| 1 |
On Mon, 05 Nov 2007 22:23:36 -0800 |
| 2 |
Xamindar <junkxamindar@×××××.com> wrote: |
| 3 |
|
| 4 |
> I have tried following the howto here: |
| 5 |
> http://gentoo-wiki.com/HOWTO_Packet_Shaping |
| 6 |
> But it doesn't work. First of all it ends up limiting both upload AND |
| 7 |
> download. I have tried a few different ways with all the same result. |
| 8 |
> Anyone know what is wrong here? |
| 9 |
|
| 10 |
|
| 11 |
=== |
| 12 |
Some "get annoyed and don't read from this point on" stuff :) |
| 13 |
|
| 14 |
All that follows represents my personal conclusions based on my |
| 15 |
personal trial and error experience. These conclusions may be totally |
| 16 |
wrong. They are not based on any official sources of information. |
| 17 |
Please, consider seeking answers at the "HTB Home" site: |
| 18 |
http://luxik.cdi.cz/~devik/qos/htb/ |
| 19 |
=== |
| 20 |
|
| 21 |
First. |
| 22 |
$UPRATE and "$UPCELL" should be equal to each other and *must* be at |
| 23 |
their maximum equal to your full upload bandwidth as granted by the |
| 24 |
ISP. Even better if they are a little bit smaller (let's say 5-10%, |
| 25 |
YMMW). Otherwise the shaper gives unpredictable results such as |
| 26 |
significant delays followed by uncontrolled bursts. |
| 27 |
|
| 28 |
Next. Basically you have 2 choices. (1) To share the unused bandwidth |
| 29 |
between the classes or (2) to give them strict limits. |
| 30 |
|
| 31 |
(1) Shared traffic |
| 32 |
|
| 33 |
The sum of the rates should be equal (or less) to the rate of the parent |
| 34 |
class. In your case: |
| 35 |
|
| 36 |
$P2PRATE + $PRIORATE[1-4] = $UPRATE |
| 37 |
|
| 38 |
And the individual "cells" should be equal (or less) than $UPRATE (as |
| 39 |
you have done) |
| 40 |
|
| 41 |
This way a given class has "guaranteed" speed = "its rate" and is |
| 42 |
allowed to take (borrow) the unused traffic from any other siblings (up |
| 43 |
to the size of its "cell"). |
| 44 |
|
| 45 |
In your example 128+65+46+27+8 > 152. Consider changing the values to |
| 46 |
get equal sign instead of "greater than". |
| 47 |
|
| 48 |
(2) Strict limits |
| 49 |
|
| 50 |
(I doubt you wont this) |
| 51 |
|
| 52 |
Every class has "its rate"="its cell" and the sum of the rates should |
| 53 |
not be bigger than the traffic allocated for the parent class. This |
| 54 |
means 1:1 becomes useless in you case. |
| 55 |
|
| 56 |
Last. All other cases (like "giving more than one has") seem to me |
| 57 |
irrational. |
| 58 |
|
| 59 |
Please, give some feed back on your results if you decide to follow the |
| 60 |
above info as advices. |
| 61 |
|
| 62 |
|
| 63 |
-- |
| 64 |
Best regards, |
| 65 |
Daniel |
| 66 |
-- |
| 67 |
gentoo-user@g.o mailing list |
Archives