Cosmos: big data and big challenges.
Pat Helland talked about massively parallel processing based on Dryad using a Petabyte store and made the point that these massive systems process information differently. Database processing in this environment involves tools like SCOPE, which is an SQL-like language that has been optimized for execution over Dryad. Saving this data long term is a problem because they are worried about bit rot. Cosmos keeps at least three copies of the data, checks them regularly, and replaces data that is damaged. Interesting how close this is to LOCKSS (which saves 7 copies).
In talking about how faster processing is not always the solution to processing problems, a speaker quoted Henry Ford as saying: “If I had asked my customers what they wanted, they would have said faster horses... “ (source: Eric Haseltine Long Fuse and Big Bang.)
NUI (Natural User Interfaces)
One of the speakers distinguished between making user interfaces imitate nature and making them feel natural. Non-verbal clues that convey meaning need to be part of the interaction. Another speaker said that we need better feedback systems. An example is a touch screen with a single button. If a user touches it and nothing happens, the user will hit it again and again. When designers changed the button to send out sparkles when touched, the repeated touching stopped, even though sparkles are not a natural result of touching a screen.
In the discussion someone said that we aren't doing science if we can't go back to the data. This suggests a clear separation of data and processing that speakers about very large data sets said is no longer really possible, since the data is usable only with a degree of processing. The speaker was doing fundamentally social science research, which is more human-readable than Big Science data.
Other comments of interest:
Microsoft clearly has a strong interest in image management, particularly three-dimensional images such are used in medical imaging (doubtless a good market) or gaming. They are dividing a picture into quadrants and creating mathematical representations of the edges in each square to create a hash to search for similar photos. Photosynth.net was also demoed -- it allows the creation of three dimensional images from multiple photos.
Evening Cruise
Microsoft has planned a dinner cruise for the evening. It should be pleasant (I will comment tomorrow), but many of us wanted to go back to the hotel to leave computers, etc., and to change clothes because it is fairly chilly out (despite the heat wave in the rest of the US).
In the discussion someone said that we aren't doing science if we can't go back to the data. This suggests a clear separation of data and processing that speakers about very large data sets said is no longer really possible, since the data is usable only with a degree of processing. The speaker was doing fundamentally social science research, which is more human-readable than Big Science data.
Other comments of interest:
- Do we need to take the “good” into account in our interpretation of the “natural”?
- It's not a machine that we are interfacing to any more, though the speaker is not sure what it is exactly. There is no machine, but a task.
Microsoft clearly has a strong interest in image management, particularly three-dimensional images such are used in medical imaging (doubtless a good market) or gaming. They are dividing a picture into quadrants and creating mathematical representations of the edges in each square to create a hash to search for similar photos. Photosynth.net was also demoed -- it allows the creation of three dimensional images from multiple photos.
Evening Cruise
Microsoft has planned a dinner cruise for the evening. It should be pleasant (I will comment tomorrow), but many of us wanted to go back to the hotel to leave computers, etc., and to change clothes because it is fairly chilly out (despite the heat wave in the rest of the US).
No comments:
Post a Comment