Thanks Geng for your interesting comments, please see my reply inline below.

-Qin
发件人: Li, Geng [mailto:[email protected]]
发送时间: 2017年6月29日 9:34
收件人: [email protected]; [email protected]
主题: Review on draft-ietf-alto-performance-metrics-01 by Geng


Dear authors,



Your work is quite excellent, and here are some suggestions I have.



1.     The ALTO server should have the capability to track and record the most 
frequently used or requested cost metrics of each ALTO client. I’m not sure if 
it is true, that each client may only run one or two certain applications at 
most of the time. Then, the benefit is that the ALTO server can prepare the 
cost metrics in advanced and respond the clients fast after receiving the 
request.

[Qin]: This draft allows ALTO server collect and record a bunch of data on time 
period covering past and present, but I am not sure how you can prepare cost 
metrics in advance, unless you can predict the change pattern of some 
performance metric or predict the possible value of cost metric in the future.



2.     Another benefit of recording the type of cost metrics of each client is 
that the ALTO server can detect abnormal behavior of client or application, 
such as database resource leaks, illegal occupation, or other security issues.

[Qin]: The performance metric we exposed to the client is used to help the 
client to select the right access point or select appropriate site in the data 
center based on network performance criteria. I am not sure we can used these 
performance metric to detect abnormal behavior of client or prevent any 
resource leak unless you provide me  a clear use cases for that.

3. A variety of performance cost metrics also raise an interesting scheduling 
problem. For example, client “A” requests two metrics, client “B” just requests 
one, and client “C” requests three metrics, then how the ALTO server schedules 
to respond them in the constraint of given resources for workload processing 
and network transmitting. And for each client, what is the order of the cost 
metric types to be returned?

[Qin]: I am not sure we need a sophisticated scheduling feature here. The ALTO 
server will process request from each client by order, who request first, whose 
request will be processed first.



Hope these are helpful.



Best, Geng

Computer Science, Yale
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to