[asterisk-biz] Experimental/new VoIP rate search engine.

[Trixter aka Bret McDanel]

On Sun, 2009-01-04 at 22:33 -0800, Nitzan Kon wrote:
> In other words, if a provider tests well from one point in
> Europe - chances are they'll be OK from anywhere within the
> EU.
> 

that may change, in amsterdam this year they are deploying residential
gigabit fiber to most of the city.  The plan is at least 50% early this
year and expand to the rest later on or something.  One problem they
noticed is that many desktops cant sustain gigabit downloading, the disk
is too slow, remember we are talking residential here.  But if you see
500k people or so start using that you will see saturation and unless
the peering is beefed up to compensate there will be issues.

On the flipside, AMS-IX charges something like 2500 euro/month for a
10gbps link or something, so at least locally providers can increase
peering, but getting that data outside of the local networks may require
some retooling on the network providers side and that is likely to take
some time.

But then the same concept still applies, tests done here are not that
applicable to people outside of this area.  People in south africa,
pakistan, japan, etc will have little use for them.  So unless they are
done everywhere they become somewhat meaningless.  Additionally
destination numbers are likely to have some effect in terms of quality,
if I call a DE number it may go over a different provider than if I call
a NL number.  

Basically it means that there is a lot of testing that has to be done,
and the testing has to be blind to the providers to prevent any of them
from rigging the test.  Not saying that they will, but to remove the
potential they cant know which ones are the test calls.  

for the rate search engine, to use a quality metric in its output, you
would have to have servers all over the world, and they would have to
call numbers all over the world or the test will be somewhat incomplete
and not usable to anyone that isnt somewhere local to a test facility,
placing calls to the same types of numbers that were tested.  I do not
see this level of testing as practical, especially for a volunteer
effort given the scope that it would take to make it a global metric.


Further latency isnt as big of a deal, its loss and jitter.  If latency
was a killer you would never be able to send a fax from north america to
say india or something :)  Jitter can be dealt with by way of a jitter
buffer, providing its not too bad.  Loss is harder to deal with, some
codecs like ilbc have a higher MOS score with loss, at a higher cpu cost
though, while others do not conceal loss at all or do it poorly, but are
more cpu friendly.  If you look at g.729 vs ilbc for example, g.729
falls much faster than ilbc in its MOS score with moderate loss.  


-- 
Trixter http://www.0xdecafbad.com     Bret McDanel
pgp key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8AE5C721

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url : http://lists.digium.com/pipermail/asterisk-biz/attachments/20090105/914b3d1a/attachment.pgp