3 ms·

I think they receive a little too much credit in many cases for invention when their primary contribution has been popularization. That's being a little pedant

by karterk 13y ago

I think they receive a little too much credit in many cases for invention when their primary contribution has been popularization. That's being a little pedantic. Applying X for Y is definitely being inventive. That's how largely every domain works. We stand on the shoulders of giants etc.

jandrewrogers 13y ago
The weakness of this contribution is often cited as the reason many types of software patents should be invalid. Putting a shiny wrapper on an old idea in an obvious but slightly different context is something upon which many software hackers would normally cast a jaundiced eye. My point was largely that e.g. MapReduce had literally been used in the HPC world for quite some time in countless variations. Something similar occurred with Spanner where it was a minor variant implementation on a concept that existed in many systems decades old. I worked on systems that used both of these "inventions" in the 1990s; no one thought it was novel then either. When I think of "invention" in computer science, I assume some hard problem in computer science was solved. That is not the case for most of what Google is credited with in the realm of computer science. The absence of meaningful novelty in Google's systems strikes me as a good basis for being skeptical of claims of invention. Relative obscurity of computer science does not confer invention claims on the first to popularize it. People frequently attribute more computer science invention to me than is warranted (I do a lot of high-end algorithm work). I always try to make it clear that much of what they see as "invention" is their first exposure to obscure computer science that I happen to be familiar with and know how to exploit. Much of what is attributed to Google is similar in character; people conflate the source of their introduction with the inventor of the concept.
- chubot 13y ago
  Can you give a citation for the MapReduce and Spanner predecessors? I've seen similar papers that predate MapReduce (I'm curious if you will find the ones I have). But I've also noticed that most people who think MapReduce is old hat have never used it and/or misunderstand what it actually is. And if it was already so well known, there wouldn't be hundreds of CS researchers continuing to publish both theoretical and systems of papers about it. The MapReduce paper has 7992 citations so far, and the number continues to increase every year! http://scholar.google.com/citations?view_op=view_citation&hl=en&user=NMS69lQAAAAJ&citation_for_view=NMS69lQAAAAJ:BqipwSGYUEgC http://scholar.google.com/citations?view_op=view_citation&hl... What paper should they be citing instead? It's hard to believe that 7995 sets of authors would ignore prior research, especially since some of those authors are a decade or two older than the MapReduce authors.
  - jandrewrogers 13y ago
    As a general point, a lot of computer science is poorly documented or difficult to find in the literature but well-known among the practitioners that use it. A great example is the recent flurry of activity (including patent activity) around "hyperdimensional hashing" and similar concepts. The irony is that the concepts are so old that some of the original patents have expired but the work is sufficiently forgotten that academics are attempting to re-patent those ideas. I am familiar with it because those algorithms are de rigueur for some types of massively parallel systems (I've been designing parallel systems for ages). No one publishes papers on it though because it is not new to those people. A large part of it is that much of the original literature is lost in paper form and cannot be found electronically. I've been burned by this personally, having done work that I later discovered had been previously done in the 1980s by a respected academic but which is not cited in any modern literature; from the standpoint of modern literature, the work is completely forgotten. But it had been done and rare people exist that are familiar with it. So more specifically (I got back from vacation a couple hours ago, forgive the terseness): Spanner is an old, old idea in literature and a number of bespoke systems were built on the concept. Ancient transaction processing literature mentions it, I've worked with a few crusty mainframe environments that used variants of it. It was considered a stupid idea for decades because the hardware required for a reasonable implementation was outrageously expensive for most applications. Atomic clocks and satellite-delivered time codes don't cost what they used to. I used to have a transaction processing book from the 1980s that described a couple specific implementations that were quite close to Spanner (written by a couple guys with Indian surnames, wish I remember the book). It had been written off so long ago for inexpensive commodity systems that apparently it took Google to rediscover it but it never fell out of use. Similarly, I think a lot of people in the HPC world were surprised to discover that MapReduce was patentable. Not every problem in HPC looks like computational fluid dynamics (CFD), which is often the stereotype. If you are actually an academic then you should know better than anyone that the number of cites is misleading. The MapReduce model (though not under that name) was a standard design pattern for some non-CFD problems in HPC in every respect going back to at least the 1990s. The specific novelty claimed by MapReduce has never been clear to me. A lot of the (terrible) compilers for obscure HPC platforms back in the day transformed the code into a canonical MapReduce model. Maybe I am missing what makes Google's version unique?
- jamesaguilar 13y ago
  I don't think you've got the right understanding of why people are against sw patents. The argument is that applying an idea to a slightly different domain already provides enough value that it doesn't need to be encouraged by patent protection. I don't hear many people saying it has little value. The number of companies supporting themselves completely on the proceeds of selling key-value store software and services trivially shows that there is a lot of economic value in non-innovative implementations of ideas.
- cromwellian 13y ago
  If Spanner is a "variant" than it introduced something new, an improvement. You sure are jumping through hoops trying to diminish things. Most of computer science was invented in the 50, 60s and 70s when the low hanging fruit was easy to get, now a lot of it is incremental gains. Just like the astounding pace of advances in mathematics in the 1700s/1800s/1900s can't be bested in the 00s, progress is slower as the easy stuff has been mapped out already. Besides which, there's a difference between inventing say, Newtonian Physics, and building an interplanetary ship for human space flight. It is in the application of much of computer science to extremely large systems where most of the work is being done, and frankly, experienced gathered building the world's largest database and most trafficked site gives you insight into new ways to engineer things that you don't necessarily figure out from basic application of CS principles. I could probably design a basic bridge or house given what I learned of civil engineering, but chances are I could never design the Burj Dubai with that information. My point is, the way Google applied things like MapReduce was based on their experience of solving web services problems. HPC hyper-computing typically looks at problems of simulation, like QED/QCD, nuclear explosions, climate, finance, oil geology, et al. The difference focus leads to different pattern of application, and there is new value created in the papers published by both sides. Google's datacenter configurations don't look anything like say, BlueGene. And when you look at what Google is doing with machine learning, few organizations have the resources and data to pull it off. Sure, lots of people have run the algorithms before on small sets, but few actually have daily data from a billion people flowing into them.

by karterk 13y ago

jandrewrogers 13y ago

The weakness of this contribution is often cited as the reason many types of software patents should be invalid. Putting a shiny wrapper on an old idea in an obvious but slightly different context is something upon which many software hackers would normally cast a jaundiced eye. My point was largely that e.g. MapReduce had literally been used in the HPC world for quite some time in countless variations. Something similar occurred with Spanner where it was a minor variant implementation on a concept that existed in many systems decades old. I worked on systems that used both of these "inventions" in the 1990s; no one thought it was novel then either. When I think of "invention" in computer science, I assume some hard problem in computer science was solved. That is not the case for most of what Google is credited with in the realm of computer science. The absence of meaningful novelty in Google's systems strikes me as a good basis for being skeptical of claims of invention. Relative obscurity of computer science does not confer invention claims on the first to popularize it. People frequently attribute more computer science invention to me than is warranted (I do a lot of high-end algorithm work). I always try to make it clear that much of what they see as "invention" is their first exposure to obscure computer science that I happen to be familiar with and know how to exploit. Much of what is attributed to Google is similar in character; people conflate the source of their introduction with the inventor of the concept.

chubot 13y ago
Can you give a citation for the MapReduce and Spanner predecessors? I've seen similar papers that predate MapReduce (I'm curious if you will find the ones I have). But I've also noticed that most people who think MapReduce is old hat have never used it and/or misunderstand what it actually is. And if it was already so well known, there wouldn't be hundreds of CS researchers continuing to publish both theoretical and systems of papers about it. The MapReduce paper has 7992 citations so far, and the number continues to increase every year! http://scholar.google.com/citations?view_op=view_citation&hl=en&user=NMS69lQAAAAJ&citation_for_view=NMS69lQAAAAJ:BqipwSGYUEgC http://scholar.google.com/citations?view_op=view_citation&hl... What paper should they be citing instead? It's hard to believe that 7995 sets of authors would ignore prior research, especially since some of those authors are a decade or two older than the MapReduce authors.
- jandrewrogers 13y ago
  As a general point, a lot of computer science is poorly documented or difficult to find in the literature but well-known among the practitioners that use it. A great example is the recent flurry of activity (including patent activity) around "hyperdimensional hashing" and similar concepts. The irony is that the concepts are so old that some of the original patents have expired but the work is sufficiently forgotten that academics are attempting to re-patent those ideas. I am familiar with it because those algorithms are de rigueur for some types of massively parallel systems (I've been designing parallel systems for ages). No one publishes papers on it though because it is not new to those people. A large part of it is that much of the original literature is lost in paper form and cannot be found electronically. I've been burned by this personally, having done work that I later discovered had been previously done in the 1980s by a respected academic but which is not cited in any modern literature; from the standpoint of modern literature, the work is completely forgotten. But it had been done and rare people exist that are familiar with it. So more specifically (I got back from vacation a couple hours ago, forgive the terseness): Spanner is an old, old idea in literature and a number of bespoke systems were built on the concept. Ancient transaction processing literature mentions it, I've worked with a few crusty mainframe environments that used variants of it. It was considered a stupid idea for decades because the hardware required for a reasonable implementation was outrageously expensive for most applications. Atomic clocks and satellite-delivered time codes don't cost what they used to. I used to have a transaction processing book from the 1980s that described a couple specific implementations that were quite close to Spanner (written by a couple guys with Indian surnames, wish I remember the book). It had been written off so long ago for inexpensive commodity systems that apparently it took Google to rediscover it but it never fell out of use. Similarly, I think a lot of people in the HPC world were surprised to discover that MapReduce was patentable. Not every problem in HPC looks like computational fluid dynamics (CFD), which is often the stereotype. If you are actually an academic then you should know better than anyone that the number of cites is misleading. The MapReduce model (though not under that name) was a standard design pattern for some non-CFD problems in HPC in every respect going back to at least the 1990s. The specific novelty claimed by MapReduce has never been clear to me. A lot of the (terrible) compilers for obscure HPC platforms back in the day transformed the code into a canonical MapReduce model. Maybe I am missing what makes Google's version unique?
jamesaguilar 13y ago
I don't think you've got the right understanding of why people are against sw patents. The argument is that applying an idea to a slightly different domain already provides enough value that it doesn't need to be encouraged by patent protection. I don't hear many people saying it has little value. The number of companies supporting themselves completely on the proceeds of selling key-value store software and services trivially shows that there is a lot of economic value in non-innovative implementations of ideas.
cromwellian 13y ago
If Spanner is a "variant" than it introduced something new, an improvement. You sure are jumping through hoops trying to diminish things. Most of computer science was invented in the 50, 60s and 70s when the low hanging fruit was easy to get, now a lot of it is incremental gains. Just like the astounding pace of advances in mathematics in the 1700s/1800s/1900s can't be bested in the 00s, progress is slower as the easy stuff has been mapped out already. Besides which, there's a difference between inventing say, Newtonian Physics, and building an interplanetary ship for human space flight. It is in the application of much of computer science to extremely large systems where most of the work is being done, and frankly, experienced gathered building the world's largest database and most trafficked site gives you insight into new ways to engineer things that you don't necessarily figure out from basic application of CS principles. I could probably design a basic bridge or house given what I learned of civil engineering, but chances are I could never design the Burj Dubai with that information. My point is, the way Google applied things like MapReduce was based on their experience of solving web services problems. HPC hyper-computing typically looks at problems of simulation, like QED/QCD, nuclear explosions, climate, finance, oil geology, et al. The difference focus leads to different pattern of application, and there is new value created in the papers published by both sides. Google's datacenter configurations don't look anything like say, BlueGene. And when you look at what Google is doing with machine learning, few organizations have the resources and data to pull it off. Sure, lots of people have run the algorithms before on small sets, but few actually have daily data from a billion people flowing into them.