Project

General

Profile

Actions

Bug #713

closed

Shaper Wizard: When backlogged, high priority queues get zero bandwidth

Added by Aaron Berland over 13 years ago. Updated over 13 years ago.

Status:
Resolved
Priority:
Normal
Assignee:
-
Category:
Traffic Shaper (ALTQ)
Target version:
Start date:
07/02/2010
Due date:
% Done:

100%

Estimated time:
Plus Target Version:
Release Notes:
Affected Version:
2.0
Affected Architecture:

Description

Shaper wizard results for basic config:

qOthersLow - LinkShare: 5%, 500ms, 5% < This is a flat curve and redundant., 5% for the first 500ms and then 5% after that.
qOthersHigh - LinkShare: 0b, 200ms, 10% < No traffic allowed until service curve opens up!
qAck - LinkShare: 0b, 100ms, 20% < No ACKs allowed until service curve opens up!

Fix: remove m1 and d service curve values to allow traffic to flow immediately. Advanced users can set service curves if they want/need them. Don't prevent traffic from flowing by default!


Files

Actions #1

Updated by Ermal Luçi over 13 years ago

Can you please be more detailed on what the error is?

Actions #2

Updated by Aaron Berland over 13 years ago

There is no error message output.

The shaper config functions, but logically is not correct. During congestion, having a service curve start at 0 bandwidth drops packets so the low priority queue gets the bandwidth while the higher priority queues are blocked for a moment. Performance is dramatically reduced for acks and high priority traffic during congestion.

I believe this is why qDefault and qOthersLow can get better performance than the higher priority queues and not allow maximum performance.

Actions #3

Updated by Ermal Luçi over 13 years ago

  • Status changed from New to Feedback
  • % Done changed from 0 to 100
Actions #4

Updated by Josh Stompro over 13 years ago

I can confirm this as resolved. The qACK and QOthersHigh get created without the Link Share M1 or D options set now. So qACK and QOthersHigh traffic does not get starved for a certain number of ms when the queue is full.

Tested with Dec 2nd nanobsd snapshot.

Thanks Ermal
Josh

Actions #5

Updated by Ermal Luçi over 13 years ago

  • Status changed from Feedback to Resolved
Actions

Also available in: Atom PDF