[Codel] CoDel Algorithm comments and suggestions

[Codel] CoDel Algorithm comments and suggestions

[Agarwal, Anil]

[-- Attachment #1: Type: text/plain, Size: 844 bytes --]

All,

Here are a number of suggestions to improve CoDel algorithm performance, especially under extreme traffic conditions. The suggested changes are in the spirit of the CoDel algorithm and help to make state variable changes more smooth and continuous, rather than abrupt. I confess I have not gone thru all the exchanges on the mailing list, so some items may have been discussed and disposed of before.

Also, see enclosed another document with some simple analysis on the dynamics of CoDel. Nothing earth-shaking, but it was fun and useful to derive the equations.

I have tested the algorithm - with and without the changes - with my own quick implementation of CoDel - with simple unresponsive UDP traffic.
Not tested in ns2 simulations with realistic traffic and various TCP flavors.

Regards,
Anil Agarwal
ViaSat Inc.

 

[-- Attachment #2: aqm-codel-comments.docx --]
[-- Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document, Size: 49519 bytes --]

[-- Attachment #3: aqm-codel-analysis.docx --]
[-- Type: application/vnd.openxmlformats-officedocument.wordprocessingml.document, Size: 52353 bytes --]

Re: [Codel] CoDel Algorithm comments and suggestions

[Agarwal, Anil]

Please ignore item #1 in the comments doc - I misread the CoDel spec. Sorry.

Anil

-----Original Message-----
From: codel-bounces@lists.bufferbloat.net [mailto:codel-bounces@lists.bufferbloat.net] On Behalf Of Agarwal, Anil
Sent: Tuesday, June 16, 2015 1:12 PM
To: codel@lists.bufferbloat.net
Subject: [Codel] CoDel Algorithm comments and suggestions

All,

Here are a number of suggestions to improve CoDel algorithm performance, especially under extreme traffic conditions. The suggested changes are in the spirit of the CoDel algorithm and help to make state variable changes more smooth and continuous, rather than abrupt. I confess I have not gone thru all the exchanges on the mailing list, so some items may have been discussed and disposed of before.

Also, see enclosed another document with some simple analysis on the dynamics of CoDel. Nothing earth-shaking, but it was fun and useful to derive the equations.

I have tested the algorithm - with and without the changes - with my own quick implementation of CoDel - with simple unresponsive UDP traffic.
Not tested in ns2 simulations with realistic traffic and various TCP flavors.

Regards,
Anil Agarwal
ViaSat Inc.