![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
> Is QoS on TCP/IP reliable enough for applications that are > bit-rate-sensitive (like real-time video and toll quality voice)? The empirical answer is yes. I (I work at Cisco Systems via acquisition of Precept Software) find IP a perfectly adequate vehicle to move near-broadcast quality audio/video streams over IP networks, even networks with a significant number of routers and links of diverse types. Last May I put together a multi-vendor QoS/RSVP test in which we ran RTP/RTCP based audio/video and IP telephony over a network with four brands of routers connected by links of such diverse types as T-1, ADSL, fast ethernet, and standard ethernet. We used a number of host platforms, including Win 98, which, as shipped, includes RSVP. Interoperability of RSVP/QoS was quite good, indeed better than I had expected. It was about as non-proprietary as one can get. Of course, we still have a very long way to go before we achieve QoS across broad areas of the Internet. Aggregation techniques, such as diff-serve, are needed to keep router state within reasonable bounds. And we need to get policy controls in place so that we can prevent everyone from simply asking for, and receiving, preferential QoS values. And instability of routes is going to make it difficult to maintain QoS once a path is established. As one who has developed QoS enabled applications -- I suggest that one should not build an application that depends entirely on the network for QoS, for smoothing packets into isochronous bit streams. Rather, applications need to take an active part in working with the net and should be able to gracefully deal with failures to obtain the desired QoS. And applications need to avoid doing silly things (like forming IP fragments) or carving their media representations across packet boundaries so that a single lost packet causes the invalidation of data in adjacent parts of the media stream. QoS networking requires that one think in terms of the network as a whole, as *all* the layers working together. --karl--
Note Well: Messages sent to this mailing list are the opinions of the senders and do not imply endorsement by the IETF.