In 1987 I was offerred the opportunity to play bass for a nationally recognized folk performer. I had just started at my first software job a week before, but would have to go on tour a month later if I took the offer.
I said no.
I regret the decision to this day, despite being pretty successful in software.
Don't be an idiot - grab your dream job when you can. It's a lot better to be able to fail and say you tried than regret that...etc.
Having heard a few "behind the scenes" stories, Mr. Shore, apparently, needed a fair ammount of babysitting/coaxing to finish the ROTK score. A Newline Exec apparently spent a number of months in London up into November making sure that Mr. Shore showed up when he was needed.
I, too, found the ROTK score a bit of a rehash of the first two scores, so the rumors seem to be borne out by the final product.
Just to point out that most people with phone-service in "the boondocks" are rural, and most of those people are involved with providing the rest of us leeches with our food. If they all move away from "the boondocks", we starve.
Even if you use RSA or ttssh, that only hides the timing of your password keystrokes during session log in. If you use sudo or su and type a password, I'm pretty sure timing analysis could detect those events.
Don't forget that more than passwords are vulnerable. I use ssh to prevent eavesdropping, but apparently, ssh needs to do some keystroke timing hiding to improve security for the entire interactive session. (I'll let the expert paranoids figure out the best way to do that).
"Huntsville workers for Boeing accidentally threw away the two $375,000 tanks last month and later found a piece of their protective covering in the Huntsville landfill.
Boeing has said that if the tanks must be replaced, then NASA, not Boeing, must pay for them, due to terms of a contract between the organizations."
....
By coincidence, this month is "Property Awareness Month" at NASA MSFC...
This is a personal anecdote where I would say that the compiler generated better code than I would have.
I worked on an embedded network analyzer product using the intel gcc compiler for the i960. It contained about 16000 lines of C code (not trivial, but not large) that included a bare bones operating environment, network drivers, and the actual analysis code. Of these ~16000 lines, about 500 of them were in the "common critical path" and would execute on every incomming packet.
The intel gcc compiler for the i960 has this is really cool feature called "profile based optimization". You build an instrumented version of the executable, run a set of test data through it, get the profile data, and re-compile with the profile data. This optimization feature had the added advantage of closely integrating both the compiler and the linker.
The main benefits of this optimization included:
Setting the branch prediction bits for the most common cases.
Inlining functions with high hit counts : both within a module (traditional inlining), and across many modules (an advantage of linker integration)
Placing variables with high hit counts in faster memory (the i960 had 1k of on board, single-clock DRAM, and our card had both 2-clock SRAM and 4-clock DRAM)
Inlining all functions that get called only once in the executable.
I saw a 150% performance improvement over the non-profile based, but otherwise fully optimized builds.
The key 500 lines were interspaced through about six functions in four modules, often in functions where half the function was error/unusual path handling. Although I could have rewritten those 500 lines in asm, it would have been very hard to keep it integrated with the rest of the code. Also when I examined the instruction trace of those 500 lines, I found only two or three places where I would want to change things.
Intel's x86 compiler is supposed to have similar a set of profiling optimization features and another feature that's even more important on the x86 that lays out functions and data in the executable to reduce cache misses and page faults based on the profiling data.
In the "me too" catagory: Optimizing for Alpha processors is a nightmare best left to the compiler. Just try to single-instruction-step through some Alpha code and see how much you move around within the C source.
Slashdot Mirror
In 1987 I was offerred the opportunity to play bass for a nationally recognized folk performer. I had just started at my first software job a week before, but would have to go on tour a month later if I took the offer.
I said no.
I regret the decision to this day, despite being pretty successful in software.
Don't be an idiot - grab your dream job when you can. It's a lot better to be able to fail and say you tried than regret that...etc.
Peter
Having heard a few "behind the scenes" stories, Mr. Shore, apparently, needed a fair ammount of babysitting/coaxing to finish the ROTK score. A Newline Exec apparently spent a number of months in London up into November making sure that Mr. Shore showed up when he was needed.
I, too, found the ROTK score a bit of a rehash of the first two scores, so the rumors seem to be borne out by the final product.
peter
Just to point out that most people with phone-service in "the boondocks" are rural, and most of those people are involved with providing the rest of us leeches with our food. If they all move away from "the boondocks", we starve.
Not good.
peter
Even if you use RSA or ttssh, that only hides the timing of your password keystrokes during session log in. If you use sudo or su and type a password, I'm pretty sure timing analysis could detect those events.
Don't forget that more than passwords are vulnerable. I use ssh to prevent eavesdropping, but apparently, ssh needs to do some keystroke timing hiding to improve security for the entire interactive session. (I'll let the expert paranoids figure out the best way to do that).
4 March 2000: Boeing's missing tanks not explosive, Huntsville Times
"Huntsville workers for Boeing accidentally threw away the two $375,000 tanks last month and later found a piece of their protective covering in the Huntsville landfill.
Boeing has said that if the tanks must be replaced, then NASA, not Boeing, must pay for them, due to terms of a contract between the organizations."
By coincidence, this month is "Property Awareness Month" at NASA MSFC ...
This is a personal anecdote where I would say that the compiler generated better code than I would have.
I worked on an embedded network analyzer product using the intel gcc compiler for the i960. It contained about 16000 lines of C code (not trivial, but not large) that included a bare bones operating environment, network drivers, and the actual analysis code. Of these ~16000 lines, about 500 of them were in the "common critical path" and would execute on every incomming packet.
The intel gcc compiler for the i960 has this is really cool feature called "profile based optimization". You build an instrumented version of the executable, run a set of test data through it, get the profile data, and re-compile with the profile data. This optimization feature had the added advantage of closely integrating both the compiler and the linker.
The main benefits of this optimization included:
Setting the branch prediction bits for the most common cases.
Inlining functions with high hit counts : both within a module (traditional inlining), and across many modules (an advantage of linker integration)
Placing variables with high hit counts in faster memory (the i960 had 1k of on board, single-clock DRAM, and our card had both 2-clock SRAM and 4-clock DRAM)
Inlining all functions that get called only once in the executable.
I saw a 150% performance improvement over the non-profile based, but otherwise fully optimized builds.
The key 500 lines were interspaced through about six functions in four modules, often in functions where half the function was error/unusual path handling. Although I could have rewritten those 500 lines in asm, it would have been very hard to keep it integrated with the rest of the code. Also when I examined the instruction trace of those 500 lines, I found only two or three places where I would want to change things.
Intel's x86 compiler is supposed to have similar a set of profiling optimization features and another feature that's even more important on the x86 that lays out functions and data in the executable to reduce cache misses and page faults based on the profiling data.
In the "me too" catagory: Optimizing for Alpha processors is a nightmare best left to the compiler. Just try to single-instruction-step through some Alpha code and see how much you move around within the C source.
Peter (first /. post!)