Amen! If you're not already sweating optimization (i.e., you've got some supremely high-performance code), then it's better to write straightforward code. Especially with C, which has pretty mature compilers.
Another thing to consider is that compilers for a modern RISC architecture have pretty intense optimization built in just to handle the instruction scheduling (re-ordering instructions to avoid pipeline stalls, etc.) that any trivial optimizations you might make would be "lost in the noise" anyway.
That said, the big optimizations will always be worthwhile: cacheing results so you don't have to read from a file/database again, using lazy initialization to avoid populating data structures you may not use, validating inputs so you don't get halfway through an expensive operation and then have to roll back the transaction and throw an error, etc. But moving loop invariants? Maybe in a new language with an immature compiler, or a scripting language (just how efficient is PHP, anyway? Python?), but any modern compiler will make that irrelevant.
programmers could by default write their programs to automatically run in....
HA HA HA HA HA HA!!! Obviously someone who has NO IDEA how to write software, or how much more difficult it is to coordinate parallel processes. This has been a hot research topic since I first used Concurrent Euclid back in the early 80s, and while the pthreads library and the Thread class in Java make it less difficult, it's still hard to do it well (and easy to do it so poorly you actually get worse performance).
Automatically? Oh yeah, I'll just drop in a #define SMP_MODE=TRUE and it'll just happen....
Well, up to a point. I'm getting pretty tired of my P-II/450 running Windows. OTOH, I still enjoy working on my 200MHz PPro running NetBSD. Soon I hope to enjoy the best of all worlds and pick up a Mac Mini. Which will probably be my main machine for another 4-5 years...
What makes you think the database is CPU-bound instead of I/O-bound?
If he's hosting his database on an 8-way server, I would imagine he's got the performance monitoring tools to figure out what it's doing and where the bottlenecks are. If you've got a decent RAID set-up (0+1, or "RAID 10" in some circles, hooked up with FibreChannel or the like), then the burden almost always shifts back onto the CPU.
Things to avoid are places where you walk in and they tell you "We'll do your site in $TECHNOLOGY, and nothing else."
I don't think this is really the big point that many people are making it out to be. If you've got a small firm that has a proven PHP/Java/ASP infrastructure, then I wouldn't count them out of the running. If you can't hire a large firm, then it's better to get a firm with a half-dozen $TECHNOLOGY developers than a firm with a half-dozen developers of which only two know any given technology. Then you'll likely have your technology dictated by which developers are available, rather than by requirements.
The company ended up choosing a consulting company who had just started, and didn't have any previous experience...
It is also several years late, and the final cost ended up being about 3 times what was originally estimated.
That's why you want to go with a major consulting firm -- they'll get your system delivered several years late, and five times the original estimate!
Seriously, I used to do LIMS for pharmaceutical companies, and the number of failed systems created by Anderson/D&T/PWC/E&Y is just staggering. My line used to be "Give me what $FIRM is charging you for the prototype, and I'll deliver a complete system." My first job was paid for out of the completion bonus the big company failed to earn when their completed system didn't meet *any* of its targets (functionality, security, response time, or hardware requirements). After my client threw US$3M down the tubes on that one, I picked up the slack for well under US$250K.
Oddly enough, I'm working with an e-commerce company now who does exactly what you're looking for. They've been around since before the dot-bomb meltdown, and have some fairly impressive names on their resume. Impressive enough that I'll assume you've already heard of them, so I won't shill for them here. Bottom line: contact their current clients and ask. If they hem and haw or say "we don't comment on vendor performance", then cross them off the list. Happy clients are good references and shouldn't be afraid to share their experiences.
Every part on a car would need to be tested for...
Nope. Just the major structures (frame, doors, engine, seats). Unless you want us to believe that the screws holding the glove box door on are vitally important to the accuracy of the simulation.
One of the things you have to realize about serious impacts is how the forces on major components (frames, subframes, friable joints and crush zones) overshadow the forces on other parts of the car.
And let's not forget that it was the bit-slice AMD 2901 at the heart of the VAX-11/780.
Still the nicest machine I've ever worked on, I'm looking forward to getting an emulator running (at probably 10x the speed of the original).
he [JMS] seemed to have trouble working with other writers
I suspect that anyone capable of working with Harlan Ellison is going to have trouble working with anyone else....
Actually, it's the "red tape" that makes the smaller contractors certain to lose. You need a small army of people just to manage the blizzard of forms and documents required, let alone do the real work of researching and developing a vehicle. And don't think that NASA's going to let them get away with a bunch of FEAs and flight sims, they're genna have to build and crush a few airframes to get real data.
Parent's not entirely wrong, though, a smaller company won't have to share the overhead of managing other divisions and projects as well as pay the salaries of people who have been since wings were made of fabric....
So you're saying that a CISC CPU will actually recompile my code like a hotspot-style JVM? It'll keep track of my execution profile across context switches and reoptimize execution paths based on data acces patterns? Gosh, and all those RISC guys have going for them is branch prediction with speculative execution.
Care to give me an example of a modern CISC CPU with this marvelous ability? Is this one of the Sun MAJC chips? Or is is possible you don't know a JVM from a CPU? Or VLIW from CISC? Your ass from a hole in the ground?
I mean, surely someone as smart as you knows that CISC makes ILP more difficult, which leads to issues with threading latency and bus scheduling on multi-CPU (and now multi-core) systems. I mean, they covered all that in your community college course, right? Personally, I miss CISC CPUs. I remember hand-writing assembler for the VAX-11, now there was an instruction set! What other CPU had a CRC instruction? Or the infamous EDITPC (edit packed character)? Or a single instruction to load a process context? And what modern architecture has the FPD bit for restartable instructions (or restartable instructions, even)?
Oh, and I do know what JIT means. I also know that, while PBO has been around for years, no CPU does it: the code has to be specially instrumented and run so that statistics can be generated and the resulting profile applied to a subsequent optimization pass (read: recompile and re-link).
Now why don't you go cry to your homeroom teacher and let the big people talk, ok honey?
Don't know how "scientific" this one is, all the brands listed below Coke are businesses offering services or products to other businesses. I doubt your average football (pronounced "soccer") fan would recognize Intel, but they sure as hell know Marlboro.
I posit that the most recognized brands are going to be those of common consumer goods, especially those that are bad for you. E.g., Marlboro, Budweiser (in N.A.), Bacardi et. al. Also McDonalds, Starbucks, and other places ordinary people are likely to visit. Biggest audience == most recognized (by definitions). As such, I don't think the brands most recognized by suits cut it.
sending CISC instructions to the CPU which are then converted into RISC is faster than RISC because the CPU can figure out good ways to rearrange the ops
Which is what the compiler should be doing. And what a good RISC compiler does. The benefit is obviously that a better compiler will give better performance. It's easy to change compilers, a bit trickier to rewrire your CISC CPU when somebody finds a new optimization.
Compile-time optimizations for a RISC CPU involve grouping and reordering instructions to ensure that all pipelines to all execution units stay full. This is appropriate to ALL programs, not just graphics. A lot of CISC compilers still seem to think that moving loop invariants is "edgy"...
Slashdot Mirror
Another thing to consider is that compilers for a modern RISC architecture have pretty intense optimization built in just to handle the instruction scheduling (re-ordering instructions to avoid pipeline stalls, etc.) that any trivial optimizations you might make would be "lost in the noise" anyway.
That said, the big optimizations will always be worthwhile: cacheing results so you don't have to read from a file/database again, using lazy initialization to avoid populating data structures you may not use, validating inputs so you don't get halfway through an expensive operation and then have to roll back the transaction and throw an error, etc. But moving loop invariants? Maybe in a new language with an immature compiler, or a scripting language (just how efficient is PHP, anyway? Python?), but any modern compiler will make that irrelevant.
Hey, dude -- it's 2005 calling. Why don't you drop by and see us sometime?
HA HA HA HA HA HA!!! Obviously someone who has NO IDEA how to write software, or how much more difficult it is to coordinate parallel processes. This has been a hot research topic since I first used Concurrent Euclid back in the early 80s, and while the pthreads library and the Thread class in Java make it less difficult, it's still hard to do it well (and easy to do it so poorly you actually get worse performance).
Automatically? Oh yeah, I'll just drop in a #define SMP_MODE=TRUE and it'll just happen....
Well, up to a point. I'm getting pretty tired of my P-II/450 running Windows. OTOH, I still enjoy working on my 200MHz PPro running NetBSD. Soon I hope to enjoy the best of all worlds and pick up a Mac Mini. Which will probably be my main machine for another 4-5 years...
If he's hosting his database on an 8-way server, I would imagine he's got the performance monitoring tools to figure out what it's doing and where the bottlenecks are. If you've got a decent RAID set-up (0+1, or "RAID 10" in some circles, hooked up with FibreChannel or the like), then the burden almost always shifts back onto the CPU.
I don't think this is really the big point that many people are making it out to be. If you've got a small firm that has a proven PHP/Java/ASP infrastructure, then I wouldn't count them out of the running. If you can't hire a large firm, then it's better to get a firm with a half-dozen $TECHNOLOGY developers than a firm with a half-dozen developers of which only two know any given technology. Then you'll likely have your technology dictated by which developers are available, rather than by requirements.
It is also several years late, and the final cost ended up being about 3 times what was originally estimated.
That's why you want to go with a major consulting firm -- they'll get your system delivered several years late, and five times the original estimate!
Seriously, I used to do LIMS for pharmaceutical companies, and the number of failed systems created by Anderson/D&T/PWC/E&Y is just staggering. My line used to be "Give me what $FIRM is charging you for the prototype, and I'll deliver a complete system." My first job was paid for out of the completion bonus the big company failed to earn when their completed system didn't meet *any* of its targets (functionality, security, response time, or hardware requirements). After my client threw US$3M down the tubes on that one, I picked up the slack for well under US$250K.
Oddly enough, I'm working with an e-commerce company now who does exactly what you're looking for. They've been around since before the dot-bomb meltdown, and have some fairly impressive names on their resume. Impressive enough that I'll assume you've already heard of them, so I won't shill for them here. Bottom line: contact their current clients and ask. If they hem and haw or say "we don't comment on vendor performance", then cross them off the list. Happy clients are good references and shouldn't be afraid to share their experiences.
Me too.
And if I need to fix it, I know I can always rely on this to get it sorted out.
Every part on a car would need to be tested for...
Nope. Just the major structures (frame, doors, engine, seats). Unless you want us to believe that the screws holding the glove box door on are vitally important to the accuracy of the simulation.
One of the things you have to realize about serious impacts is how the forces on major components (frames, subframes, friable joints and crush zones) overshadow the forces on other parts of the car.
And let's not forget that it was the bit-slice AMD 2901 at the heart of the VAX-11/780. Still the nicest machine I've ever worked on, I'm looking forward to getting an emulator running (at probably 10x the speed of the original).
he [JMS] seemed to have trouble working with other writers
I suspect that anyone capable of working with Harlan Ellison is going to have trouble working with anyone else....
Yeah, but you've got a beautiful glossy coat!
Actually, it's the "red tape" that makes the smaller contractors certain to lose. You need a small army of people just to manage the blizzard of forms and documents required, let alone do the real work of researching and developing a vehicle. And don't think that NASA's going to let them get away with a bunch of FEAs and flight sims, they're genna have to build and crush a few airframes to get real data. Parent's not entirely wrong, though, a smaller company won't have to share the overhead of managing other divisions and projects as well as pay the salaries of people who have been since wings were made of fabric....
So you're saying that a CISC CPU will actually recompile my code like a hotspot-style JVM? It'll keep track of my execution profile across context switches and reoptimize execution paths based on data acces patterns? Gosh, and all those RISC guys have going for them is branch prediction with speculative execution.
Care to give me an example of a modern CISC CPU with this marvelous ability? Is this one of the Sun MAJC chips? Or is is possible you don't know a JVM from a CPU? Or VLIW from CISC? Your ass from a hole in the ground?
I mean, surely someone as smart as you knows that CISC makes ILP more difficult, which leads to issues with threading latency and bus scheduling on multi-CPU (and now multi-core) systems. I mean, they covered all that in your community college course, right? Personally, I miss CISC CPUs. I remember hand-writing assembler for the VAX-11, now there was an instruction set! What other CPU had a CRC instruction? Or the infamous EDITPC (edit packed character)? Or a single instruction to load a process context? And what modern architecture has the FPD bit for restartable instructions (or restartable instructions, even)? Oh, and I do know what JIT means. I also know that, while PBO has been around for years, no CPU does it: the code has to be specially instrumented and run so that statistics can be generated and the resulting profile applied to a subsequent optimization pass (read: recompile and re-link). Now why don't you go cry to your homeroom teacher and let the big people talk, ok honey?
Don't know how "scientific" this one is, all the brands listed below Coke are businesses offering services or products to other businesses. I doubt your average football (pronounced "soccer") fan would recognize Intel, but they sure as hell know Marlboro. I posit that the most recognized brands are going to be those of common consumer goods, especially those that are bad for you. E.g., Marlboro, Budweiser (in N.A.), Bacardi et. al. Also McDonalds, Starbucks, and other places ordinary people are likely to visit. Biggest audience == most recognized (by definitions). As such, I don't think the brands most recognized by suits cut it.
A confusing and infuriating comment coming from someone with a "help me get a free Mac-Mini" link in their sig...
Well, if you're going to actually READ the article...
No Nike, no Marlboro, no Coke? Who picked the brands in that article, the readers of Us magazine?
sending CISC instructions to the CPU which are then converted into RISC is faster than RISC because the CPU can figure out good ways to rearrange the ops Which is what the compiler should be doing. And what a good RISC compiler does. The benefit is obviously that a better compiler will give better performance. It's easy to change compilers, a bit trickier to rewrire your CISC CPU when somebody finds a new optimization. Compile-time optimizations for a RISC CPU involve grouping and reordering instructions to ensure that all pipelines to all execution units stay full. This is appropriate to ALL programs, not just graphics. A lot of CISC compilers still seem to think that moving loop invariants is "edgy"...
WOULD have been Andy's, but he was just bought out by Carole's Tats & Wraps ("Color-fast and low-carb!").