing hours which corresponded exactly with the period of time that Weiss was in India on his honeymoon, shifting from US east coast time to India time and then back again when Weiss returned to the US. There was general consensus on the identification once that evidence (from the normal, public edit logs) was collected, analyzed, and published.
None of the other Wikipedians involved is involved in the Financial industry or any form of serious investor. Nor did anyone have any particular interest in the topic before the fight broke out on-wiki and off.
It is factually true that Overstock.com CEO Patrick Byrne opposes naked short selling and publically blames it for low stock prices. It's factually true that he and at least one Overstock employee openly and then pseudonymously started a content war on Wikipedia on the article about it. It's generally concluded that the "other side" of this fight primarily was led by Gary Weiss, under two Wikipedia accounts now linked fairly unambiguously to him, though there hasn't been a public admission that I know of. Weiss had a conflict of interest in this matter, as a journalist covering the topic area.
Weiss never had Wikipedia administrator status, and thus the actions which Byrne blames for "censorship" were done by the other Wikipedia participants, mostly actual site administrators, who did not have conflicts of interest over the topic area. Byrne and his employee's accounts were permanently blocked from editing, and hundreds of known "sockpuppet" accounts created and used by them were also blocked. They were blocked because they threatened numerous Wikipedia volunteers, exposed alledged real names (sometimes wrong, sometimes right) of pseudonymous volunteers and personal information both of pseudonymous and openly identified individuals. Threatening phone calls were made to volunteers and their employers, viruses and various web tracking mechanisms were placed onto Byrne's website to try and help ID his alledged persecutors, and illegal access to some of the volunteers computers was made by Byrne and/or his employee. At least one other volunteer in California was subjected to threats and behavior that rose to the level of felony stalking here, though I was unable to get them to file police reports.
Byrne believes that this was all OK, because it did turn out that Weiss was credibly the person behind the two Wikipedia accounts. I for one believe that Byrne's behavior rises to the level of criminal, and that he displays behavior patterns most commonly associated with sociopaths in his online interactions.
His having been correct about Weiss does not change the fact that he is a scary, dangerous person who has little regard for other people's safety or feelings.
They've tripped over some previously known issues before, yes. However, it's not like large aerospace companies and NASA don't do the same thing all the time as well. And the total money input into SpaceX for the Falcon 1, (cancelled 5), and 9 vehicles so far is less than it would take to finish the paper studies part of any traditional aerospace companies' new launch vehicle.
Imperfect, yes, but no worse than many others, and much much cheaper.
Network architects and admins with clue are currently at the "Depression" stage (4th stage).
Why Slashdot feels that putting up a commentary authored by someone who's still in the first stage ("Denial") is useful to anyone is beyond me.
IPv4 exhaustion is coming. CIDR got us from the mid-90s until now. But it's coming now. Please stop denying, being angry, trying to bargain it away. Hopefully we'll all move past depression into acceptance (as vendors and infrastructure gets ready) before it hits. But I know a lot of smart people who would prefer to retire in the next 2 years instead of be there when it hits.
I have seen multiple server rooms full of systems ruined by A/C related failures (directly via overheating, indirectly by fires or sprinkler discharges after A/C failures).
Don't skimp!
Tell your employer to spend the money needed to get a proper A/C installation. "portable" units are ok - if they're professional grade, not home grade, and properly drained and wired and vent ducted. Permanent installation units are better, generally - cheaper for given tonnage of rating.
In either case, put in environmental monitoring ($400 or less) to send out alerts if it gets too hot or something else goes wrong.
IT folks tend to either not understand or not believe in the significance of getting facilities right. Experienced IT folks, with decades in the industry, know better. Take our advice - spend the money, get it right.
Figure out what your actual power is, what the UPS capacity is (remember, if the A/C isn't on UPS but the systems are, after a power failure the room will start heating up until the UPS is exhausted!). Definitely install monitoring (can be cheaper than any rackmount server is).
The mechanics of how these work make producing compact high speed arrays easy.
The circuit element is just two stacked planar layers between an underlying and overhead wire. Look at the electron microscope images to see what a row of them looks like... they're no bigger than the contact areas of the wires. A chip of these would be a grid of vertical wires, the active layers, then a grid of horizontal layers. The packing density is approximately wire spacing density.
Speed is good - you send a moderate voltage down one side and see if you get strong or weak signals out the far side, so it's essentially no delay other than speed of electron travel.
Skepticism is one thing. There could be all sorts of gotchas going from a small test area to large chips of this. But the fundamental method of operation is fast and the fundamental area is small, and it works at test scale. This is an extremely promising technology.
(As context, Nicholas was tangentally involved with Hans while we were all at Berkeley together, and I knew Hans more closely because I was there earlier / closer in age to Hans).
I'm not in touch with a whole lot of the rest of the crowd that knew him in the late 80s / early 90s at Cal. However, the people I am still talking to had a range of opinions... Hans was wierd, but not wierd in the way that would make you think he'd hurt or kill someone eventually. There was doubt - Nina was clearly wierd, too, as were several other people involved (Sturgeon, for one, made a better potential killer). Many other things could have been the underlying factual truth. I was personally hoping that I hadn't gone to school with someone who later became a murderer.
Plenty of innocent people have been caught up in situations that made them look guilty with various evidence and eyewitness reports... Hence the current spate of DNA evidence based overturned convictions. Think how many other innocent people were convicted of things but can't prove it because the real murderer didn't leave DNA that was found...
Yes, it was always suspicious. I don't know anyone who didn't at least put significant weight on the possibility he had killed her. I hoped not, and I'm very disappointed, and sad for his kids, and their grandparents, and for Nina.
This isn't a situation to be getting self-righteous over. No matter what the "right"/"true" answer was, it was a terrible situation, and this was not the best possible outcome. I know several geek community people that I hope this pushes into relationships counseling and anger management counseling.
Actually, I used to work at Intel (around the time of 0.6um) and one could, and indeed, did sometimes shrink chips just by "shrinkenating", or perhaps shrinkenating followed by a design rule check. The result was a chip that was cheaper to manufacture, and in most cases, ran faster.
I know what you were saying, but for the benefit of the general audience:
That works better if all the geometries scale linearly (line separation, aspect ratios, layer thicknesses, etc). As a general rule, that changes slightly from one generation to another, but there are often significant changes.
And going from 0.6u to 0.35u to 0.25u to 0.18u to 0.13u to 90 nm to 65 nm to 45 nm is a few too many steps for that assumption to work....
Particularly given that modern chip photomasks are a completely different phase-shift tech than the older ones. You couldn't size down older masks to new process at all.
Back to your main point, on why use P54 anyways... My guess is that they really want to kickstart their many-many-core work with this and walked back along their product line until they came to something with enough features, few enough transistors, and modern enough logic model / HDL or Verilog code that they could have a fair chance of translating and resynthesizing it rapidly.
But that's stretching the available leak knowledge a ways. Someone will eventually go on record with the real details.
One does not "shrink" a chip by taking photomasks and shrinkenating. One redoes the design / layout process, generally. The P5 series went from 0.8 um to 0.25 um over its lifetime (through Tillamook), stepping through 0.6, 0.35, and finally 0.25 um.
It was 148 mm^2 at 0.6 um, so the process shrink should bring it down to a floorplan of around a square millimeter or so a core. Not sure how big the die will be for Larrabee, but the extra space will probably support the simple wide data unit per core and more cache. If the SIMD is simple it could be another 3-4 million transistors / 1 square mm or so. For a 100 mm^2 chip that gives you another 30 mm^2 or so for I/O and cache (either shared, or parceled out to the cores).
There's an old saying, along the lines of "To see something, you must look at it, and then you must see it."
A lot of what you look at you could match / comprehend properly but don't. In many cases, parts of the brain used for the pattern recognition do fire, but the process doesn't complete (due to overload, fatigue, etc).
Having something mechanical flag those for you will help with the final seeing part.
Of course, it has to be tuned right. A lot of the brain's pattern match stuff fires on things which aren't a real pattern. Every edge is autodetected early in the processing, etc. I don't know how wide the actual useful range for this is, where fatigue or overload will prevent the last couple of steps but enough happened that one can meaningfully statistically say "you should look at that again" to the person in the loop...
The South African bomb doesn't resemble the Israeli bombs whatsoever - it was a uranium gun-type bomb, and the Israeli bombs are plutonium implosion bombs.
There was some cooperation, but it's not clear how relevant that ended up being for the actual weapon project they built, as opposed to general R&D ideas.
The number includes the estimated cost of the weapon related enrichment program, which built somewhat on some power reactor uranium programs but had to build its own facility from scratch. The number was validated by doing a headcount of people who worked on the program - it was amazingly small (technical core team about 50 engineers and scientists, for about 10 years of work before the weapons were made).
The bomb design itself was simple and robust, easy to design and highly reliable. Not as compact or light as a more advanced gun type weapon could have been (it weighed about a ton - US advanced nuclear artillery shells with gun-type uranium operating mechanisms got down to about 100 kilograms total weight in the W33). Simple, reliable, easy, and very cheap for them to design and engineer.
They only made 7 bombs (1 "test prototype", though not fired, and 6 sets of the weaponized bomb), but could have kept making one every 6 to 8 months or so without additional infrastructure and at costs of only $10-20 million per year to keep the factories and plants running.
There's a great myth out there in the public mind that nuclear weapons are very very hard to design and hard to produce. They are only hard to design, and expensive to produce. Don't believe the myth.
One of the advantages of being a consultant is that sometimes, you can throw down and say the truth where an organization doesn't want to hear it, because you can walk away when they don't want to hear it.
I've been told to leave a company I was consulting at after a series of development related SNAFUs. I was the IT architect standing between the development project and live, internet-customer facing production, and after re-doing the QA work and benchmarking and reliability testing on the product determined that it wasn't ready to go live.
My reputation inside was dinged badly by saying that once in a wide meeting right before the demo site was supposed to go up. I was overridden, and it went up, and then went down for a week. Telling the truth right then, right before it embarrassed a lot of people, was Not Ok within the political environment.
They kept patching and pushing it forwards, and I kept pushing back where it wasn't ready yet. The final straw was when all the performance analysis and benchmarking indicated that NewApp was only half as fast as OldApp and therefore needed twice as many servers, and I told the ops VP to buy another thousand servers to support the first limited production environment of the app. This was overridden by senior VP, who pulled new numbers out of his ass, asked me to leave, and shortly thereafter asked my VP to leave.
NewApp went live, turned out to run half as fast as OldApp in real customer facing production environments, and the company ended up spending many many millions of dollars on new server hardware and datacenter space to support NewApp, which they launched anyways. Everyone who had stood up and told the truth about it was gone a year later. So was the dev team leadership - they didn't "win" either, as it was obviously FUBAR code. Company was bought out eighteen months after I left.
But I was comfortably working elsewhere those 18 months.
Slashdot Mirror
Um, no. Plain low-Z foams are well documented in a number of nuclear weapons - are perfectly straighforwards to make, use, and test. Fogbank's described properties pretty much only match an aerogel of some sort with suspended high-Z material, and extensive (non-public) analysis by a physicist with inertial confinement fusion experience indicates that Fogbank could be extremely critical to the operation of highly compact thermonuclear secondaries (the second, fusion implosion stage after the primary fission fires). Very thin radiation cases and energy buffered into the aerogel's high-Z constituent apparently allow you to effectively push the secondary without having a thick (heavy) radiation case to contain the primary's energy for much longer. There are a number of weapons that didn't have that level of compact secondary still in use - B61s, B83s, W62s, etc. However, a number of the very tight tolerances secondaries in use - W76, W87, W88, possibly W80 and the other B61 derivatives with stepped radiation cases, possibly W89 and RRW derivatives, probably don't work without the suspended high-Z aerogel material. Could we redesign them with thicker radiation cases instead? Sure. Add ... 20% perhaps to overall weight. Oh, and we'd have to withdraw from the nuclear test ban treaty and the threshold test ban treaty to test the redesigned weapons, because that redesign is NOT a minor issue with reliability, it's a fundamental physics/engineering change, even if the primary and secondary are the same. It's changing the dynamics of the energy capture from the primary and the timing and intensity of the energy pulse delivered to the secondary, in a radical manner. So, you need to test it.
Or we could go back to earlier, heavier designs, like the B61 and B83s. Except that all our current ICBM warheads appear to use Fogbank now. Oops. Error. Try again.
The Nvidia GPUs are large MIMD vector machines - look at the specs, and what they're doing with CUDA. That they're mostly actually used to draw texels of monsters and walls and bullets in flight doesn't mean that they're not a highly capable general purpose vector processor...
Many people are (almost certainly correctly) stating that Nvidia wouldn't do that if they didn't think they had to, or didn't think that this would make them more money / market penetration than not doing it. Suggesting that sticking a small x86 on the corner of their big graphics chips is somehow an architectural black hole is silly - they're not dumb, they're chasing their market. The CPU won't slow anything else down.
Better metrics to use (field-proven at a bunch of dot-coms you may have heard of) - these are rough numbers, but more precise ones require a detailed customer analysis and usually detailed log analysis after you start the actual business up live.
Assume "business day" of 10 hours - most businesses have a high point during the day that corresponds to this type of distribution.
Assume peak load in a given second will be 10x of that value.
So, if you expect 100,000 visitors a day (the story's corrected load), "business day average" will be 10,000 visitors per hour, about 2.7/sec. Assume 27/sec for true peak load.
Multiply by pages viewed per visitor per day - do they typically hit 2 pages, 8, 16? Depends on the content, usage model, etc. This gives you pages per second. A good conservative number is 10, which makes calculation easy. So that's peak page views of 270/sec.
Multiply by the number of hits per page based on number of images and other page components (navigation bars, graphics, frames, etc) - typical numbers on this range from 2.2 or so (straight HTML with a background image and a sitewide favicon) to 25 (complex multilayer navigation implimented with many graphic icons, etc). Again, assume 10 for a conservative rough order of magnitude estimate.
That gives you static hits per second. So something like 10x hits/page is 2,700 hits/sec peak rate.
If your pages are static, that's all and you're done. If they're dynamic, you have to consider the "app server" load (whether that's PHP or Tomcat or CGIs on the webserver, or back ending onto a different Tomcat or Websphere or Weblogic server or some such).
10 years ago, we built a large ecommerce website for a large bricks and mortar company from the ground up using that modeling. At that time, the right answer for the servers was Solaris/SPARC. At that time, the performance we got (on 400ish MHz CPUs) was 230 dynamic page generations per second per CPU using Weblogic as the app server, and around 2,000 static pages per second per CPU using the C2 Apache Stronghold and HTTPS. Our build spec was significantly more than 100,000 users - we used a Sun E10K cluster with separate domains on each system for web and app servers.
Obviously, faster modern CPUs do better, and the amount of memory and CPU power available in even cheap desktop or rackmount servers is plenty for the stated user load requirements on these servers.
You can screw that up by using the lousiest PHP or CGI scripts in the world, using excessively large pages or numbers of navigation images, etc. But even a site which is an order of magnitude slower than it should be will support 100,000 users on a handful of systems, if not on a handful of CPU cores with 2 or 3 boxes for redundancy.
So - Figure out what your static and dynamic content balance is, figure out a pair of "load balancers" out in front, put at least 2 servers at each level (either web, or web+app servers separately - if you track user accounts etc, then add a redundant database layer too). The cheapest servers you can get will be more powerful than you need for pure load considerations but you need to think about degraded mode performance when one falls down.
If you're hosting this at a hosting facility get the cheapest 1U rackmount servers you can. If you're hosting it in an office then small tower cases are marginally cheaper, but computer room space may be at a premium pushing you to racks anyways.
Set up a master server with the master copy and backup / older copies of the content and configurations. Distribute those out (rsync, cfengine, whatever) to the "live" servers. Put one server with large hard drives out in a completely different location and back up the files there nightly or more often, for disaster recovery.
In terms of clustering - you want to load balance the web and app servers. Probably don't want to cluster the database - easier to set up a slave DB server than true clustering and true DB clustering leads to many failure modes that non-experts often fall into, leaving them less reliable than master/slave plus manual failover options.
The general concept doesn't totally suck - one could build an offshore hotel on an oil platform, and some of the winning design's components look reasonable to me.
Rough seas and high winds at times, and the occasional hurricane, aren't impossible to work around at a resort (it's not like parts of Florida, the Caribbean, Mexico etc don't have those problems on land). It's going to need more evacuations and will have an operations impact, but you just have to plan and budget for that. You give people refunds / free future stays / etc if their visit is cancelled due to a storm, etc.
But you can't build something with a blatant blaring flaw, and letting something with a blatant blaring flaw win a design competition even if it was a neat concept doesn't make sense to me.
Whoever the designer of this was, has no experience with offshore platforms.
You NEVER put the helicopter platform on the side. It's never in clear air then, you can't safely approach the pad and land. And it was too close to hotel rooms anyways, where they have it, if the pilot goofs he flies the rotor into the rooms.
The China Airlines 747 was severely damaged and nearly had to be scrapped. Not due to supersonic flight loads, but due to damage from the high-G pullout required to recover from the out of control power dive towards the ocean.
Among other things the landing gear locks pulled out of their fuselage mounts and the gear extended partly during the dive pullout, damaging the gear and gear doors.
The pullout encountered 5.1 and 4.8 G peaks, which exceed the normal structural limits, and the aircraft's wings were permanently bent upwards 2-3 inches.
The horizontal tailfins also were partially shredded - see pictures and more incident data at:
http://aviation-safety.net/database/record.php?id=19850219-0
Also NTSB report available at:
http://www.rvs.uni-bielefeld.de/publications/Incidents/DOCS/ComAndRep/ChinaAir/AAR8603.html
Some of you may remember the 1980s and early 1990s, where PCs started out costing $5,000 and declined slowly to around $2,500 for name brand models.
Around 1995, CPUs exceeded the GUI requirements of all the apps then popular (this is pre-modern gaming, of course). Around 1996 and into 1997 the prices of PCs fell off a cliff, down to $1,000.
Those who fail to remember history...
Of course, the scary thing from that perspective is the W-33 atomic artillery projectile. 8 inch (20 cm) diameter, 32 inches long, 241 pounds or so. And, like Little Boy, a gun-type weapon, though a much much smarter design than Little Boy.
W33 (nuclear weapon) at Wikipedia
Fits in a missile reentry vehicle just fine, thank you...
John posts on his blog that he's busy all day with real life and a sick kid, and you go and slashdot him...
TANJ.
Good book, though.
Not enough solar energy to warm a blimp that way, but radioactive heat sources do nicely, and yes people have studied hot hydrogen balloons / blimps on Jupiter, Saturn, etc. They seem to work ok, if you stay out of the regions with high wind shear (flying a blimp into a hurricane is a bad life path choice...)
Reactors are better, but little radioactive heater units will work in a pinch.
Canter and Siegel were kicked off their ISPs in decently short order 14 years ago (1994) after starting to spam. See:
https://secure.wikimedia.org/wikipedia/en/wiki/Canter_and_siegel
Anyone familiar with the history of spamfighting will be able to point to numerous examples every year since then, of escalating size and complexity.
Vigilantism is acting extrajudicially AND illegally as a community group to right a wrong or combat a criminal. It's an inappropriate model here - the response was entirely legal. It was done by people who, contrary to your assertion, were openly identified and stood and stand by their information.
If people were assassinating botnet operators or burning McColo datacenters down, THAT would be vigilantism. This is just community response.
One can produce nuclear weapons using reactor grade plutonium, the APS article notwithstanding.
Yes, the weapon predetonates. One can overcome predetonation with deuterium-tritium gas boosting - the "minimum yield" listed on the APS article corresponds to 0.027 of "nominal yield". Nominal yield is going to be 10 to 20 kilotons depending on basic fission configuration and fissile mass used. Boosting effects work at yields of at least roughly 250 tons TNT .... which is 0.025 of a nominal 10-kiloton design. Once the boost fusion reactor ignites, it proceeds much faster than the fission reaction, and boosting up to 10 kt nominal yield is only a gram or two or tritium (and equivalent deuterium).
The US fired a reactor-grade plutonium bomb to test that, and it worked just fine.
One can also use at least all of gas centrifuge, vortex, or electromagnetic separation on Plutonium to produce weapons grade plutonium from reactor grade. The facility is intensely radioactive and hard to work in... but very small, because you only need 6 kg of output per weapon or less, and you get on the order of 25% of the mass out as useful weapons material (as opposed to Uranium enrichment, where you get around 0.3% out, about two orders of magnitude more efficient, and you need 3-4 times more material for Uranium bombs, making it about 300-400 times the effective material processing capacity).
I don't oppose reprocessing - I think it can be done safely and with little proliferation risk if treated seriously. But this particular argument, that it's not a proliferation risk, is bull.
He was engaging in issue advocacy (as were Mr Byrne and Mr Bagley).
Mr Byrne's interest in the topic is an open conflict of interest - if disclosed, others can monitor to make sure that one doesn't insert improper bias. Byrne disclosed where he was coming from, and in that regard wasn't causing a problem. Weiss' edits, as a journalist and columnist covering the issue and editorializing on the issue, were helping to try and establish his personal opinion on the topic as the neutral, standard opinion represented in the Wikipedia article.
A reporter who does not engage in writing opinion columns or editorials would have little conflict of interest, as long as they didn't use their own writing as a "reliable independent source" without disclosing that, or use the Wikipedia article as a source for their outside writing. Weiss was editorializing and writing opinion columns and advocating on the topic. Doing so secretly, using a cover identity and intentionally lying to people about the identity, forms a conflict of interest.
I have seen emails and heard recordings of voicemail and phone calls left by Mr Byrne and Mr Bagley. I am familiar with California stalking and terroristic threats laws due to an unrelated incident. Had victims in California filed police reports, I believe both Mr Byrne and Mr Bagley would have had felony warrants outstanding.
I've seen multiple independent reports of web tracking bugs much more sophisticated than mere tracking images, including viruses and malware, and at least one report of a keylogger virus uploaded from their website. Tracking images are merely slimy. The other activity appears to have broken federal computer crime laws.
I am discounting in this any claims made by the persons we now reasonably believe were Weiss. This is reports from other, completely unrelated people. Even discounting pseudonymous user Slimvirgin's experience, there were many many victims in this, and many instances that rise to the level of computer crime or stalking.
Evidence was collected that pretty unambiguously identifies Gary Weiss as the person behind two Wikipedia accounts which were pushing a point of view on the "Naked Short Selling" article, including a shift in editing hours which corresponded exactly with the period of time that Weiss was in India on his honeymoon, shifting from US east coast time to India time and then back again when Weiss returned to the US. There was general consensus on the identification once that evidence (from the normal, public edit logs) was collected, analyzed, and published.
None of the other Wikipedians involved is involved in the Financial industry or any form of serious investor. Nor did anyone have any particular interest in the topic before the fight broke out on-wiki and off.
It is factually true that Overstock.com CEO Patrick Byrne opposes naked short selling and publically blames it for low stock prices. It's factually true that he and at least one Overstock employee openly and then pseudonymously started a content war on Wikipedia on the article about it. It's generally concluded that the "other side" of this fight primarily was led by Gary Weiss, under two Wikipedia accounts now linked fairly unambiguously to him, though there hasn't been a public admission that I know of. Weiss had a conflict of interest in this matter, as a journalist covering the topic area.
Weiss never had Wikipedia administrator status, and thus the actions which Byrne blames for "censorship" were done by the other Wikipedia participants, mostly actual site administrators, who did not have conflicts of interest over the topic area. Byrne and his employee's accounts were permanently blocked from editing, and hundreds of known "sockpuppet" accounts created and used by them were also blocked. They were blocked because they threatened numerous Wikipedia volunteers, exposed alledged real names (sometimes wrong, sometimes right) of pseudonymous volunteers and personal information both of pseudonymous and openly identified individuals. Threatening phone calls were made to volunteers and their employers, viruses and various web tracking mechanisms were placed onto Byrne's website to try and help ID his alledged persecutors, and illegal access to some of the volunteers computers was made by Byrne and/or his employee. At least one other volunteer in California was subjected to threats and behavior that rose to the level of felony stalking here, though I was unable to get them to file police reports.
Byrne believes that this was all OK, because it did turn out that Weiss was credibly the person behind the two Wikipedia accounts. I for one believe that Byrne's behavior rises to the level of criminal, and that he displays behavior patterns most commonly associated with sociopaths in his online interactions.
His having been correct about Weiss does not change the fact that he is a scary, dangerous person who has little regard for other people's safety or feelings.
They've tripped over some previously known issues before, yes. However, it's not like large aerospace companies and NASA don't do the same thing all the time as well. And the total money input into SpaceX for the Falcon 1, (cancelled 5), and 9 vehicles so far is less than it would take to finish the paper studies part of any traditional aerospace companies' new launch vehicle.
Imperfect, yes, but no worse than many others, and much much cheaper.
Network architects and admins with clue are currently at the "Depression" stage (4th stage).
Why Slashdot feels that putting up a commentary authored by someone who's still in the first stage ("Denial") is useful to anyone is beyond me.
IPv4 exhaustion is coming. CIDR got us from the mid-90s until now. But it's coming now. Please stop denying, being angry, trying to bargain it away. Hopefully we'll all move past depression into acceptance (as vendors and infrastructure gets ready) before it hits. But I know a lot of smart people who would prefer to retire in the next 2 years instead of be there when it hits.
They probably won't, but would like to...
I have seen multiple server rooms full of systems ruined by A/C related failures (directly via overheating, indirectly by fires or sprinkler discharges after A/C failures).
Don't skimp!
Tell your employer to spend the money needed to get a proper A/C installation. "portable" units are ok - if they're professional grade, not home grade, and properly drained and wired and vent ducted. Permanent installation units are better, generally - cheaper for given tonnage of rating.
In either case, put in environmental monitoring ($400 or less) to send out alerts if it gets too hot or something else goes wrong.
IT folks tend to either not understand or not believe in the significance of getting facilities right. Experienced IT folks, with decades in the industry, know better. Take our advice - spend the money, get it right.
Figure out what your actual power is, what the UPS capacity is (remember, if the A/C isn't on UPS but the systems are, after a power failure the room will start heating up until the UPS is exhausted!). Definitely install monitoring (can be cheaper than any rackmount server is).
The mechanics of how these work make producing compact high speed arrays easy.
The circuit element is just two stacked planar layers between an underlying and overhead wire. Look at the electron microscope images to see what a row of them looks like... they're no bigger than the contact areas of the wires. A chip of these would be a grid of vertical wires, the active layers, then a grid of horizontal layers. The packing density is approximately wire spacing density.
Speed is good - you send a moderate voltage down one side and see if you get strong or weak signals out the far side, so it's essentially no delay other than speed of electron travel.
Skepticism is one thing. There could be all sorts of gotchas going from a small test area to large chips of this. But the fundamental method of operation is fast and the fundamental area is small, and it works at test scale. This is an extremely promising technology.
Hi Nick.
(As context, Nicholas was tangentally involved with Hans while we were all at Berkeley together, and I knew Hans more closely because I was there earlier / closer in age to Hans).
I'm not in touch with a whole lot of the rest of the crowd that knew him in the late 80s / early 90s at Cal. However, the people I am still talking to had a range of opinions... Hans was wierd, but not wierd in the way that would make you think he'd hurt or kill someone eventually. There was doubt - Nina was clearly wierd, too, as were several other people involved (Sturgeon, for one, made a better potential killer). Many other things could have been the underlying factual truth. I was personally hoping that I hadn't gone to school with someone who later became a murderer.
Plenty of innocent people have been caught up in situations that made them look guilty with various evidence and eyewitness reports... Hence the current spate of DNA evidence based overturned convictions. Think how many other innocent people were convicted of things but can't prove it because the real murderer didn't leave DNA that was found...
Yes, it was always suspicious. I don't know anyone who didn't at least put significant weight on the possibility he had killed her. I hoped not, and I'm very disappointed, and sad for his kids, and their grandparents, and for Nina.
This isn't a situation to be getting self-righteous over. No matter what the "right"/"true" answer was, it was a terrible situation, and this was not the best possible outcome. I know several geek community people that I hope this pushes into relationships counseling and anger management counseling.
Sad day.
I know what you were saying, but for the benefit of the general audience:
That works better if all the geometries scale linearly (line separation, aspect ratios, layer thicknesses, etc). As a general rule, that changes slightly from one generation to another, but there are often significant changes.
And going from 0.6u to 0.35u to 0.25u to 0.18u to 0.13u to 90 nm to 65 nm to 45 nm is a few too many steps for that assumption to work....
Particularly given that modern chip photomasks are a completely different phase-shift tech than the older ones. You couldn't size down older masks to new process at all.
Back to your main point, on why use P54 anyways... My guess is that they really want to kickstart their many-many-core work with this and walked back along their product line until they came to something with enough features, few enough transistors, and modern enough logic model / HDL or Verilog code that they could have a fair chance of translating and resynthesizing it rapidly.
But that's stretching the available leak knowledge a ways. Someone will eventually go on record with the real details.
One does not "shrink" a chip by taking photomasks and shrinkenating. One redoes the design / layout process, generally. The P5 series went from 0.8 um to 0.25 um over its lifetime (through Tillamook), stepping through 0.6, 0.35, and finally 0.25 um.
It was 148 mm^2 at 0.6 um, so the process shrink should bring it down to a floorplan of around a square millimeter or so a core. Not sure how big the die will be for Larrabee, but the extra space will probably support the simple wide data unit per core and more cache. If the SIMD is simple it could be another 3-4 million transistors / 1 square mm or so. For a 100 mm^2 chip that gives you another 30 mm^2 or so for I/O and cache (either shared, or parceled out to the cores).
There's an old saying, along the lines of "To see something, you must look at it, and then you must see it."
A lot of what you look at you could match / comprehend properly but don't. In many cases, parts of the brain used for the pattern recognition do fire, but the process doesn't complete (due to overload, fatigue, etc).
Having something mechanical flag those for you will help with the final seeing part.
Of course, it has to be tuned right. A lot of the brain's pattern match stuff fires on things which aren't a real pattern. Every edge is autodetected early in the processing, etc. I don't know how wide the actual useful range for this is, where fatigue or overload will prevent the last couple of steps but enough happened that one can meaningfully statistically say "you should look at that again" to the person in the loop...
The South African bomb doesn't resemble the Israeli bombs whatsoever - it was a uranium gun-type bomb, and the Israeli bombs are plutonium implosion bombs.
There was some cooperation, but it's not clear how relevant that ended up being for the actual weapon project they built, as opposed to general R&D ideas.
The number includes the estimated cost of the weapon related enrichment program, which built somewhat on some power reactor uranium programs but had to build its own facility from scratch. The number was validated by doing a headcount of people who worked on the program - it was amazingly small (technical core team about 50 engineers and scientists, for about 10 years of work before the weapons were made).
The bomb design itself was simple and robust, easy to design and highly reliable. Not as compact or light as a more advanced gun type weapon could have been (it weighed about a ton - US advanced nuclear artillery shells with gun-type uranium operating mechanisms got down to about 100 kilograms total weight in the W33). Simple, reliable, easy, and very cheap for them to design and engineer.
They only made 7 bombs (1 "test prototype", though not fired, and 6 sets of the weaponized bomb), but could have kept making one every 6 to 8 months or so without additional infrastructure and at costs of only $10-20 million per year to keep the factories and plants running.
There's a great myth out there in the public mind that nuclear weapons are very very hard to design and hard to produce. They are only hard to design, and expensive to produce. Don't believe the myth.
One of the advantages of being a consultant is that sometimes, you can throw down and say the truth where an organization doesn't want to hear it, because you can walk away when they don't want to hear it.
I've been told to leave a company I was consulting at after a series of development related SNAFUs. I was the IT architect standing between the development project and live, internet-customer facing production, and after re-doing the QA work and benchmarking and reliability testing on the product determined that it wasn't ready to go live.
My reputation inside was dinged badly by saying that once in a wide meeting right before the demo site was supposed to go up. I was overridden, and it went up, and then went down for a week. Telling the truth right then, right before it embarrassed a lot of people, was Not Ok within the political environment.
They kept patching and pushing it forwards, and I kept pushing back where it wasn't ready yet. The final straw was when all the performance analysis and benchmarking indicated that NewApp was only half as fast as OldApp and therefore needed twice as many servers, and I told the ops VP to buy another thousand servers to support the first limited production environment of the app. This was overridden by senior VP, who pulled new numbers out of his ass, asked me to leave, and shortly thereafter asked my VP to leave.
NewApp went live, turned out to run half as fast as OldApp in real customer facing production environments, and the company ended up spending many many millions of dollars on new server hardware and datacenter space to support NewApp, which they launched anyways. Everyone who had stood up and told the truth about it was gone a year later. So was the dev team leadership - they didn't "win" either, as it was obviously FUBAR code. Company was bought out eighteen months after I left.
But I was comfortably working elsewhere those 18 months.