07-29-2015, 11:07 AM
I don't know how many members are tech-minded, but I am and we're talking about Waterloo, so I'm assuming it's fair to talk about :)
A problem I keep having with thinking about this construction and "Why are they doing X then Y instead of doing X+Z1 then Y+Z2 for parallelism?" is because, unlike in modern computers, the actors in this project aren't general-purpose.
CPUs (and, increasingly, GPUs (see OpenCL and GPGPU architectures)) are general-purpose and their cores can run in parallel to maximize cowork. Individual cores could be instructing the disk controller to fetch from storage, encrypting a block of memory using AES, decoding an MP3 frame for gain adjustment, and supplying crosstalk between physics hardware and a running computer game... and anything else we want. They are programmable.
Human workers aren't. Either the work is specialised so that workers are unable to do it (why don't we lay bundles of dark fibre for a rainy day municipal network when we tear up? Because it requires specialist training), or the work is governed in such a way that workers aren't allowed to do it (why don't we bury those overhead lines that are susceptible to cars and weather in addition to being eyesores? Because WNH owns the hydro lines and the poles, telcos own the comm lines, and so forth. The people tearing up the street have no power over any of those parties.).
So it becomes a matter of both resource allocation (where to send the specialists) and synchronization (getting all the permissions and crews in order before start-of-work). From anyone who's studied NP-completeness, scheduling isn't an easy problem, even in a finite system.
There are ways to cut this gordian knot: for the fibre case there are municipalities that have a self-trained populace acquiring permission to and actually renting and operating the earth movers to go about burying fibre across the hills and dales to build a municipal fibre network (paying for the interconnect fee and cabinet rental at the POP through subscriptions much lower than competing narrowband competitors). However, as density increases, these problems get harder and the complexity increases in a non-polynomial way.
Anyway, I hope this ramble helps people who aren't just me understand why there's inefficiencies in the system. The tl;dr is that this is the price for generating a complex system. In return we get this wonderful region of ours.
I think it's worth the price of admission.
A problem I keep having with thinking about this construction and "Why are they doing X then Y instead of doing X+Z1 then Y+Z2 for parallelism?" is because, unlike in modern computers, the actors in this project aren't general-purpose.
CPUs (and, increasingly, GPUs (see OpenCL and GPGPU architectures)) are general-purpose and their cores can run in parallel to maximize cowork. Individual cores could be instructing the disk controller to fetch from storage, encrypting a block of memory using AES, decoding an MP3 frame for gain adjustment, and supplying crosstalk between physics hardware and a running computer game... and anything else we want. They are programmable.
Human workers aren't. Either the work is specialised so that workers are unable to do it (why don't we lay bundles of dark fibre for a rainy day municipal network when we tear up
So it becomes a matter of both resource allocation (where to send the specialists) and synchronization (getting all the permissions and crews in order before start-of-work). From anyone who's studied NP-completeness, scheduling isn't an easy problem, even in a finite system.
There are ways to cut this gordian knot: for the fibre case there are municipalities that have a self-trained populace acquiring permission to and actually renting and operating the earth movers to go about burying fibre across the hills and dales to build a municipal fibre network (paying for the interconnect fee and cabinet rental at the POP through subscriptions much lower than competing narrowband competitors). However, as density increases, these problems get harder and the complexity increases in a non-polynomial way.
Anyway, I hope this ramble helps people who aren't just me understand why there's inefficiencies in the system. The tl;dr is that this is the price for generating a complex system. In return we get this wonderful region of ours.
I think it's worth the price of admission.