I haven't heard of any great indian computer scientist. The good ones will always survive.
How about
Arvind (he's pretty famous systems, parallel computing and architecture type), or Rajeev Motwani (famous database theory guy, well known for algorithms work too), Raj Reddy who won a Turing award for his work in 1994 for his work in Robotics/AI, Narendra Karmarkar in optimization theory (linear programming) or Raj Jain for performance analysis and network design? There are MANY famous Indian computer
scientists (don't even get me started on data mining, that community is FULL of them), and although many now live outside India, some are returning (e.g Krithi Ramamritham).
The Linux Trace Toolkit (LTT) is a neat kernel debugging tool, however, it hasn't been able to get inclusion into the main kernel source tree. I'd like to ask if there was any buzz about LTT during the conference and if there is any sense about whether LTT is likely to get included into the source tree or not any time soon?
I agree with your points, however, auto manufacture is fundamentally different from media due to technological issues. In particular, economy of scale and cost of entry also impact diversity. Some areas (like processor design) have become so expensive that technological issues are driving out the little guy (who can't just run a Fab in his garage).
Actually, multicore processors and hyperthreading (if I understand correctly) are the result of hardware manufacturer's giving up on single core models. The problem with single core models is that cache misses and incorrect branch predictions must either stall the processor or be handled by a dynamic scheduler. Dynamic scheduling has a high gate count (does it grow exponentially with the number of stalled instructions that must be tolerated?), and as such has rapidly diminishing returns (increasing cache size also has a diminishing return on hit rate in practice). With high clock rates and the need for short propagation distances, latency tolerance (the goal of hyperthreading or simultaneous multithreading) and concurrency (the goal of multicore systems) are classical approaches (albeit with new application to processor design), since the device size is shrinking, speed is increasing, but so is clock speed.
No. Or, at least, that's what we already do and it's called "factoring." Factoring large composite numbers is
hard .
Isn't factoring widely believed to be hard (in the
computational complexity sense). I don't think it is proven (in the sense that
certain problems are uncomputable or known
to be in one of the NP classes).
One slick little feature I've noticed on 9.0 is the new desktop launcher Icon in the SuS KDE Menu. The ability to start a new session without logging out a colleague who stepped away from the machine is helpful in my lab.
The real reason they're keeping the data is the way google's distributed file system work
That argument is bunk.
With a system like that, you could implement a system where "deleted" chunks get purged or overwritten on some semi-regular basis.
It is possible that the grand parent poster did not get the motivation for making the users agree to allow google to maintain the data correct. However, his assertion that backups, checkpointing, caching and distributed storage cause privacy concerns is accurate, and it is a hard problem.
HIPAA (U.S. law which includes penalties for disclosing confidential health information)and
other regulations have caused serious concerns in the database and data mining research community (contrary to popular opinion not all data miners want to strip you of your privacy). Rakesh Agrawal, Jerry Kiernan and
Ramakrishnan Srikant (major data base researchers, Srikant and Agrawal made groundbreaking contributions to the field of data mining) published recently (well about 2 years ago or so) on Hippocratic Databases (in gzipped PostScript),
where he describes the hard problem of making
the database forget information to conform to legal and ethical restrictions.
Psion's corporate history page
states that they have been making small form
factor devices for 20 years.
About 15 years ago,
if memory serves (this would be about 15 years ago or so) some Psion devices were targeting inventory management applications and warehousing. The idea was that people would carry around a large PDA form factor (or sublaptop form factor) device with an inventory entry system and then hot sync (wireless and internet weren't so widely available back then).
I'm not sure if their mindset was really geared toward consumer devices, but they've been in the small form factor mobile computing area for a long time.
I suspect that new users find Unix/Linux challenging due to historical reasons.
Two problems that stick out are command naming
conventions and command output.
It is an unfortunate artifact of Unix history that some of the commands are poorly named. The Help command did exist in Unix, but it was the help system for sccs (too bad there wasn't sccs --help or some such convention instead). There are a few other pet peeves of mine, grep might be better named search, etc.
Command output is problematic since users often
expect feedback. For example when we grep on a file and don't find the pattern, grep does not generate any output. From a programmer point of view that is definitely the right thing (especially since these commands are used as filters in a pipelined fashion), however, from a user centric point of view there may be an expectation of a report that the pattern was not found. I'm pretty used to the Unix/Linux way of
doing things, but new users are not.
Interestingly New York, New Jersey, Connecticut and Massachusetts don't have any intersections listed under the little pull down menu where you can browse by state. I'm sure driving in Boston,
NYC and surrounding Tri State area or Northern New Jersey is so safe that there aren't any statistics indicating dangerous intersections there.
However, I do agree that Philadelphia has
extrodinarily dangerous traffic patterns. I can remember many white knuckle experiences on the Schuykill expressway when I worked down by Philadelphia, but I guess exit ramps and toll booths
(especially by the bridge) don't count. I do remember taking Admiral Wilson Blvd. on the New Jersey side, and noting the large number of scantily clad hookers working along the side of the road. Even if they aren't attractive, it is really hard to avoid the instinct to look. One time I recall reading in the Inquirer that the police rounded up these girls after noticing an unusually large increase in the number of accidents in the area where they advertised.
For a more recent treatment, I found an interesting article/lecture notes, DeBeers and Beyond by Luis M. B. Cabral
of NYU's dept of economics, describing sanctions
on individuals (e.g. Israeli diamond investors and Australian Diamond Miners) and
against governments (e.g. Zaire and Russia).
Apparently the "diamonds are forever" slogan means don't resell your diamonds and don't use them for investment (since your sales reduce the DeBeers cartel's profits).
In practice, it is the standards internalized by the prof. and not some carrot/stick held by the University that motivates good teaching.
A bad administration can sap the will of even a very promising energetic new faculty.
In general, Research Universities in the U.S. have disincentives for poor teaching (e.g. if you don't know your stuff, sexually harass your students or do something really inappropriate), but don't have strong incentives for excellent teaching. Usually good researchers can explain what they do well, there are a few who are not so articulate.
So what's wrong with this old song and dance?./configure && make && make install
No easy dependency tracking, no easy uninstall, no easy upgrade, no audit trail.
Actually, from my point of view, the biggest reason to use a distribution's package manager is that (hopefully) the packager did some integration testing of software versions. I've had a number of unpleasant surprises in the past, but generally they tend to be less than if I pull down packages and just build them from source without some other guys going in and picking (hopefully) compatible versions for me.
I agree with waiting to specialize. One thing to think about in selecting an undergraduate school is what kind of environment you want, and what you want to do after graduation.
Big name research schools often focus on their
Ph.D. programs, their undergrad programs may reflect this, you may wind up being taught by
grad students (some of whom may be quite good)
instead of the big names. Making contact with
the big names can be tough.
Some of the 4 year schools can offer some good teaching (e.g. Williams, Swathmore, Haverford, Colgate, etc.) and give good preparation.
Some of the really big schools may have enormous student to faculty ratios in the introductory classes, and if you aren't a self
starter, you may find it hard there.
One reason I use opera (and Konqueror too) is for the ease of setting the user agent. The Mozilla (and Galeon) browsers I've tried don't have an easy run time solution (in their menus) nor do they have a way to set the user agent on a domain by domain basis.
Actually I wondered about this. If one thinks back to how people write on paper (papyrus and parchment in the old days I guess), they used a pen/quill, which deposited wet ink and waited for it to dry.
If a right handed person uses a left-to-right written language, the letters most recently written are on the opposite side of the way the hand travels. For right-to-left languages, wouldn't there be a tendency to smudge the wet ink as the writer's hand passes over the freshly written letters?
Re:Sounds like a good idea
on
Source Code Escrow
·
· Score: 2, Interesting
One startup that I had the misfortune of dealing with failed to properly honor their source code escrow agreement and put the wrong source in the escrow (after I was done dealing with them thank goodness). They got spanked by their larger customer in court big time and went under. Most escrow agreements allow for building from the escrowed source and testing it, so deliberately checking in a crippled version is dangerous.
However, this happened in the U.S. (the buyer was German, but had a large presence in the U.S.) If a U.S. company tries to enforce an escrow agreement against an Indian vendor, I'm not sure how that would work.
Good point, just to be explicit one of the contributing factors to trigger the investigation was that the long odds made the payoff too large
(I think it was several hundred thousand US dollars).
because Linus' hate for dump has a reason behind it.
Yeah, Linus broke it (albeit in an attempt to improve file systems performance).
It works O.K. for us because our file systems
consists of a large number of (relatively small)
seldom shared files (and a few larger but
not very frequently updated files). What other
back up tools are worth using? The admins around here vastly prefer dump. Tar,dd an cpio leave a bit to be desired as far as I can see.
Does linux have an XFS dump/restore ported to it?
That makes a difference for our installation.
Currently we use ext3fs so that we can dump to tape
(in spite of Linus's hate for dump, the admin features of dump are very useful).
I'm not an expert in DBMS systems, however, I can tell you that distributed rollback is a hard problem (meaning still open) which has been extensively researched in the area of parallel and distributed discrete event simulations (PDES). In PDES they categorize the protocols as either conservative (make sure your right and then apply the transaction) or optimistic (which speculatively executes the transaction and then either rolls back or applies an inverse transaction to undo the update in the event of an incorrect transaction application).
In practice speculatively executing the updates in a PDES system tends to result in a lot of state saving (so you use lots of memory) and if the rollback is not carefully avoided, the system may have serious performance issues due to cascading rollbacks. Last I heard the issue was still open.
I realize these may be foolish questions, but I don't own any DVD Drives (but that might change). Do DVD burners/drives have an interface standard and what is the status of Linux support (e.g. is it like cdrecord)? If not, are any of these drives supported? Have DVD drive owners been happy with Linux run time support? And finally are there any good GUI wrappers, for CDs they have xcdroast, which does what I want.
I run a lab with 16 machines. I'm very seriously considering phasing out redhat in our lab for a number of reasons. I'm considering replacing it with SuSE.
The subscription model for up2date is prohibitive.
I'm a KDE user.
SuSE Professional comes with more "stuff" than RedHat (at least 7.3/8 and 9 as far as I can tell).
The integration testing of SuSE seems better than Redhat.
I can't tolerate a 1 year end of life for my software (at least on my servers).
YAST is consistent and works better than the Redhat tools.
Our university is rolling out SuSE as its
preferred Linux distro, so going with the flow
eases implementation problems.
A number of our group work with IBM, and
IBM uses SuSE (as well as Redhat) internally.
Funnily, I only tried SuSE out of a bit of an emergency when a Redhat distro upgrade on an Athlon machine on my desktop refused to boot.
I wasn't able to rollback properly, since I had converted my file systems from ext2 to ext3 during the upgrade and ran to BestBuy at 9:30 at night.
SuSE worked and has been pretty good overall.
Slashdot Mirror
The Linux Trace Toolkit (LTT) is a neat kernel debugging tool, however, it hasn't been able to get inclusion into the main kernel source tree. I'd like to ask if there was any buzz about LTT during the conference and if there is any sense about whether LTT is likely to get included into the source tree or not any time soon?
I agree with your points, however, auto manufacture is fundamentally different from media due to technological issues. In particular, economy of scale and cost of entry also impact diversity. Some areas (like processor design) have become so expensive that technological issues are driving out the little guy (who can't just run a Fab in his garage).
Actually, multicore processors and hyperthreading (if I understand correctly) are the result of hardware manufacturer's giving up on single core models.
The problem with single core models is that cache misses and incorrect branch predictions must either stall the processor or be handled by a dynamic scheduler. Dynamic scheduling has a high gate count (does it grow exponentially with the number of stalled instructions that must be tolerated?), and as such has rapidly diminishing returns (increasing cache size also has a diminishing return on hit rate in practice). With high clock rates and the need for short propagation distances, latency tolerance (the goal of hyperthreading or simultaneous multithreading) and concurrency (the goal of multicore systems) are classical approaches (albeit with new application to processor design), since the device size is shrinking, speed is increasing, but so is clock speed.
One slick little feature I've noticed on 9.0 is the new desktop launcher Icon in the SuS KDE Menu. The ability to start a new session without logging out a colleague who stepped away from the machine is helpful in my lab.
HIPAA (U.S. law which includes penalties for disclosing confidential health information)and other regulations have caused serious concerns in the database and data mining research community (contrary to popular opinion not all data miners want to strip you of your privacy). Rakesh Agrawal, Jerry Kiernan and Ramakrishnan Srikant (major data base researchers, Srikant and Agrawal made groundbreaking contributions to the field of data mining) published recently (well about 2 years ago or so) on Hippocratic Databases (in gzipped PostScript), where he describes the hard problem of making the database forget information to conform to legal and ethical restrictions.
That is interesting, now that you pointed that out. Perhaps they got stuff for performance evaluation?
Psion's corporate history page states that they have been making small form factor devices for 20 years. About 15 years ago, if memory serves (this would be about 15 years ago or so) some Psion devices were targeting inventory management applications and warehousing. The idea was that people would carry around a large PDA form factor (or sublaptop form factor) device with an inventory entry system and then hot sync (wireless and internet weren't so widely available back then). I'm not sure if their mindset was really geared toward consumer devices, but they've been in the small form factor mobile computing area for a long time.
It is an unfortunate artifact of Unix history that some of the commands are poorly named. The Help command did exist in Unix, but it was the help system for sccs (too bad there wasn't sccs --help or some such convention instead). There are a few other pet peeves of mine, grep might be better named search, etc.
Command output is problematic since users often expect feedback. For example when we grep on a file and don't find the pattern, grep does not generate any output. From a programmer point of view that is definitely the right thing (especially since these commands are used as filters in a pipelined fashion), however, from a user centric point of view there may be an expectation of a report that the pattern was not found. I'm pretty used to the Unix/Linux way of doing things, but new users are not.
However, I do agree that Philadelphia has extrodinarily dangerous traffic patterns. I can remember many white knuckle experiences on the Schuykill expressway when I worked down by Philadelphia, but I guess exit ramps and toll booths (especially by the bridge) don't count. I do remember taking Admiral Wilson Blvd. on the New Jersey side, and noting the large number of scantily clad hookers working along the side of the road. Even if they aren't attractive, it is really hard to avoid the instinct to look. One time I recall reading in the Inquirer that the police rounded up these girls after noticing an unusually large increase in the number of accidents in the area where they advertised.
For a more recent treatment, I found an interesting article/lecture notes, DeBeers and Beyond by Luis M. B. Cabral of NYU's dept of economics, describing sanctions on individuals (e.g. Israeli diamond investors and Australian Diamond Miners) and against governments (e.g. Zaire and Russia). Apparently the "diamonds are forever" slogan means don't resell your diamonds and don't use them for investment (since your sales reduce the DeBeers cartel's profits).
In general, Research Universities in the U.S. have disincentives for poor teaching (e.g. if you don't know your stuff, sexually harass your students or do something really inappropriate), but don't have strong incentives for excellent teaching. Usually good researchers can explain what they do well, there are a few who are not so articulate.
One reason I use opera (and Konqueror too) is for the ease of setting the user agent. The Mozilla (and Galeon) browsers I've tried don't have an easy run time solution (in their menus) nor do they have a way to set the user agent on a domain by domain basis.
Actually I wondered about this. If one thinks back to how people write on paper (papyrus and parchment in the old days I guess), they used a pen/quill, which deposited wet ink and waited for it to dry. If a right handed person uses a left-to-right written language, the letters most recently written are on the opposite side of the way the hand travels. For right-to-left languages, wouldn't there be a tendency to smudge the wet ink as the writer's hand passes over the freshly written letters?
However, this happened in the U.S. (the buyer was German, but had a large presence in the U.S.) If a U.S. company tries to enforce an escrow agreement against an Indian vendor, I'm not sure how that would work.
Good point, just to be explicit one of the contributing factors to trigger the investigation was that the long odds made the payoff too large (I think it was several hundred thousand US dollars).
LISP is about the same era, it might be close.
Does linux have an XFS dump/restore ported to it? That makes a difference for our installation. Currently we use ext3fs so that we can dump to tape (in spite of Linus's hate for dump, the admin features of dump are very useful).
In practice speculatively executing the updates in a PDES system tends to result in a lot of state saving (so you use lots of memory) and if the rollback is not carefully avoided, the system may have serious performance issues due to cascading rollbacks. Last I heard the issue was still open.
I realize these may be foolish questions, but I don't own any DVD Drives (but that might change). Do DVD burners/drives have an interface standard and what is the status of Linux support (e.g. is it like cdrecord)? If not, are any of these drives supported? Have DVD drive owners been happy with Linux run time support? And finally are there any good GUI wrappers, for CDs they have xcdroast, which does what I want.
- The subscription model for up2date is prohibitive.
- I'm a KDE user.
- SuSE Professional comes with more "stuff" than RedHat (at least 7.3/8 and 9 as far as I can tell).
- The integration testing of SuSE seems better than Redhat.
- I can't tolerate a 1 year end of life for my software (at least on my servers).
- YAST is consistent and works better than the Redhat tools.
- Our university is rolling out SuSE as its
preferred Linux distro, so going with the flow
eases implementation problems.
- A number of our group work with IBM, and
IBM uses SuSE (as well as Redhat) internally.
Funnily, I only tried SuSE out of a bit of an emergency when a Redhat distro upgrade on an Athlon machine on my desktop refused to boot. I wasn't able to rollback properly, since I had converted my file systems from ext2 to ext3 during the upgrade and ran to BestBuy at 9:30 at night. SuSE worked and has been pretty good overall.