I would be extremely happy if our instrumentation had finally become powerful enough to discover experimental evidence that breaks the model.
The last huge time someone said, "Hold on--it should not be doing this!" was Planck, in 1900, when he found light quanta in black body radiation.
Basically, Planck was expecting the color of the light of a hot body to increase smoothly as the temperature went up...(infrared, visible, UV, Xray, gamma)....Unfortunately, he found that in reality, it did _not_ go up smoothly....It went up in a staircase with billions of teeny tiny steps, meaning light is *quantized*. This effed up our entire model. All of it. Before this discovery, the precession of Mercury (ended up being a relativity thing) was the only thing people were having a tough time with. Then this hit and they had to develop a system of mechanics to deal with these quanta.
Check out the next 15 years:
1901: Max Planck, determination of Planck's constant, Boltzmann's constant, Avogadro's number and the charge on electron
1904: Albert Einstein, energy-frequency relation of light quanta
1905: Albert Einstein, special relativity
1909: Robert Millikan, measured electron charge
1909: Albert Einstein, particle-wave duality of photons
1911: Ernest Rutherford, Infers the nucleus from the weird scattering of alpha particles on gold foil
1913: Niels Bohr, quantum theory of atomic orbits. Same year: radioactivity as nuclear property
1915: Albert Einstein, general relativity
Not bad for fifteen years.
Now, while we have made a lot of progress messing with these basic discoveries in cosmology, particle theory, quantum theory etc, we still have been refining these models. We haven't had to chuck the whole thing in a while.
I want another fifteen years like this. But for this to happen, the thing needs to break. In half.
Of course, I have a bias. I want zero point energy, flying cars and FTL travel. So I am praying for rain.
Thomas S. Kuhn's The Structure of Scientific Revolutions is about this very subject--not that we will never know anything, but that science is about building models, refining the models until experimental evidence no longer fits our model, throwing it out, and coming up with a better model.
Really good book, if you somehow missed it in college.
As in, this posting is not complimentary of JohnKatz.
vis-a-vis
Two actors' performances were complementary.
Spelling errers are one thing. Ineffectual command of English is another, especially in someone whose failure to grasp technical details already rankles the population. Do at least one thing well.
/RANT
Clustering != load-balancing.
Two completely different ideas.
Load balancing is simply splitting requests between multiple servers. Because HHTP is a stateless protocol--it remembers nothing between GET or POST requests, every time you go to a site, the webserver does not know you from Adam. This really sucks when you are trying to get fifty pieces of information from someone, put it in a database, and then _inform_ that person authoritatively that it went into the database.
The match made in hell between the stateless HTTP protocol and _very_ stateful, transactional datastores gave birth to an entire industry--middleware: arbitrary server-side datastores that could keep track of a user's session info, keyed on UIDs in cookies, or written in the URL.
When they tried to round-robin between their new, spiffy, stateful system, they realized that because only one server instance had the session info, the chances of hitting the same server were pretty slim.
So they began replicating the session-state between the servers, and called it clustering. To this day, it sucks and doesn't work--under load, the app server might not have time to replicate to the others in your cluster, because software is _slow_. What it _is_ good for, tho, is failover. Odds are the primary managed to at _least_ send your session-state off to the secondary before dying.
Finally, vendors like Cisco they started putting sticky-bit features in their hardware load balancers (read: LocalDirector) that make sure requests from a given IP "stick" to the first server they hit--replication is generally used for failover.
2 Scenarios:
1. w/o clustered persistence:
You have two appservers. One dies. You fail over, the app server has no idea who you are. Log in again. What state was your transaction in again? Did you buy that widget or not?
2. w/ clustered persistence:
You have two app servers, replicating session between the two. App1 fails. App2 knows about your session, picks up where the other left off.
In terms of computer tech, I would have to suggest Vernor Vinge, an excellent hard Sci-Fi writer and UC San Diego Comp Sci emeritus who deals largely with the technological possibilities of distributed systems and their subsequent effects on civilization.
Specifically, he postulated cyberspace (long before Gibson) in a novella called True Names, and also speculates on the future of mankind as the rate of technological innovation continues to accelerate, perhaps towards some sort of singularity beyond which further human endeavor will be somewhat incomprehensible to us.
He has won several awards, in particular for A Fire Upon the Deep, which looks at the "silence in the sky" problem....as in, if life is so damned plentiful in this universe, where are all the visible-forty-lights-away Bussard Ramjet flares? He uses a solution to this question as a unique premise for the novel. Awesome.
Slashdot Mirror
The last huge time someone said, "Hold on--it should not be doing this!" was Planck, in 1900, when he found light quanta in black body radiation.
Basically, Planck was expecting the color of the light of a hot body to increase smoothly as the temperature went up...(infrared, visible, UV, Xray, gamma)....Unfortunately, he found that in reality, it did _not_ go up smoothly....It went up in a staircase with billions of teeny tiny steps, meaning light is *quantized*. This effed up our entire model. All of it. Before this discovery, the precession of Mercury (ended up being a relativity thing) was the only thing people were having a tough time with. Then this hit and they had to develop a system of mechanics to deal with these quanta.
Check out the next 15 years:
1901: Max Planck, determination of Planck's constant, Boltzmann's constant, Avogadro's number and the charge on electron
1904: Albert Einstein, energy-frequency relation of light quanta
1905: Albert Einstein, special relativity
1909: Robert Millikan, measured electron charge
1909: Albert Einstein, particle-wave duality of photons
1911: Ernest Rutherford, Infers the nucleus from the weird scattering of alpha particles on gold foil
1913: Niels Bohr, quantum theory of atomic orbits. Same year: radioactivity as nuclear property
1915: Albert Einstein, general relativity
Not bad for fifteen years.
Now, while we have made a lot of progress messing with these basic discoveries in cosmology, particle theory, quantum theory etc, we still have been refining these models. We haven't had to chuck the whole thing in a while.
I want another fifteen years like this. But for this to happen, the thing needs to break. In half.
Of course, I have a bias. I want zero point energy, flying cars and FTL travel. So I am praying for rain.
Really good book, if you somehow missed it in college.
RANT reason="grammar"
Compliment != complement.
As in, this posting is not complimentary of JohnKatz.
vis-a-vis
Two actors' performances were complementary.
Spelling errers are one thing. Ineffectual command of English is another, especially in someone whose failure to grasp technical details already rankles the population. Do at least one thing well.
/RANT
Or, more ominously, all the other sapient life keeps a low EM signature to avoid divulging the locations of their homeworlds...
Meanwhile we got the music turned up nice and loud. Great.
Clustering != load-balancing.
Two completely different ideas.
Load balancing is simply splitting requests between multiple servers. Because HHTP is a stateless protocol--it remembers nothing between GET or POST requests, every time you go to a site, the webserver does not know you from Adam. This really sucks when you are trying to get fifty pieces of information from someone, put it in a database, and then _inform_ that person authoritatively that it went into the database.
The match made in hell between the stateless HTTP protocol and _very_ stateful, transactional datastores gave birth to an entire industry--middleware: arbitrary server-side datastores that could keep track of a user's session info, keyed on UIDs in cookies, or written in the URL.
When they tried to round-robin between their new, spiffy, stateful system, they realized that because only one server instance had the session info, the chances of hitting the same server were pretty slim.
So they began replicating the session-state between the servers, and called it clustering. To this day, it sucks and doesn't work--under load, the app server might not have time to replicate to the others in your cluster, because software is _slow_. What it _is_ good for, tho, is failover. Odds are the primary managed to at _least_ send your session-state off to the secondary before dying.
Finally, vendors like Cisco they started putting sticky-bit features in their hardware load balancers (read: LocalDirector) that make sure requests from a given IP "stick" to the first server they hit--replication is generally used for failover.
Session-state peristence.
2 Scenarios:
1. w/o clustered persistence:
You have two appservers. One dies. You fail over, the app server has no idea who you are. Log in again. What state was your transaction in again? Did you buy that widget or not?
2. w/ clustered persistence:
You have two app servers, replicating session between the two. App1 fails. App2 knows about your session, picks up where the other left off.
In terms of computer tech, I would have to suggest Vernor Vinge, an excellent hard Sci-Fi writer and UC San Diego Comp Sci emeritus who deals largely with the technological possibilities of distributed systems and their subsequent effects on civilization.
Specifically, he postulated cyberspace (long before Gibson) in a novella called True Names, and also speculates on the future of mankind as the rate of technological innovation continues to accelerate, perhaps towards some sort of singularity beyond which further human endeavor will be somewhat incomprehensible to us.
He has won several awards, in particular for A Fire Upon the Deep, which looks at the "silence in the sky" problem....as in, if life is so damned plentiful in this universe, where are all the visible-forty-lights-away Bussard Ramjet flares? He uses a solution to this question as a unique premise for the novel. Awesome.