Assuming a wavelength window of 1500nm to 1580nm, theres an optical frequency range of some 10 Tb/s (assuming my quick calculation is correct).
Accordingly there will need to be fancy coding methods before 6.4Gb/s and above will be reached (eg. methods that trade bandwidth for signal/noise such as used in Modems).
Actually thats all pretty accademic anyway, since the limits for non-linear interactions between the wavelengths will be reached much sooner. Causing at best loss of signal, at worse distruction of the fibre....
Its all vey good to say 80 channels in one breath, and 80 Gb/s per channel in the next. Its an invalid assumption however to assume they can be combined since the effect of modulating one channel results in wavelength "sidebands" that affect the adjacent channels.
This is from my experiences years ago, but I dont think the laws of physics have changed since then...
Actually the table needs to be quite a bit bigger than 8MB, as there will be multiple routes to destinations, and other route properties (also isnt 2^28 bits == 32MBytes not 8).
Looking this up fast is hard, even using the techniques you describe which are already used in modern routers with distributed switching architectures.
The problem in both telephone and internet networks boils down to a good clean hierarchy versus the ability to change providers and keep old addresses. This later obviously breaks any hierarchy that may have been in the system. Using namelookup mechanisms is a saving grace, as people can renumber and preserve structure. Having 128 bits of address space helps a little (as users dont have to come back for more addresses). IPv6 isnt the whole answer though - indeed some of the IPv6 address allocation procedures Ive seen look worse than what we currently have for IPv4.
Any one who uses "class A/B/C" should be immediately corrected because they cause immense confusion. Its no harder to say "slash 8" than "class A", a lot clearer, and try saying "2 bits longer than a class B" instead of "slash 18". Its amazing really how many "supposed experts" keep refering to class A/B/C in a non-historical context.
Unfortunately theres still some software that contains classful bugs. All Solaris up to at least 2.5.1 (they route without a netmask !), DNS-servers, and Ciscos IOS for example..... Linux has remarkably few:)
Slashdot Mirror
Assuming a wavelength window of 1500nm to 1580nm, theres an optical frequency range of some 10 Tb/s (assuming my quick calculation is correct).
Accordingly there will need to be fancy coding methods before 6.4Gb/s and above will be reached (eg. methods that trade bandwidth for signal/noise such as used in Modems).
Actually thats all pretty accademic anyway, since the limits for non-linear interactions between the wavelengths will be reached much sooner. Causing at best loss of signal, at worse distruction of the fibre....
Its all vey good to say 80 channels in one breath, and 80 Gb/s per channel in the next. Its an invalid assumption however to assume they can be combined since the effect of modulating one channel results in wavelength "sidebands" that affect the adjacent channels.
This is from my experiences years ago, but I dont think the laws of physics have changed since then...
Cheers
Dave P
Actually the table needs to be quite a bit bigger than 8MB, as there will be multiple routes to destinations, and other route properties (also isnt 2^28 bits == 32MBytes not 8).
:)
Looking this up fast is hard, even using the techniques you describe which are already used in modern routers with distributed switching architectures.
The problem in both telephone and internet networks boils down to a good clean hierarchy versus the ability to change providers and keep old addresses. This later obviously breaks any hierarchy that may have been in the system. Using namelookup mechanisms is a saving grace, as people can renumber and preserve structure. Having 128 bits of address space helps a little (as users dont have to come back for more addresses). IPv6 isnt the whole answer though - indeed some of the IPv6 address allocation procedures Ive seen look worse than what we currently have for IPv4.
Any one who uses "class A/B/C" should be immediately corrected because they cause immense confusion. Its no harder to say "slash 8" than "class A", a lot clearer, and try saying "2 bits longer than a class B" instead of "slash 18". Its amazing really how many "supposed experts" keep refering to class A/B/C in a non-historical context.
Unfortunately theres still some software that contains classful bugs. All Solaris up to at least 2.5.1 (they route without a netmask !), DNS-servers, and Ciscos IOS for example..... Linux has remarkably few