Apparently, the "study" is an exercise in pulling numbers out of thin air then.
How long ago was it that the MS/Hotmail internal paper was leaked showing that administration of the large server farm was a nightmare with Windows 2000 and that with Open Source software (FreeBSD in this case, ISTR), it was vastly simpler and consequently required far fewer administration resources?
If OSS takes a fraction of the admin resources, and is robust and reliable, offering potentially lower downtime, *and* by their own volition these account for the vast majority of the cost (also disputed in the MS/Hotmail paper), then unless they're paying the OSS admins six-figure salaries and the Windows admins are on minimum wage, then it simply doesn't add up.
Judging by the results that this technique has yielded so far, I'd say the results are very impressive, especially given than the Mr. Rossmo takes great pains to point out the limitations and that the software is only truly effective as one part of the array of mathods being used in such cases.
It is apparent from your comment that you have a poor understanding of the large and frightening scope of mental illness. Being mentally ill does not necessarily make the victim totally irrational. Consequently, it is not at all surprising that such a technique might well be effective. If your illness is so debilitating that you effectively behave "randomly", I submit that you would not long evade capture.
Regarding the unabomber. Is there any evidence to suggest that technique was employed in that case (I don't think so) ? How do you know if this might not have been effective there ?
The difference is you're saying what should be, and Grishnakh is saying what is. Intel *should* be doing whatever it takes to make the code that gcc produces for ia64 optimal, but they won't because they're being greedy and believe that they can sell the compiler as well.
Maybe they're right, and maybe they're wrong, but given the lack of uptake on ia64 so far, and given the relative amounts of money that they stand to make on the compiler software versus the ia64 hardware, I think they're making a big mistake.
It rather looks like a case of company politics, with nobody able or willing to see the big picture and "do the right thing" regarding "giving away" the compiler technology.
Tim
Pure science and funding.
on
Ununoctium Wrapup
·
· Score: 2, Interesting
It seems to me that a big problem regarding objectivity in science (which, let's face it, is a fundamental aspect of science), is the source of funding. Scientists are much more likely to attempt fraudulent activity, or cling to theories based around results which fit their model, conveniently suppresing those that do not, if they stand to lose their funding if they do "the right thing".
Not sure that there's an obvious solution here, although peer review works well a lot of the time, but it seems to me that this is becoming more of an issue.
If you're talking about OCFS for Linux (announced at LinuxWorld), it's nothing like AdvFS or a "real" clustered filesystem. It is intended to allow Oracle 9i RAC to operate in a filesystem environment and is explicitly documented as not being intended for use as a true filesystem replacement. A brief glance at the current implementation will show you why. Although it offers most of the standard filesystem semantics, performance for use as a general-purpose filesystem (particularly metadata-update intensive) would be quite horrible. Since there is hardly any metadata activity when running Oracle on the filesystem, this would not be a problem.
I have to concur with this. I very much prefer the IBM 'T' series laptops, because they are, IMHO, vastly better engineered, and they don't rip you off to upgrade. You want to swap the HD in the IBM. Undo the screw, swap the drive - you're done. The Sony ? Oh, that's a $700 return to factory upgrade/repair.
The Sony machines look cute, but the blend of gratuitously incompatible hardware, expense repair/upgrade costs etc. leave me unimpressed. Oh, I also hate the keyboard layout on the PCG-R505 - the tiny right-hand shift key makes typing very unpleasant.
Happy to say, someone will be taking my Sony, and I'll be getting an IBM in return pretty soon:-)
You are quite correct. All the currently shipping products have a metadata server which introduces both a performance bottleneck and in most cases a single point of failure.
<Shameless plug>
I work for a company developing a filesystem which is truly symmetric, and hence does not suffer from these issues. If you are at all interested, check out http://www.polyserve.com/products/pmxs.html for details. It's currently in Beta.
</Shameless plug>
Don't ever wire up a cable with (1,2) (3,4), (5,6), (7,8) as pairs if you expect it to work and it;s any kind of length. I found this out the hard way. It has to be (1,2), (3,6), (4,5), and (7,8). Kind of obvious when you know which conductors get used, but rather painful if you don't !
I can't speak to rejection of stories or otherwise, but AIX 5L is not presently available for x86, and I'm pretty sure that this situation is not going to change. You can lookup the details at the AIX 5L Version 5.1 web site.
On the contrary, the GNU license makes it expressly clear that you must make the sources available. It doesn't have to be for free. They don't have to be up on the web. You're allowed to charge a "reasonable" fee for providing them, but if you "distribute" (i.e. provide to people outside of you organiztaion) binaries of GPL'd code, you are required to make the source available.
ecause unless you're terribly interested in the exact code that's running underneath, your perception of what it is is determined by what you see.
If the API etc. presented is indistinguishable for that which you would find on a "real" Linux system, then, as far as you are concerned, it's a Linux system.
Of course, in reality, life is a little more complex. Things like fileystem layout, system management related code etc. tend to not match, but it's possible to create a "Linux" system that doesn't use the Linux kernel. In extermis, one could re-implement that Linux kernel from scratch. I have no idea why you would want to, but it would, assuming you did it correctly, for that moment in time be indistinguishable from "the real thing":)
Nope, AIX5L was developed entirely within IBM. I'm all but certain that there's no Caldera tie-in.
AIX5L doesn't run on ia32 hardware, so the focus is quite different. Caldera have the ability to run Linux ia32 binaries, wherea AIX5L foxusses on providing a familiar GNU environment and a sufficiently rich library/header environment as to allow easy porting of software that already runs under Linux.
One tiny correction. There is not a single MPEG chip, there are two. The IBM chip is the decoder and is also well-documented. The MPEG2 encoder is a Sony chip and documentation does not appear to be freely available.
So, recording and playing different streams is not hard with this architecture. Getting it all to work without underruns is not entirely trivial.
Personally, I'm not a fan of the current patent system what grants many "obvious" ideas patents, but in the grand scheme of things, the TiVo patents have more to them than many. At least, that's my opinion, I could be wrong:-)
Just another happy TiVo user (hacked HDR212 with 100GB disk space)
1. What utter nonsense ! The CPUs in Intel 4-way and 8-way boxes are the same, and furthermore they dominate the cost. Check the price of a 2MB L2 cache Xeon to verify this. The only difference is the addition of the Corollary Profusion chipset - hardly a big hitter in cost. Admittedly more 4-way boxes are sold, so you'll get a slight break here, but your first point is not terribly valid. Check IBM or Compaq's prices to see.
2. That's probably true. Linux support for hot-plug SCSI is weak right now. Don't expect it to stay that way for long. It isn't exactly rocket science to develop a driver to talk to the SCSI hot-plug controller:-)
3. Actually, it wasn't just OS cost in this case. Check the database software cost. It's pretty significant. I still think you're right and that Win2k will win on a single 4-way at the moment, but the scalability of the clustered solution is impressive. Of course, I suspect this has a lot to do with DB2 EEE edition, since it is highly optimized for this kind of setup (think IBM SP2).
I know that work is going on regarding Linux support for HP printers. The one major frustration I still face is the lack of support of the HP PhotoSmart S20 scanner (the USB one).
It's a neat little scanner, producing pretty decent quality scans (not as good as the new Nikon scanners, but...). However, unlike it's SCSI predecessor, it doesn't use HP SCL, and there's no documentation, making completely useless under Linux - no support in SANE.
Anybody know if this situation is ever likely to change or how one might obtain documentation to fix it ???
Kbits/s or Kbytes/s ?
I recently stayed at the Marriott Marquis in Manhatten for the Linuxworld Expo and the STSN hookup there was getting ~70Kbytes/s !
That's a great deal faster than dialup and allowed me to download ~200MB of updates after I had to re-install my laptop following an APM/UDMA snafu (incredibly badly trashed root partition).
Dunno how the latest snapshots are doing, but historically, the ia32 gcc compiler has produced *really* lousy code for 64-bit long long data.
So much so, that Linus has shunned use of 'long long' in the kernel as much as possible. This is rather a pity. I hope that someone improves this in the future.
Hi Rob,
the T20 contains a Lucent winmodem (assuming it's the mini-pci modem supplied with the machine).
Lucent don't exactly support it, but it appears that there's at least one person at Lucent who supports the Linux community. They produced a Linux driver (infrastructure source, binary dsp module), and it's been tweaked (debugged) by some folks out on the net.
For general Linux "winmodem" information, check out the linmodems website. This site links to the updated Lucent drivers. I'm using them successfully on my T20.
The newer Thinkpads don't have an MWave chip at all. IIRC the last model to use it was the 600E. The newer models don't actually integrate a modem at all. The sound chipsets tend to be Cirrus Logic based (CS42xx + AC97), and the modem is supplied as a (replaceable) mini-pci card.
The card supplied is a Lucent "winmodem". You can also get a card that is a combined ethernet/modem mini-pci card.
Oh for heavens sake !
How is this informative ?
It has been pointed out on more than one occasion that the OS/2 version of JFS is actually newer than the AIX version. You make it sound as if it were somehow inferior.
If you go and actually grab the JFS code you will notice that case sensitivity was on the todo list, and is fixed as of something like drop 13.
Elsewhere, somebody posted a link to an article from the Linux Gazette that gives a carefully reasearched comparison. Your characterization of JFS implies you know little about it:-(
Well, I'm certainly not an astroturfer, but I have had no problems with RADSL in Portland. I'm on the west side in GTE (now Verizon) territory. Initial setup had a slight hiccup due to the local loop part being owned by a different company, but once we got past that, I've had practically uninterrupted service.
Recently, I had an outage, and phoned up the helpdesk. I was almost immediately connected to a tech who was polite and helpful. He quickly traced that there were frames on the line from my ADSL modem, and I found to my embarrassment that the crossover cable had come loose (probably the cat:-)
So, whilst there do seem to be a large number of horror stories out there, it is possible to get decent service with DSL. I hope you are lucky and get the same.
So,
the National average was $45,000, but the local salary was $35,000. Unless you're trying to tell me that everywhere pays the same, or that, conveniently, the particular example just happened to be at the median, then this is completely meaningless. You can bet that the amount is higher in California than, say, Ohio !
As to the fees, unless you're suggesting that H1-B hires paid for their own travel etc., you'll find that a combination of air fares, accomodation, shipping of personal effects, hire car etc. usually amount to at least $20,000. Hardly the $1000 quoted.
From what you quoted, I find their research to be seriously flawed.
Tim
RAM memory in servers generally has ECC circuitry which will correct all single-bit errors, and detect all double-bit errors. In fact, you'd be highly ill-advised to buy a server that you used for serious business that didn't have this.
However, the problem here is not the main memory. It's the cache memory (data & tags) for the CPU. The whole point of the cache memory is that it's supposed to be fast. Adding ECC would slow it down too much. That's why it's only parity protected. You detect a single-bit error, but at that point you're toast.
The suggested solution mentioned elsewhere is to mirror the cache, i.e. everything goes to both sides, and if you get a parity error on one side, you take data from the other and "scrub" the bad side.
Slashdot Mirror
Hmmm...
wonder how careful they were in programming the server app that listens for these "tattle-tale" CDs.
Just a thought
Tim
Apparently, the "study" is an exercise in pulling numbers out of thin air then.
How long ago was it that the MS/Hotmail internal paper was leaked showing that administration of the large server farm was a nightmare with Windows 2000 and that with Open Source software (FreeBSD in this case, ISTR), it was vastly simpler and consequently required far fewer administration resources?
If OSS takes a fraction of the admin resources, and is robust and reliable, offering potentially lower downtime, *and* by their own volition these account for the vast majority of the cost (also disputed in the MS/Hotmail paper), then unless they're paying the OSS admins six-figure salaries and the Windows admins are on minimum wage, then it simply doesn't add up.
So who funded this "study" ?
Judging by the results that this technique has yielded so far, I'd say the results are very impressive, especially given than the Mr. Rossmo takes great pains to point out the limitations and that the software is only truly effective as one part of the array of mathods being used in such cases.
It is apparent from your comment that you have a poor understanding of the large and frightening scope of mental illness. Being mentally ill does not necessarily make the victim totally irrational. Consequently, it is not at all surprising that such a technique might well be effective. If your illness is so debilitating that you effectively behave "randomly", I submit that you would not long evade capture.
Regarding the unabomber. Is there any evidence to suggest that technique was employed in that case (I don't think so) ? How do you know if this might not have been effective there ?
Tim
The difference is you're saying what should be, and Grishnakh is saying what is. Intel *should* be doing whatever it takes to make the code that gcc produces for ia64 optimal, but they won't because they're being greedy and believe that they can sell the compiler as well.
Maybe they're right, and maybe they're wrong, but given the lack of uptake on ia64 so far, and given the relative amounts of money that they stand to make on the compiler software versus the ia64 hardware, I think they're making a big mistake.
It rather looks like a case of company politics, with nobody able or willing to see the big picture and "do the right thing" regarding "giving away" the compiler technology.
Tim
It seems to me that a big problem regarding objectivity in science (which, let's face it, is a fundamental aspect of science), is the source of funding. Scientists are much more likely to attempt fraudulent activity, or cling to theories based around results which fit their model, conveniently suppresing those that do not, if they stand to lose their funding if they do "the right thing".
Not sure that there's an obvious solution here, although peer review works well a lot of the time, but it seems to me that this is becoming more of an issue.
If you're talking about OCFS for Linux (announced at LinuxWorld), it's nothing like AdvFS or a "real" clustered filesystem. It is intended to allow Oracle 9i RAC to operate in a filesystem environment and is explicitly documented as not being intended for use as a true filesystem replacement. A brief glance at the current implementation will show you why. Although it offers most of the standard filesystem semantics, performance for use as a general-purpose filesystem (particularly metadata-update intensive) would be quite horrible. Since there is hardly any metadata activity when running Oracle on the filesystem, this would not be a problem.
Tim
I have to concur with this. I very much prefer the IBM 'T' series laptops, because they are, IMHO, vastly better engineered, and they don't rip you off to upgrade. You want to swap the HD in the IBM. Undo the screw, swap the drive - you're done. The Sony ? Oh, that's a $700 return to factory upgrade/repair.
:-)
The Sony machines look cute, but the blend of gratuitously incompatible hardware, expense repair/upgrade costs etc. leave me unimpressed. Oh, I also hate the keyboard layout on the PCG-R505 - the tiny right-hand shift key makes typing very unpleasant.
Happy to say, someone will be taking my Sony, and I'll be getting an IBM in return pretty soon
Tim
<Shameless plug>
I work for a company developing a filesystem which is truly symmetric, and hence does not suffer from these issues. If you are at all interested, check out http://www.polyserve.com/products/pmxs.html for details. It's currently in Beta.
</Shameless plug>
Regards,
Tim
pins 1&2 are one pair, 3&6 are the other.
:-)
Don't ever wire up a cable with (1,2) (3,4), (5,6), (7,8) as pairs if you expect it to work and it;s any kind of length. I found this out the hard way. It has to be (1,2), (3,6), (4,5), and (7,8). Kind of obvious when you know which conductors get used, but rather painful if you don't !
Just FYI
Tim
Not true.
They're coming out as we speak. Everquest "Shadows of Luclin" is out on December the 4th and it's a DX8-based game.
Tim
Tim
Tim
If the API etc. presented is indistinguishable for that which you would find on a "real" Linux system, then, as far as you are concerned, it's a Linux system.
Of course, in reality, life is a little more complex. Things like fileystem layout, system management related code etc. tend to not match, but it's possible to create a "Linux" system that doesn't use the Linux kernel. In extermis, one could re-implement that Linux kernel from scratch. I have no idea why you would want to, but it would, assuming you did it correctly, for that moment in time be indistinguishable from "the real thing" :)
Tim
Nope, AIX5L was developed entirely within IBM. I'm all but certain that there's no Caldera tie-in.
AIX5L doesn't run on ia32 hardware, so the focus is quite different. Caldera have the ability to run Linux ia32 binaries, wherea AIX5L foxusses on providing a familiar GNU environment and a sufficiently rich library/header environment as to allow easy porting of software that already runs under Linux.
Tim
So, recording and playing different streams is not hard with this architecture. Getting it all to work without underruns is not entirely trivial.
Personally, I'm not a fan of the current patent system what grants many "obvious" ideas patents, but in the grand scheme of things, the TiVo patents have more to them than many. At least, that's my opinion, I could be wrong :-)
Just another happy TiVo user (hacked HDR212 with 100GB disk space)
Tim
2. That's probably true. Linux support for hot-plug SCSI is weak right now. Don't expect it to stay that way for long. It isn't exactly rocket science to develop a driver to talk to the SCSI hot-plug controller :-)
3. Actually, it wasn't just OS cost in this case. Check the database software cost. It's pretty significant. I still think you're right and that Win2k will win on a single 4-way at the moment, but the scalability of the clustered solution is impressive. Of course, I suspect this has a lot to do with DB2 EEE edition, since it is highly optimized for this kind of setup (think IBM SP2).
Tim
It's a neat little scanner, producing pretty decent quality scans (not as good as the new Nikon scanners, but ...). However, unlike it's SCSI predecessor, it doesn't use HP SCL, and there's no documentation, making completely useless under Linux - no support in SANE.
Anybody know if this situation is ever likely to change or how one might obtain documentation to fix it ???
Tim
Kbits/s or Kbytes/s ?
:-)
I recently stayed at the Marriott Marquis in Manhatten for the Linuxworld Expo and the STSN hookup there was getting ~70Kbytes/s !
That's a great deal faster than dialup and allowed me to download ~200MB of updates after I had to re-install my laptop following an APM/UDMA snafu (incredibly badly trashed root partition).
YMMV
Tim
Dunno how the latest snapshots are doing, but historically, the ia32 gcc compiler has produced *really* lousy code for 64-bit long long data.
So much so, that Linus has shunned use of 'long long' in the kernel as much as possible. This is rather a pity. I hope that someone improves this in the future.
Tim
the T20 contains a Lucent winmodem (assuming it's the mini-pci modem supplied with the machine).
Lucent don't exactly support it, but it appears that there's at least one person at Lucent who supports the Linux community. They produced a Linux driver (infrastructure source, binary dsp module), and it's been tweaked (debugged) by some folks out on the net.
For general Linux "winmodem" information, check out the linmodems website. This site links to the updated Lucent drivers. I'm using them successfully on my T20.
Tim
The card supplied is a Lucent "winmodem". You can also get a card that is a combined ethernet/modem mini-pci card.
Tim
Oh for heavens sake ! :-(
How is this informative ?
It has been pointed out on more than one occasion that the OS/2 version of JFS is actually newer than the AIX version. You make it sound as if it were somehow inferior.
If you go and actually grab the JFS code you will notice that case sensitivity was on the todo list, and is fixed as of something like drop 13.
Elsewhere, somebody posted a link to an article from the Linux Gazette that gives a carefully reasearched comparison. Your characterization of JFS implies you know little about it
Tim
Well, I'm certainly not an astroturfer, but I have had no problems with RADSL in Portland. I'm on the west side in GTE (now Verizon) territory. Initial setup had a slight hiccup due to the local loop part being owned by a different company, but once we got past that, I've had practically uninterrupted service. :-)
Recently, I had an outage, and phoned up the helpdesk. I was almost immediately connected to a tech who was polite and helpful. He quickly traced that there were frames on the line from my ADSL modem, and I found to my embarrassment that the crossover cable had come loose (probably the cat
So, whilst there do seem to be a large number of horror stories out there, it is possible to get decent service with DSL. I hope you are lucky and get the same.
Regards,
Tim
So,
the National average was $45,000, but the local salary was $35,000. Unless you're trying to tell me that everywhere pays the same, or that, conveniently, the particular example just happened to be at the median, then this is completely meaningless. You can bet that the amount is higher in California than, say, Ohio !
As to the fees, unless you're suggesting that H1-B hires paid for their own travel etc., you'll find that a combination of air fares, accomodation, shipping of personal effects, hire car etc. usually amount to at least $20,000. Hardly the $1000 quoted.
From what you quoted, I find their research to be seriously flawed.
Tim
RAM memory in servers generally has ECC circuitry which will correct all single-bit errors, and detect all double-bit errors. In fact, you'd be highly ill-advised to buy a server that you used for serious business that didn't have this.
However, the problem here is not the main memory. It's the cache memory (data & tags) for the CPU. The whole point of the cache memory is that it's supposed to be fast. Adding ECC would slow it down too much. That's why it's only parity protected. You detect a single-bit error, but at that point you're toast.
The suggested solution mentioned elsewhere is to mirror the cache, i.e. everything goes to both sides, and if you get a parity error on one side, you take data from the other and "scrub" the bad side.
Tim