Hi,
We have been discussing the Tequila draft and have two points that we'd like
to raise.
1) Section 3.5 of the draft discusses Performance Guarantees using four
parameters: delay, jitter, packet loss and throughput and these service
guarantees may be quantitative or qualitative.
Something we consider essential is a definition of time periods to use in
measuring the SLS, i.e. whether these parameters represent the overall
average for the entire duration of the SLS or whether they indicate the
worst acceptable time periods for the SLS duration.
An interpretation of an SLS over 2 months with a specified packet loss of
10E-3, could simply require the average packet loss over the entire 2 months
to be less than 10E-3. If a time period (e.g. 1 hour) is specified, then
the SLS
would require the average packet loss over any 1 hour period within the 2
months to be less than 10E-3.
Obviously, the 2 SLSs above are quite different in requirements, and it is
important for the customer to be able to clearly define the SLS service
required.
2) The second issue we would like to raise is the concept of blocking and
associated probability. We can see a use for a second type of SLS for
connection oriented services. In this SLS, the user defines the
characteristics
either for the aggregate traffic level or for the individual flow, as well
as a traffic
profile (inter-arrival times, holding times, Bandwidth, etc) and blocking
probability
for the individual connections.
When the individual connections are established, an admission control
request is made for them. The operator may thus reject individual sessions,
but the SLS defines what level of these services the operator should
dimension for. This may allow the operator to oversubscribe their network.
The advantage with this SLS for the user is they can maintain stringent QoS
per flow requirements due to the admission control, even when the network is
oversubscribed.
Regards,
Gabor and Kamal
This archive was generated by hypermail 2b29 : Wed Nov 01 2000 - 21:46:31 CET