Please note: This project is no longer active. The website is kept online for historic purposes only.
If you´re looking for a Linux driver for your Atheros WLAN device, you should continue here .

Ticket #989 (new task)

Opened 16 years ago

Last modified 15 years ago

Sample rate control should be deprecated in favour of Minstrel

Reported by: benoit.papillault@free.fr Assigned to:
Priority: major Milestone:
Component: madwifi: rate module 'sample' Version: trunk
Keywords: Cc:
Patch is attached: 0 Pending:

Description (Last modified by mrenzmann)

sample rate control works well on good links. However, if the condition are changing from a good link to a bad link and then back to a good link, the rate is kept at 1 Mbit/s. When the link is bad, the average transmission time for all rates is increased to large values and then, when the link is good again, the average transmission time is updated only for few rates and even when it's updated, it does not converge to a lower value as expected (since EWMA is used).

To me, there are in fact two bugs:

  • EWMA does not seem appropriate compare to "last 10 seconds average" as exposed in John's thesis.
  • average transmission time is not updated so often, so old values are kept a LONG time

Change History

11/06/06 07:10:59 changed by mrenzmann

  • description changed.

(follow-up: ↓ 3 ) 03/26/07 05:58:18 changed by derek@indranet.co.nz

Hi,

we had similar findings to the description above. Our key observation was that the rate could by changed from automatic (with sample) to fixed, and choosing carefully, the throughput could be doubled (or more).

The simplest way to measure throughput was to copy a large file from one box to another, and see how long it took.

We wrote a rate control module based on sample, and on an earlier release in 2005 http ://sourceforge.net/mailarchive/forum.php?forum_id=33966&max_rows=25&style=flat&viewmonth=200501&viewday=5 In our assesment, the EWMA is the correct approach because it reduces the significance of old data and places more emphasis on newer data. We did try minor modifications to sample, in an endeavour to fix it. However, it became clear that a rewrite was required, as so much had to be changed.

Minstrel has been extensively tested in an adhoc environment. It has run on several different machines for weeks (3 * 7 * 24 hours).

This patch is against the svn of 2216.

Enjoy! Derek Smithies, <derek@indranet.co.nz>

(in reply to: ↑ 2 ) 03/26/07 06:16:36 changed by mrenzmann

I've opened a new ticket for Minstrel, to keep reviewing comments out of this ticket. The patch has also been moved there. Please follow up to #1222.

05/15/07 01:24:08 changed by mentor

  • type changed from defect to task.
  • summary changed from sample rate control is buggy to Sample rate control should be deprecated in favour of Minstrel.

Sample rate control module is buggy: yes. We now have an alternative, so we need to test minstrel to be able to move it to the default rate control module.