there's some very fancy codes that digital fountain and other
companies doing FEC based reliable multicast have - see tornado codes
and other references
can this and the TCP discussion move to end2end-interest
[EMAIL PROTECTED]
coz its a more appropriate (focussed) place!
In message <[EMAIL PROTECTED]>, Christian
Huitema typed:
>>Yes, in theory you may obtain an arbitrary residual rate for any given
>>input rate + overhead, if you are willing to make your codes long
>>enough. That may quickly become a tad pricy, however. There is also the
>>additional problem that most demonstrations assume errors are randomly
>>distributed. If the error distribution is heavy tailed, which is the
>>case on very many technologies, all bets are off... This is one of the
>>ways to not achieve the desired result.
>>
>>-- Christian Huitema
>>
>>> -----Original Message-----
>>> From: James P. Salsman [mailto:[EMAIL PROTECTED]]
>>> Sent: Friday, February 09, 2001 4:22 PM
>>> To: Christian Huitema
>>> Cc: [EMAIL PROTECTED]
>>> Subject: ECC limits? (was RE: An alternative to TCP ...)
>>>
>>>
>>> >... To reach 10 Gbps, you will need a BER of 1.E-14....
>>> > There are very many ways to not achieve that....
>>>
>>> Are there any limits to what error correcting codes can provide?
>>>
>>> If you have 15 Gbps with a natural error rate of 1 in 100,
>>> I think you can still get better than 1 in 1e14 bit errors
>>> at 10 Gbps with traditional techniques such as Reed-Solomon.
>>>
>>> Please correct me if I am wrong.
>>>
>>> Cheers,
>>> James
>>>
>>
cheers
jon
