Originally posted by Aim64C
I suppose this would depend, highly, upon the specifics of the system and the nature of the upgrades.
Bus and socket technology have not really changed in computers over the past 15 years. They've simply gotten more creative with the I/O waveforms.
In that case - it's really a matter of just how scalable your system is.
Bus and socket technologies have hugely changed in computers over the past 15 years, with significant speed increases and error handling along the
You have materials changes, configuration changes, grounding changes, tunneling changes - loads of improvements over the past decade alone.
What was scalable 15 years ago is not comparable to what is considered scalable today - you do not see the interconnects used 15 years ago in HPCs
used today in modern HPCs. Interconnect technology is a rapidly moving target
Yes, and no. Parallel distributed floating point processing is highly scalable and the RISC architecture makes all complex instructions virtual in
nature. Task-specific hardware is, thus, unnecessary. All of the processes conducted by modern avionics are massively parallel in nature - from
analysis of radar returns to stabilization and kinematics.
A hardware DSP will always, always *always* outperform a distributed architecture for the same power, cooling and space requirements. Always.
The only reason we do not use DSPs in everything is because they are essentially single-task orientated, while most super computers are designed to be
used for multiple tasks.
But a modern aircraft is full of tasks that can be happily processed by single-task chips - there is extremely little general purpose computing
involved in a modern aircrafts systems. Even the tasks you highlight can be handled much better by a hardware DSP than a general purpose computing
architecture - they are well defined problems with well defined solutions, you put X in and want Y out. You will always put X in and want Y out.
Therefor you do not use a general purpose computing architecture, you use specific hardware to do that.
And in a modern aircraft, you really do want tasks to be split out - you do not want your FBW system to suddenly find that its being bumped from the
schedule because the radar system wants a little more processing time.
And equally, you don't want your targetting system to be bumped because the FBW system is having a hard time keeping up with a limited amount of time
on the system.
So you are back to independent systems that tasks can monopolise.
In short, the nature of RISC programming is that you customize the code to the function and not the hardware to the function.
Which is fine in general purpose computing - you don't want your architecture to be hamstrung by decisions you made to lean the capabilities to one
task or another.
But its another thing when you know that your system isn't going to be doing general purpose computing, and instead is going to be handling a lot of
very well defined tasks.
It is not only more cost-effective, but more power-efficient, as well.
Its cost effective in certain circumstances, but it can be a huge problem in others - RISC was much vaunted in the 1980s and 1990s as the saviour of
computing, but in the end it just turned out to be another architecture that was useful in some circumstances and not others.
RISC is insanely useful in low branching, low pipeline length systems, but it does not inherently decrease the complexity of a system - you still code
in high level complex instructions, and rather than being translated in an almost 1:1 fashion by the compiler, you end up with your high level
instruction being translated to a dozen or more RISC instructions. More gets executed - which is fine if you have a short pipeline and excellent
branch prediction, but not if the architecture is poor in that area.
The F-23 had the vast majority of its avionics virtualized. This is, precisely, the trend that has been mirrored in the server industry 10-15 years
later. Not to mention I have a face-to-face source on this.
Lets remember that the F-23 never existed - there was the YF-23 and a proposal.
None of the YF-23 would have gone into production, if the F-22's and F-35's births are any indication - we would have ended up with an aircraft that
looked vaguely like the YF-23 but none of its systems would have been carried over.
So we have no idea what the F-23 would have looked like.
And yes, there has been a great virtualisation trend in the server market, but as an accomplished IT engineer who has handled budgets in the millions
of dollars for hardware and software infrastructures, there are loads of tasks I would never, *ever* virtualise and I do cast scorn on those who do
virtualise those tasks.
Virtualisation does not solve every problem, it solves some problems and creates huge ones of its own.