[Olsr-dev] [PATCH v1 0/3] Use bandwidth during gateway selection

[Teco Boot]

Op 27 jul. 2012, om 15:48 heeft Ferry Huberts het volgende geschreven:

> 
> 
> On 27-07-12 15:35, Teco Boot wrote:
>> Me, as dummy, get easily confused.
>>  - Factors should be multipliers.
>>  - Lower data rate links would have high influence, in a balanced way.
>> 
>> My proposal:
>>   Config option:                   Recommended value:   Default:
>>   ==============================   ==================   ========
>>   SmartGatewayWeightEtx            1.00                 1.00
>>   SmartGatewayWeightExitLinkUp     1.00                 0.00
>>   SmartGatewayWeightExitLinkDown   1.00                 0.00
>> 
>> And calculate such as:
>> 
>>   SmartGatewayWeightExitLinkUp
>>  ------------------------------
>> 
>>                         SmartGatewayWeightExitLinkUp
>>    costExitLinkUp = -----------------------------------
>>                     SmartGatewaySpeed(upstream) *  1000
> 
>  SmartGatewaySpeed(upstream) *  1000
> gives you bps
> 
> I think you want
>  SmartGatewaySpeed(upstream) / 1000
> right?

I didn't check your SmartGatewaySpeed.
It should be all Mbps or all kbps or all bps.


>> 
>>                         SmartGatewayWeightExitLinkDown
>>    costExitLinkDown = -----------------------------------
>>                       SmartGatewaySpeed(upstream) *  1000
>> 
> 
> same here
> 
>>    costEtx = SmartGatewayWeightEtx * PathEtxCost / 4
> 
> what's with the magic constant 4?

I assume, without any other knowledge, a WLAN ad hoc link.
With ETX=1, there is no loss at all and user throughput would be 4Mbps, or more, for a single connection on the medium.
If one knows more, (s)he should check / adjust with SmartGatewayWeightEtx.
This brings us to the ETX / LQ / LQ_multiplier discussion.

> how about
>  costEtx=(SmartGatewayWeightEtx/SmartGatewayDividerEtx)*PathEtxCost

Two weight config parameters for same purpose, same effect, inverse from each other.
What matters is background knowledge, when configuring. Your SmartGatewayDividerEtx is for experts only (recommended as 4). SmartGatewayWeightEtx is the simple to use weight, recommended as 4. I expect experts can use SmartGatewayWeightEtx for their need.


Let's wait on respond on who is using SmartGW. Take this to olsr-users?


Teco


> 
>> 
>> 
>>    gw_cost =  costExitLinkUp + costExitLinkDown + costEtx
>> 
>> 
>> Teco
>> 
>> 
>> Op 27 jul. 2012, om 14:06 heeft Ferry Huberts het volgende geschreven:
>> 
>>> 
>>> 
>>> On 27-07-12 13:59, Daniel Nitzpon wrote:
>>>> Hi!
>>>> 
>>>> Am 27.07.2012 13:21, schrieb Teco Boot:
>>>>> I also want to see comments. Maybe also:
>>>>>  - not interest in SmartGW at all
>>>>>  - interested, but not tried yet
>>>> 
>>>>> 
>>>>> Op 27 jul. 2012, om 12:50 heeft Henning Rogge het volgende geschreven:
>>>>> 
>>>>>> I would like to hear some comments about this from someone who has
>>>>>> USED the SmartGW system.
>>>> 
>>>> Haven't used it yet, sorry..
>>>> But still a comment:
>>>> 
>>>>>> On 07/26/2012 04:23 PM, Ferry Huberts wrote:
>>>>>>> - We calculate a new path cost of
>>>>>>> 
>>>>>>>     new_path_cost = (1 / (uplink + downlink)) + (tc_path_cost / 4)
>>>>>>> 
>>>>>>>   (uplink and downlink in Mbps)
>>>>>>>   (the 1 and 4 are weighing factors and we can make then configurable)
>>>> 
>>>> Freifunk experience here tells that in shared gateways via ADSL the
>>>> uplink is usually a much more limiting factor than the downlink. So I
>>>> would propose something along the lines of
>>>> 
>>>> new_path_cost = (1 / 2*(min (uplink*4, downlink))) + (tc_path_cost / 4)
>>>> 
>>>> (without having a real opinion on the weighting factors, just introduced
>>>> 1/2 to preserve the magnitude of the original suggestion)
>>>> 
>>> 
>>> Interesting point.
>>> I did think about this a few days ago, but wanted to propose the simple scenario first ;-)
>>> 
>>> How about scaling the combined bandwidth back with an asymmetry factor, something like:
>>> 
>>> 
>>> asym = (uplink > downlink) ? (downlink/uplink) : (uplink/downlink)
>>> new_path_cost = (1 / (asym * (uplink + downlink))) + (tc_path_cost / 4)
>>> 
>>>> Daniel
>>>> 
>>> 
>>> --
>>> Ferry Huberts
>>> 
>>> --
>>> Olsr-dev mailing list
>>> (spam-protected)
>>> https://lists.olsr.org/mailman/listinfo/olsr-dev
>> 
> 
> -- 
> Ferry Huberts