Ehmmm. As maxtor stopped making the 5400 "high capacity, low performance" line of disks, we've been forced to start buying 7200rpm drives. Turns out they break down more often. We've got enough disks in the office to start doing statistics now...
SCSI tends to be engineered "high performance", which translates to heat, which might cause premature breakdowns. I'm not so sure I like high performance SCSI that much.....
After RTFA, I concluded that these two companies are entering a research cooperation to try and develop something that might make it into the market within a decade or so.
What did I misread? Where does it say they are going into production???
As a matter of fact, if the atmosphere and terrain were not an issue, you COULD do an orbit a hundred feet off the ground. And you could enter this orbit by going straight sideways. It just requires moving a lot faster than a higher orbit.
A lot faster? Why?
The speed of an orbit depends on the distance from the center of the object and the mass of the object you orbit. At the distance the moon it it takes about a month (duh!). At 36000 km it takes about a day (geosynchronous) for a full orbit. At 200km from the surface (or 6578 from the center) the time is about 90 minutes. At 0km from the surface, or 6378 from the center, the time or speed for an orbit will not be significantly different.
There are however other significant objections to an orbit at surface level.
The reason for the 100km limit is that the atmosphere becomes significantly thin around that point.
You're allowed to download the MS-Caller ID spec (free-of-charge) to develop software. Then your users require an MS-licence to use the software, wether or not you want your code to be GPL or not.
they can, with reasonable accuracy, calculate how long a given scene will take to render, whereas with public distributed computing this calculation is not possible.
When the Dutch internet went from an "unreliable, variable bandwidth, unpredictable" connection between the academic network and the commercial branch to a "predictable, reliable, fixed-bandwidth" link between them, the whole internet slowed to a crawl for us.
The unreliable variable bandwidth link was 10 mbps ethernet, and was shared with a couple of supercomputers! The fixed bandwidth thingy was one or two mbps.
Anyway, as long as the unpredictable version is about 5 times faster (on average) than the predictable version, the unpredictable one may still beat the predictable by a useful amount.
Applications are quoted as "Identification", as in "who is this person?". Not as Identity verification, as in: "His pass says he's Roger Wolff, is that true?". Maybe that's a common misrepresentation by the press.
If you pose the "who is this" question to the computer, your scan will be matched against 1000 others. If the per-match chances of going wrong are 0.004 percent, then doing 1000 matches will result in about 4% error rate.
With that error rate, trying 1 million matches will result in a correct identification in about 4 in 10^18th attempts. Auch.
Now, for lots of applications it's not hard to move the problem into the "identity verification" realm. But what we on the outside don't see is wether or not the computer will just scan the database for the best-match internally, and then decide: "Hmm. best match: Spacefrog, pass says Roger Wolff, Nope not him. Access denied".
At Schiphol airport these Iris scans are used to grant automatic access through customs. From stories I understand that this is an important field where people think this will be applied. So I'm convinced it will pushed to be implemented into "large groups" soonish.
If my "schiphol pass" says: "this is Roger Wolff" and the system just checks my identity with the Iris scan on file, then a 1% failure rate can be acceptable.
But if it is used to IDENTIFY me as in: "Who is this person in front of the camera?" then the small-scale-failure rate will go up enormously when larger groups are put into the database.
... fails to correctly identify people in just 4 percent of cases...
If you do a test run with 1000 individuals,and find that 4% of the subjects are identified as someone else, then you really have a problem.
If you then scale up to 1 million people, you will find that a MUCH larger percentage of people will be misidentified: There is a much larger database of people who might have an iris that to the computer looks almost the same. That's when the shit hits the fan.
Yes! And in reality it used to stop at 2Gb with "file too large" and you'd have to do it multiple times with different files before you start rm-ing them. But you get the idea.
Here in the Netherlands, we have an institution called "Buma/Stemra". They have something to do with copyrights on music and stuff like that.
For some small monthly fee you can have up to 30 seconds of music on your site. (Mono, not encoded with more than 128kbps)
But if you read the smallprint, they only handle the copyrights on the notes and the lyrics. Not for the performing artist. They make you believe that you're allowed to put michael jackson's "bad" on your site for that fee, but you're not (unless you perform it yourself).
I store all my incoming Email. including the spam.
I could be sent to jail for that: I have kiddy porn on my drive. Just having the stuff on your drive makes you in violation of the Dutch law, and you can go to jail for that. Simple.
(The spam is useful for training filters, or to calculate statistics after the fact. If I know in advance I want to know how many spams I got every day, I can just tabulate them, but I might want to collect some other statistic and need info I didn't keep. So I just store everything. Costs me some disk space, but it's cheap nowadays.)
Even back in the days I did call support for an ISP, sometimes I'd just ask their login name and they'd just blurt out, "My login is sueray22 and my password is newyork!"
Recently I've been asked by "tech support" for some stupid websites for my username AND password. Does someone here know a site that explains the CONs about this?
One holds my employee's salaries and such. I'm perfectly happy that the support people can access that if they need to. The system can then log "helpdesk-Tom" accessed XYZ's financial data, and get possible problems after that figured out. If I give him my password, it'll look as if I used some stupid dialup with my password, and it's my word against their logs that it wasn't me....
The other case would have allowed the helpdesk guy to order goods in my name. Volume two of the catalog is over 2000 pages. Volume one is less thick (and currently not on my desk). To give you an idea about how many products he'd be able to chose from.....
There were no skid marks (Although ABS may limit them)
ABS leaves very distinct skid marks. ABS works by skidding for a short time, detecting that, and then reducing brake power a little for a short period, so that the wheels spin and skid again and things can start over. As at least one of the wheels skids periodically, it will leave marks.
If you're backing systems up, tape begins making economic sense when your backups start getting past 100Gb or so.
Ehmm. Have you found a tape system where the TAPEs currently cost less than the cost for new harddrives? Harddrives out here cost below one euro per Gb, These SOJ drives seem to cost double that. And whenever during the last year I looked into tapes, the tape-media were also about $2 per Gb. So even if I buy an infinite amout of storage the harddrives will win.
I want to buy 4 of these drives. Hook up two per interface. Then I have two drives per interface, each capable of sustaining up to 61Mbyte per second (read the specs), on an ATA133, and the on-disk-cache should allow me to alternate handling the two drives giving something like 120Mbyte per second sustained over two drives.
However all IDE controllers (even the ATA133 ones!) that I've been able to find seem to bottleneck at around 60 to 70Mbyte per second:-(.
Wrong. There were two mremap bugs. Regretfully, some people with the right background didn't have time to look at the bug and the fix before the first one went public. So a second public fix was needed.
Correct. I know the difference. I also know that some people insist on writing GB for gigabyte, while I write Gb for gigabyte. Now, what is your point again?
That worked because RLL encoded the data using a different method than MFM.
Right. MFM encodes something like 2 bits for every 4 magnetic areas on the drive, while RLL encodes 3 bits. Nowadays they code something like 16 bits for every 17 areas.
I remember the times when the geometry of the drive was compiled into the OS! When I got my RLL controller the OS (Minix) only could use the first 17 sectors of every track. But my RLL drive had 26! So eventually, I coded up a second disk driver with the new geometry. Copy the data from the old driver to the new (yes, with a live filesystem and all!), and then remove the old driver:-) Those were the days.
Working for a data-recovery company I've opened up quite a bunch of drives. So I know what's going on inside.
Depending on the form factor and the manufacturer, they can stuff 1, 2, 3, 4, 6, or even 15 platters in an enclosure.(That last is for a full height 5.25" drive. 6 only fit into the 1.7" heigh drives).
Suppose Quantum can fit 10Gb on one side of a platter. They will then make a family of drives: 10G (one platter, one head), 20Gb (one platter, two heads), 30Gb (two platters, three heads), 40Gb (two platters, 4 heads), and 60Gb. (Quantum only fits three platters in a 1" high 3.25" drive). This sequence holds for the quantum Fireball AS series by the way.
As you can see, there is half a platter (one side) unused in the 10 and 30Gb models. Quantum usually leaves that nice and shiny. IBM usually takes a sharp object and makes a big scratch on the surface....
In either case, it's quite possible that QA on that part of the disk failed, and that it would be unwise to use that part of the disk. Even if you managed to get a head able to read/write it....
from my understanding of the licence: If I want to implement a compliant implementation, I can go right ahead. (as long as I promise not to bother MS about patents that I might own on this technology).
If I then sell or distribute the software I wrote: Fine.
You however get to pay MicroSoft to use my software.
Oh, and they've included a GPL incompatible advertizing clause.
The core problem with this is that although most buffer overflow exploits currently simply use the buffer they are overflowing as the code stash, it has been shown that this is not neccesary.
Oh, the exploits might need to be tweaked a bit, but after a while there are enough examples floating around on the internet to allow the script kiddies to generate 'ploits again.
Slashdot Mirror
Also, go SCSI instead of IDE.
Ehmmm. As maxtor stopped making the 5400 "high capacity, low performance" line of disks, we've been forced to start buying 7200rpm drives. Turns out they break down more often. We've got enough disks in the office to start doing statistics now...
SCSI tends to be engineered "high performance", which translates to heat, which might cause premature breakdowns. I'm not so sure I like high performance SCSI that much.....
After RTFA, I concluded that these two companies are entering a research cooperation to try and develop something that might make it into the market within a decade or so.
What did I misread? Where does it say they are going into production???
As a matter of fact, if the atmosphere and terrain were not an issue, you COULD do an orbit a hundred feet off the ground. And you could enter this orbit by going straight sideways. It just requires moving a lot faster than a higher orbit.
A lot faster? Why?
The speed of an orbit depends on the distance from the center of the object and the mass of the object you orbit. At the distance the moon it it takes about a month (duh!). At 36000 km it takes about a day (geosynchronous) for a full orbit. At 200km from the surface (or 6578 from the center) the time is about 90 minutes. At 0km from the surface, or 6378 from the center, the time or speed for an orbit will not be significantly different.
There are however other significant objections to an orbit at surface level.
The reason for the 100km limit is that the atmosphere becomes significantly thin around that point.
It's worse.
You're allowed to download the MS-Caller ID spec (free-of-charge) to develop software. Then your users require an MS-licence to use the software, wether or not you want your code to be GPL or not.
At least that was the state when I last looked.
they can, with reasonable accuracy, calculate how long a given scene will take to render, whereas with public distributed computing this calculation is not possible.
When the Dutch internet went from an "unreliable, variable bandwidth, unpredictable" connection between the academic network and the commercial branch to a "predictable, reliable, fixed-bandwidth" link between them, the whole internet slowed to a crawl for us.
The unreliable variable bandwidth link was 10 mbps ethernet, and was shared with a couple of supercomputers! The fixed bandwidth thingy was one or two mbps.
Anyway, as long as the unpredictable version is about 5 times faster (on average) than the predictable version, the unpredictable one may still beat the predictable by a useful amount.
roger.
Applications are quoted as "Identification", as in "who is this person?". Not as Identity verification, as in: "His pass says he's Roger Wolff, is that true?". Maybe that's a common misrepresentation by the press.
If you pose the "who is this" question to the computer, your scan will be matched against 1000 others. If the per-match chances of going wrong are 0.004 percent, then doing 1000 matches will result in about 4% error rate.
With that error rate, trying 1 million matches will result in a correct identification in about 4 in 10^18th attempts. Auch.
Now, for lots of applications it's not hard to move the problem into the "identity verification" realm. But what we on the outside don't see is wether or not the computer will just scan the database for the best-match internally, and then decide: "Hmm. best match: Spacefrog, pass says Roger Wolff, Nope not him. Access denied".
At Schiphol airport these Iris scans are used to grant automatic access through customs. From stories I understand that this is an important field where people think this will be applied. So I'm convinced it will pushed to be implemented into "large groups" soonish.
If my "schiphol pass" says: "this is Roger Wolff" and the system just checks my identity with the Iris scan on file, then a 1% failure rate can be acceptable.
But if it is used to IDENTIFY me as in: "Who is this person in front of the camera?" then the small-scale-failure rate will go up enormously when larger groups are put into the database.
If you do a test run with 1000 individuals,and find that 4% of the subjects are identified as someone else, then you really have a problem.
If you then scale up to 1 million people, you will find that a MUCH larger percentage of people will be misidentified: There is a much larger database of people who might have an iris that to the computer looks almost the same. That's when the shit hits the fan.
Yes! And in reality it used to stop at 2Gb with "file too large" and you'd have to do it multiple times with different files before you start rm-ing them. But you get the idea.
Before I use Linux tools to "ghost" a partition, I always
dd if=/dev/zero of=somefile; rm somefile
first. That zeroes all the unused blocks. Those compress pretty good.
Here in the Netherlands, we have an institution called "Buma/Stemra". They have something to do with copyrights on music and stuff like that.
For some small monthly fee you can have up to 30 seconds of music on your site. (Mono, not encoded with more than 128kbps)
But if you read the smallprint, they only handle the copyrights on the notes and the lyrics. Not for the performing artist. They make you believe that you're allowed to put michael jackson's "bad" on your site for that fee, but you're not (unless you perform it yourself).
I wouldn't be surprised if this was similar....
Roger.
I store all my incoming Email. including the spam.
I could be sent to jail for that: I have kiddy porn on my drive. Just having the stuff on your drive makes you in violation of the Dutch law, and you can go to jail for that. Simple.
(The spam is useful for training filters, or to calculate statistics after the fact. If I know in advance I want to know how many spams I got every day, I can just tabulate them, but I might want to collect some other statistic and need info I didn't keep. So I just store everything. Costs me some disk space, but it's cheap nowadays.)
Even back in the days I did call support for an ISP, sometimes I'd just ask their login name and they'd just blurt out, "My login is sueray22 and my password is newyork!"
Recently I've been asked by "tech support" for some stupid websites for my username AND password. Does someone here know a site that explains the CONs about this?
One holds my employee's salaries and such. I'm perfectly happy that the support people can access that if they need to. The system can then log "helpdesk-Tom" accessed XYZ's financial data, and get possible problems after that figured out. If I give him my password, it'll look as if I used some stupid dialup with my password, and it's my word against their logs that it wasn't me....
The other case would have allowed the helpdesk guy to order goods in my name. Volume two of the catalog is over 2000 pages. Volume one is less thick (and currently not on my desk). To give you an idea about how many products he'd be able to chose from.....
There were no skid marks (Although ABS may limit them)
ABS leaves very distinct skid marks. ABS works by skidding for a short time, detecting that, and then reducing brake power a little for a short period, so that the wheels spin and skid again and things can start over. As at least one of the wheels skids periodically, it will leave marks.
How did it take them three years to figure that out? Wasn't the data right there in their hands?
The first argument heard in court would probably have been: "These devices are unreliable, it got messed up by the crash".
So it can very well take 3 years to come to the conclusion that these things ARE reliable, do NOT lie, and ARE permissable as evidence in court.
If you're backing systems up, tape begins making economic sense when your backups start getting past 100Gb or so.
Ehmm. Have you found a tape system where the TAPEs currently cost less than the cost for new harddrives? Harddrives out here cost below one euro per Gb, These SOJ drives seem to cost double that. And whenever during the last year I looked into tapes, the tape-media were also about $2 per Gb. So even if I buy an infinite amout of storage the harddrives will win.
Roger.
Yah, ATA-100 is just so much faster than ATA-133.
:-(.
Nah.
I want to buy 4 of these drives. Hook up two per interface. Then I have two drives per interface, each capable of sustaining up to 61Mbyte per second (read the specs), on an ATA133, and the on-disk-cache should allow me to alternate handling the two drives giving something like 120Mbyte per second sustained over two drives.
However all IDE controllers (even the ATA133 ones!) that I've been able to find seem to bottleneck at around 60 to 70Mbyte per second
It was the same thing and already fixed
Wrong. There were two mremap bugs. Regretfully, some people with the right background didn't have time to look at the bug and the fix before the first one went public. So a second public fix was needed.
No, the head would be missing.
Correct. I know the difference. I also know that some people insist on writing GB for gigabyte, while I write Gb for gigabyte. Now, what is your point again?
That worked because RLL encoded the data using a different method than MFM.
:-) Those were the days.
Right. MFM encodes something like 2 bits for every 4 magnetic areas on the drive, while RLL encodes 3 bits. Nowadays they code something like 16 bits for every 17 areas.
I remember the times when the geometry of the drive was compiled into the OS! When I got my RLL controller the OS (Minix) only could use the first 17 sectors of every track. But my RLL drive had 26! So eventually, I coded up a second disk driver with the new geometry. Copy the data from the old driver to the new (yes, with a live filesystem and all!), and then remove the old driver
Working for a data-recovery company I've opened up quite a bunch of drives. So I know what's going on inside.
Depending on the form factor and the manufacturer, they can stuff 1, 2, 3, 4, 6, or even 15 platters in an enclosure.(That last is for a full height 5.25" drive. 6 only fit into the 1.7" heigh drives).
Suppose Quantum can fit 10Gb on one side of a platter. They will then make a family of drives: 10G (one platter, one head), 20Gb (one platter, two heads), 30Gb (two platters, three heads), 40Gb (two platters, 4 heads), and 60Gb. (Quantum only fits three platters in a 1" high 3.25" drive). This sequence holds for the quantum Fireball AS series by the way.
As you can see, there is half a platter (one side) unused in the 10 and 30Gb models. Quantum usually leaves that nice and shiny. IBM usually takes a sharp object and makes a big scratch on the surface....
In either case, it's quite possible that QA on that part of the disk failed, and that it would be unwise to use that part of the disk. Even if you managed to get a head able to read/write it....
from my understanding of the licence: If I want to implement a compliant implementation, I can go right ahead. (as long as I promise not to bother MS about patents that I might own on this technology).
If I then sell or distribute the software I wrote: Fine.
You however get to pay MicroSoft to use my software.
Oh, and they've included a GPL incompatible advertizing clause.
Once installed, if you get a spam or virus, you should now have the Email address of the person responsible.
This makes it quite a lot easier to tell the responsible person to fix their computer than when you just have an IP.
Roger.
The core problem with this is that although most buffer overflow exploits currently simply use the buffer they are overflowing as the code stash, it has been shown that this is not neccesary.
Oh, the exploits might need to be tweaked a bit, but after a while there are enough examples floating around on the internet to allow the script kiddies to generate 'ploits again.