QOS or more bandwidth
pete at kruckenberg.com
Tue May 29 16:10:21 UTC 2001
On Tue, 29 May 2001, Ukyo Kuonji wrote:
> --- Sean Doran Wrote ---
>>In the absence of such a document, from which anyone can build
>>an interoperable "true" "end to end" QoS system into his or her
>>product, I am tempted to believe that such buzzwords are weapons
>>in a DoS attack by marketroids.
> Yeah, I have to agree with you there. I have long been
> of the opinion that all these are spouted off by
> marketing types and ATM biggots. As far as I am
> concerned, a properly designed and run network makes QoS
> and MPLS more a hassle than a benefit.
Constrain a network to a homogeneous application set, and
then make sure that expectations for reliability and
consistent performance are low, you can probably design it
any way you'd like.
Look at how telcos handled (whether done well or not)
multiple traffic types. Typical telco has multiple networks,
usually ultimately based on a voice-oriented technology
(SONET, TDM, ATM), with application-layer integrations at
some limited points (ie frame-relay<->ATM internetworking,
ISDN<->frame-relay, DSL<->ATM, modem->ATM offloading).
Some telcos have financial disincentives to adopting new
technology, but they still have demonstrated not only the
need for QoS in a multi-service network, but some of the
benefits of interconnecting networks to gain from economies
of scale and statistical multiplexing.
>>TDM. What flavour would you like? SONET/SDH? PDH? "Virtual >dark
>>fibre"? ITU-Grid optics?
> That works very well for always on type service, but in
> reality, these type of services are not always on.
> They are periodic, at best, and the time between the
> periods can be hours to days. Right now they ARE over
> TDM, but TDM is expensive to supply (payback time and
> all). Data services (ATM or IP) has a much better
> payback time. And if you can get the stat mux gain, all
> the better.
Voice-over-IP benefits from statistical multiplexing as
much, if not moreso, than any other application. A
toll-quality voice call runs at ~5-6kbps (factoring silence
suppression and RTP header compression) vs. 8kbps across
The VoIP QoS problem is interesting. Barring congestion in
the network, VoIP just has a problem with the fact that IP
communications are frame-oriented (and a VoIP packet gets
behind a 1536-byte Ethernet frame in the transmit queue).
Frame-oriented protocols make sense for shared media like
Ethernet, but why couldn't a point-to-point link (or other
'exclusive' media) allow a (low-overhead) byte- or
cell-oriented mode as well as frame-oriented mode. VoIP
would get along much better.
More information about the NANOG