Just because they don't make the CPU doesn't mean they can't continue to make SPARC systems. They'll just be Fujitsu SPARCs. And maybe having one company doing SPARC will bring Fujitsu back to profit in that division.
Apple, Dell and Compaq are doing alright selling systems with someone else's CPUs in them.
Frankly, for general-purpose computing, I've never been impressed with SPARCs, and the "64 virtual core" Niagara thing is just useless; so I'm going to do a happy-dance if I can be shut of doing Solaris SPARC support. But it's not going to be in the next few years.
It's a wild-goose-chase thing for the kid mechanic who's new to the shop. (There's a similar military tradition, too.) When you're done, you can get a round tuit.
I've got a CB radio on my motorcycle, for group rides, but before I got an audio set-up for my helmet speakers, I'd have it on channel 19 when I was going to or from a group event.
Funniest thing I ever heard on the CB was one trucker saying to another, "Westbound flatbed... you're wasting blinker fluid." He'd left his blinker on.
And yes, that means there wasn't much funny on the CB. Mostly about bears and discos, even bears at a disco, and strange things like that. (Though "do not get on I290 it's blocked at the Thruway and backed up all the way to 990" was really nice to know.)
I'll agree with you there. It's not the choice of steak or s**t for dinner, it's whose s**t you want for dinner. In my case, though, Roger's service doesn't work well in my neighbourhood; it's almost like there's one missing tower in the GTA and I live where it should go.
It doesn't help that the "competing" cell companies in my area are just re-sellers on Bell's towers. Maybe it's Bell's equipment I don't like? (At least, ever since Roger's borged Fido.)
I was so hoping Teacher's Pension would buy them and break them up for spares.
I'm having trouble thinking of an organization I trust less than Canada's telecoms companies to handle my money.
Has Bell figured out how to deal with incorrect direct payment transactions? When it happened to me, I had to have my bank block all transactions originated from Bell. Bell couldn't figure out how to identify the account making the bad transactions on their own--they actually needed the "payment refused" bounces from the bank. (They've got check-digits on account numbers now, but can they fix a problem from their end yet?)
A friend on Roger's discovered his phone had been cloned. The Roger's people thought that there was nothing odd about his phone being used in Toronto and in south Florida at the very same time. (The small claims judge did think that was odd.)
Depending on the bus design, active-low signals can be asserted by any device by turning on a transistor to ground. They are allowed to "float" high via pull-up resistors, so you get a poor-man's OR gate.
And depending on the circuitry you're using, "drive down, float up" may be much, much, much simpler than "drive up, float down". In the pre-CMOS days, for example, only N-channel FETs were available. So a transistor to drive a bus line low is cheap, but it's not pleasant driving something high. (Fortunately, I've forgotten most of the details.)
It's one of those places where reality and theory diverge.
What's even weirder is, using the $750 per song figure, my CD collection is worth $6,810,750.
But using $18 per CD, it's closer to $12,420. And that's not taking into account things like 24-discs-for-$120.
The lower figure makes much more sense, given the amount of money that's been in my life. And I've had to spend on books and DVDs and games and computers and the equipment to run it all and the house and the car and the bike and food and clothing and electricity and gas and gasoline and water and sewage and garbage and taxes and roofs and paint and furniture and dishes and appliances....
I'm not entirely sure, but I've got this suspicion that Wii Sports Boxing only works if you pull your punches. That is, you need a sharp stop to your punch.
If you would normally rely on a heavy bag or opponent's face to stop your punch, it's not going to work in the game.
And you're not supposed to punch your friend in the face playing Wii Sports Tennis but I did anyway. So much for the big warning screen.
It's worse than that. ars technica looked in to some of the numbers, and found that some of them derive from numbers that are cited because they are cited. There's some basis in a survey about losses due to "inadequate protection", and some estimates from that. But newer stuff which uses those reports drops all the contexts, jumps to the high end of estimated ranges, and so on.
So, every single number Mr. Hatch says needs to be suffixed with "[citation needed]".
Engineering still uses a bunch of Fortran. It may be called from C or C++, it may be standalone, but it's all still there.
It even optimizes better than C and C++: there are no pointers to muck up the alias sets. The language is very simple and well-defined, and that makes the compiler very, very happy. And reliable, you don't get into this "oooh, you assigned to a float from an int pointer and that breaks the ANSI Alias Rules so the compiler trashed your code for you".
It may be a bit clunky, but it is well-defined. (Compare the rampant "implementation defined" behaviours in ANSI C++....)
Sun's Performance Library is Fortran. IBM's ESSL is in Fortran. (Both are derived from the famous BLAS and LAPACK libraries.)
If you think that what many of us here do is "engineering" in any way, you've been in the same Flavor-Aid that the HR people drink.
Engineering is, to a large extent, about what goes wrong. Failure modes and effects; designing systems to fail predictably; identifying critical subsystems; all that sort of thing. The "is it strong enough" bit is trivial by comparison.
Most people with a "Software Engineer" title are programmers. They're not doing the design or analysis on the system. They aren't held accountable.
They call me a "Software Engineer" where I work. I've never done any outside of school, and believe it is actually illegal to call me that in this province.
Of course, some of that is _why_ [many] software systems suck. Actually engineering a system is slow and expensive, and software is often used as a cheap substitute for hardware. So what's the point if software becomes expensive?
On the other hand, if it's a thermal problem, you may have to rescue in "chunks". I had a disk go that could only be used for about 10 minutes before it got too warm and shut down.
On the third hand, you may have something that looks like physical damage, but when you wipe the disk with zeros to confirm the fault and get ready for RMA-time, it all magically comes back. That's a sign you got corrupted data on the disk that the ECC couldn't deal with. (And probably that you've got a drive with questionable firmware, and is reporting the wrong kind of error: Fujitsu, I'm looking at you. Especially for not recording anything in the SMART counters.)
That's because some things IE (6 and 7 at last test, don't have access to 8) just don't work the way they should.
Especially <BUTTON>; IE sends the contents of the button element, not the value attribute. And Microsoft's (original) page on the subject described the defective behaviour, and had a link to the w3c standard as if that document supported Microsoft's claims.
(I can't find that one now. What I can find is one that says IE 8 will work right in "Document compatibility mode". And you can't, necessarily, turn on document compatibility mode with a DOCTYPE of HTML 4.0 Strict. You may need an IE-specific HTTP header (or http-equiv tag). Or it may not work at all. I'm leaving all my stuff ugly for MSIE, I'm not going to go around adding headers all over the place. Because there's no way of telling if the browser really did go into HTML 4.0 Strict mode. Without doing IE-specific kludges, and I've already got those to make seriously-ugly INPUT elements instead of BUTTON elements. I'm not adding more, especially when Firefox and Opera are both fine and free.)
GP doesn't say "cite sources", it says "dig up sources". They succeeded in that part.
Citations are seriously advanced. Like, oh, grade 6 or 7. Around the same time they make you do that more-than-three-minutes-less-than-five public speaking thing in front of the whole class.
On my Mac, I'd far rather use VueScan than the crap that comes with scanners.
I had an old Microtek X6USB, the first cheap scanner with USB. It worked only with the provided drivers on Mac OS 8.1, nothing earlier, nothing later. But it worked fine with Xsane on Linux, or VueScan on Mac (and presumably Windows) as well.
I've got a much nicer scanner from Canon now, with the infra-red dust removal when scanning slides and negatives and everything. VueScan runs it nicely indeed.
My rule is, find out what hardware works with what software first. Then buy the hardware.
Don't connect/dev/random to/dev/urandom! It would be bad....
There's also a deadlock or some other lost race condition in some of the earlier 2.6 kernels that causes the entropy pool for/dev/random to dry up, and no new entropy would get added until a reboot. So, anything needing good randomness hangs waiting for/dev/random.
So, I've got a cron-job on those machines that watches the entropy pool and whines when they run out of random numbers. (I back-ported the fix to the must-run kernel on those servers, but the cron-job remains. Now it only triggers when something kills/tmp and it gets a false positive.)
A fault in the mirror logic? (Like inappropriately comparing timestamps as text, and getting one-more-digit. Happened to a number of programs when time_t got one digit longer.)
Swine flu for Servers outbreak?
They didn't have backups. They had a high-availability configuration and no disaster recovery plan.
RAID can copy an error across all mirrors faster than you can hit ^C. Live copies are not backups.
Well, in this particular case, it's fairly apparent that the uncovered corner conditions are obvious during static analysis. And the incorrectly-weighted averaging.
Code that passes static analysis is not known good.
Code that fails static analysis is known bad.
Static analysis does not mean "it's pretty". It means the logic and arithmetic work.
I dunno... that far out of town, what else are you going to do but play games?
And aren't the Americans always going on about how it's OK that their DSL and 3G coverage is so horrible _because_ there's such a large proportion of the country that is sparsely populated?
I've done the pull-a-drive-out test on Mac OS X software RAID the old fashioned way... by accidentally stepping on the power strip controlling one of the external volumes the RAID is using.
IME, the only reason to favour hardware RAID controllers is the battery-backed write-back cache. At least for mirroring modes; I don't really like parity modes, as they need too much simultaneous I/O on the disks, and disks are cheap enough today it's not worth it. (And the power consumption is WAY down from those full-height 5.25" SCSI monsters of 1993.)
But that works in JBOD (real JBOD, not "spanning with a stupid name) mode. So you can still use software RAID (Linux MD, AIX LVM mirrors, whatever) and get the benefits of the write-back cache and none of the "damn I need another card just like this one" downsides.
I was surprised that 2x the I/O on the PCI-X bus wasn't a problem. I really expected to have a measurable, consistent improvement from the hardware RAID controller, especially since it should have been able to drive two independent Ultra320 SCSI channels from a single PCI-X transaction. Especially given what the card cost.
Glad I wasn't the one paying for it....
I've done less extensive tests on other configs since then, and I can only find specific I/O patterns where the hardware card is a slight win; it's usually about the same as software RAID. You can do more by locating your hot data at the rim of the disk.
Probably not a big deal these days, but in the pin-through-hole era, you ran the risk of getting scraps of aluminum stuck to the pins on the circuit boards. The plastic anti-stat bags are resistant to that.
Slashdot Mirror
Feed arctan of 100% in degrees to Google if you want to check his math.
Of course, you can always question my vague memories of trig....
Just because they don't make the CPU doesn't mean they can't continue to make SPARC systems. They'll just be Fujitsu SPARCs. And maybe having one company doing SPARC will bring Fujitsu back to profit in that division.
Apple, Dell and Compaq are doing alright selling systems with someone else's CPUs in them.
Frankly, for general-purpose computing, I've never been impressed with SPARCs, and the "64 virtual core" Niagara thing is just useless; so I'm going to do a happy-dance if I can be shut of doing Solaris SPARC support. But it's not going to be in the next few years.
It's a wild-goose-chase thing for the kid mechanic who's new to the shop. (There's a similar military tradition, too.) When you're done, you can get a round tuit.
I've got a CB radio on my motorcycle, for group rides, but before I got an audio set-up for my helmet speakers, I'd have it on channel 19 when I was going to or from a group event.
Funniest thing I ever heard on the CB was one trucker saying to another, "Westbound flatbed... you're wasting blinker fluid." He'd left his blinker on.
And yes, that means there wasn't much funny on the CB. Mostly about bears and discos, even bears at a disco, and strange things like that. (Though "do not get on I290 it's blocked at the Thruway and backed up all the way to 990" was really nice to know.)
I'll agree with you there. It's not the choice of steak or s**t for dinner, it's whose s**t you want for dinner. In my case, though, Roger's service doesn't work well in my neighbourhood; it's almost like there's one missing tower in the GTA and I live where it should go.
It doesn't help that the "competing" cell companies in my area are just re-sellers on Bell's towers. Maybe it's Bell's equipment I don't like? (At least, ever since Roger's borged Fido.)
I was so hoping Teacher's Pension would buy them and break them up for spares.
I'm having trouble thinking of an organization I trust less than Canada's telecoms companies to handle my money.
Has Bell figured out how to deal with incorrect direct payment transactions? When it happened to me, I had to have my bank block all transactions originated from Bell. Bell couldn't figure out how to identify the account making the bad transactions on their own--they actually needed the "payment refused" bounces from the bank. (They've got check-digits on account numbers now, but can they fix a problem from their end yet?)
A friend on Roger's discovered his phone had been cloned. The Roger's people thought that there was nothing odd about his phone being used in Toronto and in south Florida at the very same time. (The small claims judge did think that was odd.)
Depending on the bus design, active-low signals can be asserted by any device by turning on a transistor to ground. They are allowed to "float" high via pull-up resistors, so you get a poor-man's OR gate.
And depending on the circuitry you're using, "drive down, float up" may be much, much, much simpler than "drive up, float down". In the pre-CMOS days, for example, only N-channel FETs were available. So a transistor to drive a bus line low is cheap, but it's not pleasant driving something high. (Fortunately, I've forgotten most of the details.)
It's one of those places where reality and theory diverge.
What's even weirder is, using the $750 per song figure, my CD collection is worth $6,810,750.
But using $18 per CD, it's closer to $12,420. And that's not taking into account things like 24-discs-for-$120.
The lower figure makes much more sense, given the amount of money that's been in my life. And I've had to spend on books and DVDs and games and computers and the equipment to run it all and the house and the car and the bike and food and clothing and electricity and gas and gasoline and water and sewage and garbage and taxes and roofs and paint and furniture and dishes and appliances....
I'm not entirely sure, but I've got this suspicion that Wii Sports Boxing only works if you pull your punches. That is, you need a sharp stop to your punch.
If you would normally rely on a heavy bag or opponent's face to stop your punch, it's not going to work in the game.
And you're not supposed to punch your friend in the face playing Wii Sports Tennis but I did anyway. So much for the big warning screen.
I'm Canadian and it's at least a 7 hour drive to get north of the 49th, you insensitive clod!
Though I did set up a proxy so an American friend could watch Discovery Canada shows on the intertubes....
It's worse than that. ars technica looked in to some of the numbers, and found that some of them derive from numbers that are cited because they are cited. There's some basis in a survey about losses due to "inadequate protection", and some estimates from that. But newer stuff which uses those reports drops all the contexts, jumps to the high end of estimated ranges, and so on.
So, every single number Mr. Hatch says needs to be suffixed with "[citation needed]".
Engineering still uses a bunch of Fortran. It may be called from C or C++, it may be standalone, but it's all still there.
It even optimizes better than C and C++: there are no pointers to muck up the alias sets. The language is very simple and well-defined, and that makes the compiler very, very happy. And reliable, you don't get into this "oooh, you assigned to a float from an int pointer and that breaks the ANSI Alias Rules so the compiler trashed your code for you".
It may be a bit clunky, but it is well-defined. (Compare the rampant "implementation defined" behaviours in ANSI C++....)
Sun's Performance Library is Fortran. IBM's ESSL is in Fortran. (Both are derived from the famous BLAS and LAPACK libraries.)
If you think that what many of us here do is "engineering" in any way, you've been in the same Flavor-Aid that the HR people drink.
Engineering is, to a large extent, about what goes wrong. Failure modes and effects; designing systems to fail predictably; identifying critical subsystems; all that sort of thing. The "is it strong enough" bit is trivial by comparison.
Most people with a "Software Engineer" title are programmers. They're not doing the design or analysis on the system. They aren't held accountable.
They call me a "Software Engineer" where I work. I've never done any outside of school, and believe it is actually illegal to call me that in this province.
Of course, some of that is _why_ [many] software systems suck. Actually engineering a system is slow and expensive, and software is often used as a cheap substitute for hardware. So what's the point if software becomes expensive?
Turn off the new crap. Then it works.
I forget how I did that, though. Classic Index in Preferences?
On the other hand, if it's a thermal problem, you may have to rescue in "chunks". I had a disk go that could only be used for about 10 minutes before it got too warm and shut down.
On the third hand, you may have something that looks like physical damage, but when you wipe the disk with zeros to confirm the fault and get ready for RMA-time, it all magically comes back. That's a sign you got corrupted data on the disk that the ECC couldn't deal with. (And probably that you've got a drive with questionable firmware, and is reporting the wrong kind of error: Fujitsu, I'm looking at you. Especially for not recording anything in the SMART counters.)
Oh no it won't.
That's because some things IE (6 and 7 at last test, don't have access to 8) just don't work the way they should.
Especially <BUTTON>; IE sends the contents of the button element, not the value attribute. And Microsoft's (original) page on the subject described the defective behaviour, and had a link to the w3c standard as if that document supported Microsoft's claims.
(I can't find that one now. What I can find is one that says IE 8 will work right in "Document compatibility mode". And you can't, necessarily, turn on document compatibility mode with a DOCTYPE of HTML 4.0 Strict. You may need an IE-specific HTTP header (or http-equiv tag). Or it may not work at all. I'm leaving all my stuff ugly for MSIE, I'm not going to go around adding headers all over the place. Because there's no way of telling if the browser really did go into HTML 4.0 Strict mode. Without doing IE-specific kludges, and I've already got those to make seriously-ugly INPUT elements instead of BUTTON elements. I'm not adding more, especially when Firefox and Opera are both fine and free.)
GP doesn't say "cite sources", it says "dig up sources". They succeeded in that part.
Citations are seriously advanced. Like, oh, grade 6 or 7. Around the same time they make you do that more-than-three-minutes-less-than-five public speaking thing in front of the whole class.
"Bibliography" is such a long word, after all.
On my Mac, I'd far rather use VueScan than the crap that comes with scanners.
I had an old Microtek X6USB, the first cheap scanner with USB. It worked only with the provided drivers on Mac OS 8.1, nothing earlier, nothing later. But it worked fine with Xsane on Linux, or VueScan on Mac (and presumably Windows) as well.
I've got a much nicer scanner from Canon now, with the infra-red dust removal when scanning slides and negatives and everything. VueScan runs it nicely indeed.
My rule is, find out what hardware works with what software first. Then buy the hardware.
Don't connect /dev/random to /dev/urandom! It would be bad....
There's also a deadlock or some other lost race condition in some of the earlier 2.6 kernels that causes the entropy pool for /dev/random to dry up, and no new entropy would get added until a reboot. So, anything needing good randomness hangs waiting for /dev/random.
So, I've got a cron-job on those machines that watches the entropy pool and whines when they run out of random numbers. (I back-ported the fix to the must-run kernel on those servers, but the cron-job remains. Now it only triggers when something kills /tmp and it gets a false positive.)
If I'm on the 'net, I can order food....
And if there'd been a fire?
An earthquake?
A flood?
A fault in the mirror logic? (Like inappropriately comparing timestamps as text, and getting one-more-digit. Happened to a number of programs when time_t got one digit longer.)
Swine flu for Servers outbreak?
They didn't have backups. They had a high-availability configuration and no disaster recovery plan.
RAID can copy an error across all mirrors faster than you can hit ^C. Live copies are not backups.
Well, in this particular case, it's fairly apparent that the uncovered corner conditions are obvious during static analysis. And the incorrectly-weighted averaging.
Code that passes static analysis is not known good.
Code that fails static analysis is known bad.
Static analysis does not mean "it's pretty". It means the logic and arithmetic work.
I dunno... that far out of town, what else are you going to do but play games?
And aren't the Americans always going on about how it's OK that their DSL and 3G coverage is so horrible _because_ there's such a large proportion of the country that is sparsely populated?
I've done the pull-a-drive-out test on Mac OS X software RAID the old fashioned way... by accidentally stepping on the power strip controlling one of the external volumes the RAID is using.
But yes, if it isn't tested it WILL fail.
If it is tested, it MIGHT fail.
IME, the only reason to favour hardware RAID controllers is the battery-backed write-back cache. At least for mirroring modes; I don't really like parity modes, as they need too much simultaneous I/O on the disks, and disks are cheap enough today it's not worth it. (And the power consumption is WAY down from those full-height 5.25" SCSI monsters of 1993.)
But that works in JBOD (real JBOD, not "spanning with a stupid name) mode. So you can still use software RAID (Linux MD, AIX LVM mirrors, whatever) and get the benefits of the write-back cache and none of the "damn I need another card just like this one" downsides.
I was surprised that 2x the I/O on the PCI-X bus wasn't a problem. I really expected to have a measurable, consistent improvement from the hardware RAID controller, especially since it should have been able to drive two independent Ultra320 SCSI channels from a single PCI-X transaction. Especially given what the card cost.
Glad I wasn't the one paying for it....
I've done less extensive tests on other configs since then, and I can only find specific I/O patterns where the hardware card is a slight win; it's usually about the same as software RAID. You can do more by locating your hot data at the rim of the disk.
Probably not a big deal these days, but in the pin-through-hole era, you ran the risk of getting scraps of aluminum stuck to the pins on the circuit boards. The plastic anti-stat bags are resistant to that.