Michael Jordan is only better off mowning his own lawn if he would have to sacrifice basketball playing time to do the lawn mowning (and with the size of his lawn, that might be the case.)
However, if the decision is to mow your own lawn or sit on your ass and pay the neighbor's kid to do it, you're better off doing it yourself.
Applying this to the tech mythology of the 90s, and people who would, say, spend money eat out so they don't have to waste time cooking and cleaning, when they could be thinking of the next great internet business. (and of course, 99.9% of engineers didn't come up with their own internet businesses) The "my time is more valuable than the cost of XXXX" idea only applies if you have zero free time, and are earning additional YYYY for every hour worked.(insert 'wages', 'gratitude', 'happy meals', 'respect', or other generic economic unit of prosperity for YYYY above)
Actually, that isn't quite complete, and it varies by state. Most states that have this refer to it as a Homestead act. Every state that has a Homestead act has a dollar limit on the value of the home that may be protected from lawsuit forfeiture.
I know Colorado has no such act, so you can lose your house in a lawsuit, no matter how small. I believe Arizona has a limit of about $100K on the value of the home protected by their Homestead act, but that is from memory.
"And in all the places I've worked, no one has ever cared what snaps anyone took or what they did with them."
Therein lies the problem. I know at my job I am forbidden from bringing a camera of any kind into any building, it was spelled out for me along with a hundred other regulations.
I'm sure this guy signed the same papers, or at least was aware of the rules.
I think the origin of their "biomedical" studies were bio-electrical in nature, hence the tie to EE. 60 years ago the EE department was simply a branch of the physics department at MIT too... as each sub-specialty grows, eventually it warrants its own department. When my dad went to MIT, there was no such thing as computer science... stuff we take for granted in a CS education was a fringe part of EE back then.
Interest in the biomedical/bioelectrical classes was rather limited when I attended school there, though maybe that has changed. (Biology is now required for all students I believe, whereas I got to choose organic chemistry "5.11" or inorganic chemistry "3.091" instead) (In that example, note that inorganic chemistry is part of the materials science department (course 3), and not part of the chemistry department. (course 5)) Also, most of the math required by computer science majors have designations in both EECS (course 6) and mathematics (course 18), if for no other reason than to make the registration software more complicated.
"While SCSI drives feature superior mechanics, their server orientation forces them to trade away firmware optimized for highly-localized patterns in favor of strategies that maximize returns in random access scenarios. In the Raptor, WD faces much of the same quandary."
There is no cache optimization for random access scenario, since you're guaranteed to almost never get a read cache hit.
Maximizing random performance = mechanics. Maximizing local performance = scheduling.
> 40% on-disk errors before low-level (factory) format?
Coming from a firmware engineer from a disk drive company, that sentence makes no sense whatsoever.
The "stretched capacity formats" that drive companies are using to reach their 200GB or larger drives are almost purely a function of the heads used in the drive, and have almost nothing to do with the specific media. From the plots I have seen, if media had 1% surface defects I would be surprised...
WD doesn't make any disk drives with 4 platters in them, they only use 3 platters.
The 200GB drive is a "format enhanced" version of the disk drive. When a vendor says their technology is 60GB/platter, that is a relative measurement and simply refers to the generation of the technology. There are lots of things they can do to determine how much data they actually put on those platters.
In this case the WD 200 is 3 platter / 6 heads at 66.7GB/head. I'm sure they have formats of 66.7GB for their best builds, 60GB for standard ones, and some sort of fallback 50ish GB format on drives that have marginal heads...
A 2 platter drive with 50GB/platter would still be 100GB, not THAT shabby, and still cheaper to manufacture than their previous generation of 40GB/platter drives.
> If you define education as sitting in an > institution doing whatever your teacher tells > you and being graded on your obedience, then > MIT fits in well. If you define education as > accumulating knowledge, then you do not need to > go to an institution to do that, just pick up > books and start reading.
Clearly you don't know what you're talking about.
I would say that about one-third to one-half of my classes at MIT were of the first kind.
The rest were much more free-form, with open-ended 2-month long projects, creative works, etc, that allowed us to get into whatever we wanted to focus on, limited only by what we could learn in the given time. The amount of open-endedness only increases as you get into more advanced classes, and its really only the basic 6 (calc I, calc II, bio, physics I(statics+dynamics), physics II (electromagnetism+waves), and chemistry where the professors hold your hand and insist on a certain way of doing things. (Even those ways get argued by some of the students, who are given credit for being right when they are)
Re:SerialATA doesn't seem very advanced
on
Serial ATA Coming
·
· Score: 5, Informative
> There are a number of issues that it seems that > SerialATA doesn't address that it should:
> 1) Power to the device is still separated from the data connection.
This is because the motors being currently used in hard drives pull a few amps at spinup time, and the wire guage used for signalling cannot possibly carry this much current. There is a thought that once the market is serial ATA native, the HD manufacturers will then standardize on a 5V, low current motor instead of the current 12V beasts, however, things like Microsoft's "on now" spec and other crap specify a minimum spinup time that force us to slam the motor to get up to speed.
Single connector vs two connectors has little bearing on whether you can hot swap. It is a function of how you isolate/protect your power circuitry from not having all the conductors touching at the same time. (e.g., in current ATA, what happens when of the +12,0,+5,-5, only the +12 and the -5 are connected because the pins are slightly out of tolerance?)
> 2) Because it is backwards compatible with > regular ATA it appears it will have the same > limitations on the number of devices you can > connect, i.e. 2 per channel.
It is point to point. The notion of channels will disappear, and BIOSs in the future will simply allocate an 8-word I/O space address for the device, instead of todays "primary IDE" at 0x1F0 or whatever.
> 3) It is unusable for external devices
I don't believe that is correct, however, since it is point to point, a box of external drives (similar to a SCSI enclosure) would need a cable running to it for every drive in the enclosure.
There will be no daisy chaining or hub or star network of SATA devices.
Re:Still has 137GB Limitation
on
Serial ATA Coming
·
· Score: 4, Insightful
The 48-bit command set is part of the ATA-6 specification that you can read at www.t13.org. Serial ATA will support this command set.
Most vendors don't need to support 48-bits yet because they don't have drives that are big enough. Many manufacturers do not make 4-platter IDE drives anymore, and with the current technology of 40GB/platter, the 3-platter disks are only 120GB.
When the next generation comes in at 60 or 80 GB/platter, they'll support 48-bit commands as needed.
Re:Why is serial ATA faster?
on
Serial ATA Coming
·
· Score: 5, Informative
> Pardon my ignorance, but can somebody explain > why serial ATA is faster than the current > (parallel) ATA? > > On PC's, parallel ports are significantly > faster than serial ports because they transmit > 8 bits at a time instead of serial's one bit at > a time. Wouldn't the same thing hold true for > parallel vs serial ATA? > > Please explain.
In a perfect world, parallel would always be faster than serial. However, what happens is that due to outside factors (shape of the cable, EM interference at the time, etc) when you send those 8 bits down a parallel port, they don't all arrive at the destination at exactly the same time. The faster you send them, the more likely they are to not arrive when you send them since your tolerances get lower. This is referred to as signal skew.
Serial ATA borrows a technique from LVD SCSI devices which is low-voltage differential signaling. They send the pulses down 2 lines polarity reversed. By using 2 wires instead of 9 (8 data bits + a clock) or in UDMA land 20 wires (16 data lines, IOR, IOW, DMARQ, DREQ) the chances of them being significantly different than one another is less, because they're closer to following a consistent path through space. This allows them to toggle the lines MUCH faster in LVD applications than parallel applications, which gives us much higher data rates.
Another thing is that the IDE bus still uses TTL signal levels (5V/0V), meaning that it takes a HUGE amount of power to wiggle all those 40 conductors up and down to get some data across. The little chipsets on motherboards these days have trouble supplying enough internal power to do that, so LVD will help them make less complicated circuits in the chipset to talk serial ATA.
> Cabling issues will be a big benefit to OEMs, > and end users who have fewer tech issues with > adding drives, as it will be impossible to wire > improperly and master/slave/cs jumpers will no > longer exist.
Actually, that was solved with the cable-select jumper years ago. Almost all major OEMs buy their drives configured as such these days.
If you think it will take more than a week to have the hardware hacked to read ripped CDs and CD/Rs you're crazy.
Linux is under the GPL, and we, the consumer, are free to extend it under the same GPL. Once this is out, the 2.4 series of kernels should be on this architecture in a month and other drivers not too long after that, including CD/R capability.
This thing is going to get hacked like no tomorrow.
Wonder how a PS/2 emulator would run in linux on a PS/2...
The issue is that the chinese government doesn't care one whit whether Alice or Bob was trying to send the original message. They can just as easily proxy Alice AND Bob out of the picture, then there's 2 less people for them to worry about in the future.
Never underestimate the bandwidth of a camel loaded up with 80 gig hard drives.
--eric
Re:Works great if you have a clean room available
on
Clear Hard Drive Mods
·
· Score: 5, Informative
It is 747 in size, not in mass.
A GMR head is visible to the naked eye (tiny tiny black speck on the end of the actuator) The actual read element is not visible to the eye.
It is basically shaped like a huge surfboard, where there is a tiny element on the back end of the surfboard that does the actual reading / writing.
There are 20-60 data tracks at present in modern drives within the thickness of a piece of paper. That is the required lateral accuracy.
Vertical accuracy is assisted by the use of an "air bearing". This is common to all drives to my knowledge. Basically, our head flies along similar to the way a low altitude helicopter flies upon a cushion of its own prop wash. The heads are designed like wings, and they channel a tiny bit of air underneath the head. If the head drops lower due to a bump in the media, the air pressure increases, forcing the head to resist the change to a lower altitude. Similarly, no cushion of air is created at high altitude, so this causes the head to settle down on the platter.
When you compare the size of the head to its distance from the ground, a 747 at 1/4" was accurate as of 18 months ago. Now it is even lower.
As to the rocky mountains, ok, sure, the bumps might not be quite that big, but they're at least the Appalachians. On the fractional micron scale that everything works at, there's no practical way to flatten the media that perfectly. Besides, if the media were perfectly smooth it would create too much surface tension and the head would stick to it (called "stiction"), so actually parts of the media are intentionally textured to reduce the amount of surface that might actually touch the head at any given time.
As to your F16 comment, just how maneuverable do you think the head of a drive is? We can't steer our heads onto a track, we can only recognize that we missed the track and need to adjust.
Perhaps I should have said we were an oil tanker flying mach 4 at 1/2" of altitude down an olympic slalom course.
Hell I am amazed the things even work, knowing what I know about them.
Re:Works great if you have a clean room available
on
Clear Hard Drive Mods
·
· Score: 5, Informative
.01" ? ROFL
I am a firmware engineer for Maxtor...
The heads on our drives, and everyone else's in IDE land are currently flying at some fraction of a micron, if they aren't burnished already (sliding through the layer of lubrication on the surface of the platter).
Put to scale, the head of a disk drive is like a 747 jumbo jet flying at mach 4 at an altitude of 1/4" over the rocky mountains.
A single particle of dust inside the drive is HUGE, and can easily cause catastrophic data failure. If the head touches the media at all, you can basically forget the adjacent 10-20 tracks on each side, which on a modern drive is roughly 15 megabytes at least. If the strike happened while the drive was seeking, you get a radial scratch which can be destructive to a much larger area of the drive.
Bottom line: don't do it, no matter how cool you think you might be. They're fragile enough as it is.
Yes, you're right, and the drive is actually slower issuing queued commands than standard read DMA commands.
Standard ATA queue depth is 32, which I believe is the same as SCSI, though there are side projects in ATA land to increase this because of potential performance gains. (more things you can choose from, better odds of being able to choose something easy)
Slashdot Mirror
There is one caveat to this point...
Michael Jordan is only better off mowning his own lawn if he would have to sacrifice basketball playing time to do the lawn mowning (and with the size of his lawn, that might be the case.)
However, if the decision is to mow your own lawn or sit on your ass and pay the neighbor's kid to do it, you're better off doing it yourself.
Applying this to the tech mythology of the 90s, and people who would, say, spend money eat out so they don't have to waste time cooking and cleaning, when they could be thinking of the next great internet business. (and of course, 99.9% of engineers didn't come up with their own internet businesses) The "my time is more valuable than the cost of XXXX" idea only applies if you have zero free time, and are earning additional YYYY for every hour worked.(insert 'wages', 'gratitude', 'happy meals', 'respect', or other generic economic unit of prosperity for YYYY above)
--eric
Actually, that isn't quite complete, and it varies by state. Most states that have this refer to it as a Homestead act. Every state that has a Homestead act has a dollar limit on the value of the home that may be protected from lawsuit forfeiture.
I know Colorado has no such act, so you can lose your house in a lawsuit, no matter how small. I believe Arizona has a limit of about $100K on the value of the home protected by their Homestead act, but that is from memory.
Quicksilver was good... but it took me about 400 pages to get "into" the story, and the ending 100 pages dragged for all but the final two sentences.
The book was good, but I wouldn't put it on my "best of" list. It ended with a soft fizzle the same as Snow Crash.
"And in all the places I've worked, no one has ever cared what snaps anyone took or what they did with them."
Therein lies the problem. I know at my job I am forbidden from bringing a camera of any kind into any building, it was spelled out for me along with a hundred other regulations.
I'm sure this guy signed the same papers, or at least was aware of the rules.
A company just started that has legal ROM downloads for a reasonable price if you want to legally play some of the old classics.
http://www.starroms.com/
Just to add a bit...
I think the origin of their "biomedical" studies were bio-electrical in nature, hence the tie to EE. 60 years ago the EE department was simply a branch of the physics department at MIT too... as each sub-specialty grows, eventually it warrants its own department. When my dad went to MIT, there was no such thing as computer science... stuff we take for granted in a CS education was a fringe part of EE back then.
Interest in the biomedical/bioelectrical classes was rather limited when I attended school there, though maybe that has changed. (Biology is now required for all students I believe, whereas I got to choose organic chemistry "5.11" or inorganic chemistry "3.091" instead) (In that example, note that inorganic chemistry is part of the materials science department (course 3), and not part of the chemistry department. (course 5)) Also, most of the math required by computer science majors have designations in both EECS (course 6) and mathematics (course 18), if for no other reason than to make the registration software more complicated.
--eric
Sorry you had such luck... I got my Money/Turbo Tax refunds (both) within about 5 weeks after I mailed in my stuff.
Sorry to hear you got screwed.
--eric
"While SCSI drives feature superior mechanics, their server orientation forces them to trade away firmware optimized for highly-localized patterns in favor of strategies that maximize returns in random access scenarios. In the Raptor, WD faces much of the same quandary."
There is no cache optimization for random access scenario, since you're guaranteed to almost never get a read cache hit.
Maximizing random performance = mechanics.
Maximizing local performance = scheduling.
I think the ANSI standard term for 2^30 bytes is Gibibyte, or GiB.
Gigabyte, or 1,000,000,000 bytes, is GB.
I could be wrong, but that was my understanding. Of course, very few people actually write GiB in practice...
Hasn't this been thought of before? Small-scale parallel processing sounds a lot like the queues in SEDA:
SEDA Homepage
they aren't currently making 4 platter drives.
I have no idea what they were doing 3 years ago, I've only been in the industry now for about 20 months.
> 40% on-disk errors before low-level (factory) format?
Coming from a firmware engineer from a disk drive company, that sentence makes no sense whatsoever.
The "stretched capacity formats" that drive companies are using to reach their 200GB or larger drives are almost purely a function of the heads used in the drive, and have almost nothing to do with the specific media. From the plots I have seen, if media had 1% surface defects I would be surprised...
WD doesn't make any disk drives with 4 platters in them, they only use 3 platters.
The 200GB drive is a "format enhanced" version of the disk drive. When a vendor says their technology is 60GB/platter, that is a relative measurement and simply refers to the generation of the technology. There are lots of things they can do to determine how much data they actually put on those platters.
In this case the WD 200 is 3 platter / 6 heads at 66.7GB/head. I'm sure they have formats of 66.7GB for their best builds, 60GB for standard ones, and some sort of fallback 50ish GB format on drives that have marginal heads...
A 2 platter drive with 50GB/platter would still be 100GB, not THAT shabby, and still cheaper to manufacture than their previous generation of 40GB/platter drives.
> If you define education as sitting in an
> institution doing whatever your teacher tells
> you and being graded on your obedience, then
> MIT fits in well. If you define education as
> accumulating knowledge, then you do not need to
> go to an institution to do that, just pick up
> books and start reading.
Clearly you don't know what you're talking about.
I would say that about one-third to one-half of my classes at MIT were of the first kind.
The rest were much more free-form, with open-ended 2-month long projects, creative works, etc, that allowed us to get into whatever we wanted to focus on, limited only by what we could learn in the given time. The amount of open-endedness only increases as you get into more advanced classes, and its really only the basic 6 (calc I, calc II, bio, physics I(statics+dynamics), physics II (electromagnetism+waves), and chemistry where the professors hold your hand and insist on a certain way of doing things. (Even those ways get argued by some of the students, who are given credit for being right when they are)
> There are a number of issues that it seems that > SerialATA doesn't address that it should:
> 1) Power to the device is still separated from the data connection.
This is because the motors being currently used in hard drives pull a few amps at spinup time, and the wire guage used for signalling cannot possibly carry this much current. There is a thought that once the market is serial ATA native, the HD manufacturers will then standardize on a 5V, low current motor instead of the current 12V beasts, however, things like Microsoft's "on now" spec and other crap specify a minimum spinup time that force us to slam the motor to get up to speed.
Single connector vs two connectors has little bearing on whether you can hot swap. It is a function of how you isolate/protect your power circuitry from not having all the conductors touching at the same time. (e.g., in current ATA, what happens when of the +12,0,+5,-5, only the +12 and the -5 are connected because the pins are slightly out of tolerance?)
> 2) Because it is backwards compatible with
> regular ATA it appears it will have the same
> limitations on the number of devices you can
> connect, i.e. 2 per channel.
It is point to point. The notion of channels will disappear, and BIOSs in the future will simply allocate an 8-word I/O space address for the device, instead of todays "primary IDE" at 0x1F0 or whatever.
> 3) It is unusable for external devices
I don't believe that is correct, however, since it is point to point, a box of external drives (similar to a SCSI enclosure) would need a cable running to it for every drive in the enclosure.
There will be no daisy chaining or hub or star network of SATA devices.
Have your read the official 1394 spec?
It's like 1400 pages. Holy transaction overhead.
The 48-bit command set is part of the ATA-6 specification that you can read at www.t13.org. Serial ATA will support this command set.
Most vendors don't need to support 48-bits yet because they don't have drives that are big enough. Many manufacturers do not make 4-platter IDE drives anymore, and with the current technology of 40GB/platter, the 3-platter disks are only 120GB.
When the next generation comes in at 60 or 80 GB/platter, they'll support 48-bit commands as needed.
> Pardon my ignorance, but can somebody explain
> why serial ATA is faster than the current
> (parallel) ATA?
>
> On PC's, parallel ports are significantly
> faster than serial ports because they transmit
> 8 bits at a time instead of serial's one bit at
> a time. Wouldn't the same thing hold true for
> parallel vs serial ATA?
>
> Please explain.
In a perfect world, parallel would always be faster than serial. However, what happens is that due to outside factors (shape of the cable, EM interference at the time, etc) when you send those 8 bits down a parallel port, they don't all arrive at the destination at exactly the same time. The faster you send them, the more likely they are to not arrive when you send them since your tolerances get lower. This is referred to as signal skew.
Serial ATA borrows a technique from LVD SCSI devices which is low-voltage differential signaling. They send the pulses down 2 lines polarity reversed. By using 2 wires instead of 9 (8 data bits + a clock) or in UDMA land 20 wires (16 data lines, IOR, IOW, DMARQ, DREQ) the chances of them being significantly different than one another is less, because they're closer to following a consistent path through space. This allows them to toggle the lines MUCH faster in LVD applications than parallel applications, which gives us much higher data rates.
Another thing is that the IDE bus still uses TTL signal levels (5V/0V), meaning that it takes a HUGE amount of power to wiggle all those 40 conductors up and down to get some data across. The little chipsets on motherboards these days have trouble supplying enough internal power to do that, so LVD will help them make less complicated circuits in the chipset to talk serial ATA.
eric
> Cabling issues will be a big benefit to OEMs,
> and end users who have fewer tech issues with
> adding drives, as it will be impossible to wire
> improperly and master/slave/cs jumpers will no
> longer exist.
Actually, that was solved with the cable-select jumper years ago. Almost all major OEMs buy their drives configured as such these days.
If you think it will take more than a week to have the hardware hacked to read ripped CDs and CD/Rs you're crazy.
Linux is under the GPL, and we, the consumer, are free to extend it under the same GPL. Once this is out, the 2.4 series of kernels should be on this architecture in a month and other drivers not too long after that, including CD/R capability.
This thing is going to get hacked like no tomorrow.
Wonder how a PS/2 emulator would run in linux on a PS/2...
The issue is that the chinese government doesn't care one whit whether Alice or Bob was trying to send the original message. They can just as easily proxy Alice AND Bob out of the picture, then there's 2 less people for them to worry about in the future.
Never underestimate the bandwidth of a camel loaded up with 80 gig hard drives.
--eric
It is 747 in size, not in mass.
A GMR head is visible to the naked eye (tiny tiny black speck on the end of the actuator) The actual read element is not visible to the eye.
It is basically shaped like a huge surfboard, where there is a tiny element on the back end of the surfboard that does the actual reading / writing.
There are 20-60 data tracks at present in modern drives within the thickness of a piece of paper. That is the required lateral accuracy.
Vertical accuracy is assisted by the use of an "air bearing". This is common to all drives to my knowledge. Basically, our head flies along similar to the way a low altitude helicopter flies upon a cushion of its own prop wash. The heads are designed like wings, and they channel a tiny bit of air underneath the head. If the head drops lower due to a bump in the media, the air pressure increases, forcing the head to resist the change to a lower altitude. Similarly, no cushion of air is created at high altitude, so this causes the head to settle down on the platter.
When you compare the size of the head to its distance from the ground, a 747 at 1/4" was accurate as of 18 months ago. Now it is even lower.
As to the rocky mountains, ok, sure, the bumps might not be quite that big, but they're at least the Appalachians. On the fractional micron scale that everything works at, there's no practical way to flatten the media that perfectly. Besides, if the media were perfectly smooth it would create too much surface tension and the head would stick to it (called "stiction"), so actually parts of the media are intentionally textured to reduce the amount of surface that might actually touch the head at any given time.
As to your F16 comment, just how maneuverable do you think the head of a drive is? We can't steer our heads onto a track, we can only recognize that we missed the track and need to adjust.
Perhaps I should have said we were an oil tanker flying mach 4 at 1/2" of altitude down an olympic slalom course.
Hell I am amazed the things even work, knowing what I know about them.
.01" ? ROFL
I am a firmware engineer for Maxtor...
The heads on our drives, and everyone else's in IDE land are currently flying at some fraction of a micron, if they aren't burnished already (sliding through the layer of lubrication on the surface of the platter).
Put to scale, the head of a disk drive is like a 747 jumbo jet flying at mach 4 at an altitude of 1/4" over the rocky mountains.
A single particle of dust inside the drive is HUGE, and can easily cause catastrophic data failure. If the head touches the media at all, you can basically forget the adjacent 10-20 tracks on each side, which on a modern drive is roughly 15 megabytes at least. If the strike happened while the drive was seeking, you get a radial scratch which can be destructive to a much larger area of the drive.
Bottom line: don't do it, no matter how cool you think you might be. They're fragile enough as it is.
Yes, you're right, and the drive is actually slower issuing queued commands than standard read DMA commands.
Standard ATA queue depth is 32, which I believe is the same as SCSI, though there are side projects in ATA land to increase this because of potential performance gains. (more things you can choose from, better odds of being able to choose something easy)