Thanks Xin, see reply inline below.
发件人: xin wang [mailto:[email protected]]
发送时间: 2017年6月28日 10:06
收件人: [email protected]
抄送: IETF ALTO
主题: Here is a review of draft-ietf-alto-performance-metrics-01
Dear authors of "ALTO Performance Cost Metrics”,
Your draft about extending cost metrics of ALTO looks very useful, especially
for varieties of applications with different criteria to make decisions for
traffic engineering. The following is a review of your draft:
a. Link Maximum Reservable Bandwidth and Link Residue Bandwidth
These three cost metrics look very similar but have different meanings
according to your draft. In my understanding, based on the concept of bandwidth
oversubscription, a reserved bandwidth could be larger than current remaining
bandwidth (equals to max bandwidth subtract the sum of all reserved bandwidth).
Then, Link Maximum Reservable Bandwidth should be slightly larger than Link
Residue Bandwidth. Is this right?
[Qin]: Short answer is Yes. Long answer is:
Residbw = maxbw - tunnel reservation bw
Maxresbw can be larger than maxbw if the link is oversubscirbed
Therefore Maxresbw could be larger than Residbw.
b. Unitless performance scores
In section 2.1, the draft suggests providing unitless performance scores for
privacy issues. In ALTO cost mode, it defines “ordinal” which indicates the
ranking of the cost values instead of actual costs. Do you suggest to add more
intelligent abstraction of cost values besides of ranking?
[Qin]: I think adding more intelligent abstraction of cost values besides of
ranking useful, but we need to identify use case to show deficiency of ranking
before introducing new intelligent abstraction.
c. Packet Delay Variation
The draft uses the minimum delay observed as the definition of packet delay
variation. I am not sure it is a good solution. There are some other approaches
you may consider, such as standard deviation.
[Qin]: The minimum delay observed is just used as reference delay, see the
complete definition as follows:
“
Metric name:
Packet Delay Variation
Metric Description:
To specify spatial and temporal aggregated jitter (packet delay
variation) with respect to the minimum delay observed on the
stream over the specified source and destination.
”
You are right, standard deviation will be used.
d. The unit of Periodic One Way Delay
The draft uses seconds as the unit of Periodic One Way Delay. I am not sure it
is typos or not since the example shows the powdelay cost values are 10, 20,
and 30.
[Qin]: It is not typo, it is intended. See section 7.4.3 of
draft-ietf-ippm-initial-registry-03.
We don’t define any new metric with new measurement unit here.
Also there are some typos you may fix:
Abstraction: add space between "BGP-LS,OSPF-TE";
Page 2: “delay sensitive” -> “delay-sensitive”;
Page 2: “a set new” -> “a set of new”;
Page 2: “ explicitely” -> “ explicitly”
Page 4: “ invlove” -> “ involve"
[Qin]: Fixed, thanks.
Best,
Xin
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto
