Re: [RPRWG] The upstream is more aggressive than the downstream
Winnie,
Please see my comments below.
Thanks.
Necdet
Shao, Winnie wrote:
>Hi all,
>
>Recently I simulate the fairness scenarios in annex I. The effect is a little interesting. Could you give me some comments or suggestion?
>
>Configuration:
>
> flow(s1->s3) 100% --------------------------------------------->
>
> flow(s2->s3) 100% ----------------->
>
> --> s1---------------------------> s2----------------------------> s3 --------------------->
>
> <----------------------SCFF messages (s1<-s2)
>
>
>unreservedRate=maxAllowedRate=shaperD=linkRate, reserved=0 stqLowThreshold = 1MTU stqMedThreshold = 2MTU stqHighThreshold = 3MTU rampUpCoef =6, rampDownCoef =6
>
>weight of s1 : weight of s2 = 2:1 (the output of aggressive mode and conservative mode is very close.)
>1. packet size =64 s1->s3:s2->s3 = 1.94 :1
>2. packet size = 512 s1->s3:s2->s3 = 2.08 :1
>3. packet size = 1024 s1->s3:s2->s3 = 2.13 :1
>
>weight of s1 : weight of s2 = 1:2 (the output of aggressive mode and conservative mode is very close.)
>1. packet size =64 s1->s3:s2->s3 = 1:1.26
>2. packet size = 512 s1->s3:s2->s3 ~= 1:1
>3. packet size = 1024 s1->s3:s2->s3 ~= 1:1
>
>unreservedRate=maxAllowedRate=80% linkRate reserved=20% linkRate shaperD=100%linkRate
>Other Configurable Variable is unchanged
>weight of s1 : weight of s2 = 1:2 (the output of aggressive mode)
>1. packet size =64 s1->s3:s2->s3 = 1: 1.9
>2. packet size = 512 s1->s3:s2->s3 = 1:1.65
>3. packet size = 1024 s1->s3:s2->s3 = 1: 1.6
>
>weight of s1 : weight of s2 = 1:2 conservative mode
>1. packet size =64 s1->s3:s2->s3 = 1.14:1
>2. packet size = 512 s1->s3:s2->s3 = 1.38:1
>3. packet size = 1024 s1->s3:s2->s3 = 1.2 :1
>
>The questions:
>1. Why upstream is so greed?
>
The reason that upstream station (S1) is getting as much bandwidth as
downstream station (S2) is that there was a division operation (that was
needed to fix the unfairness when the sum of the weights of upstream
stations is less than the weight of the local station) that we
eliminated. Table 9.6 row 12 condition of ...fwRate > addRate ... ,
needs to be changed to ...fwRate > addRate/localWeight ....to fix this.
There was a note about this in the draft that I could not find in D3.0,
but it was on page 202 line 22 on D2.2. Another reason is that your STQ
buffer is too small.
>2. Is there any other parameter which will impact the effect of FA?
>
Yes, more stations in the simulation and a larger buffer.
>3. How to decide ageCoef? -- Based on Table9.6 in D3.0, OC3--ageCoef=8, OC12-ageCoef=8, OC48-ageCoef=2, OC192-ageCoef=1/2
>
I could not find such a table in D3.0. The default value of ageCoef for
all rates is 4.
> It seems rateCoef * ageCoef always equal to 1, so normCoef =localWeight, is it right?
>
No, ageCoef can be independent of rateCoef. And rateCoef is fixed based
on the line rate.
>Thanks,
>Winnie Shao
>Senior Software Engineer
>IXA Development Center, Shenzhen Branch, PRC
>I-net: 8-754-1008
>
>
>
>
>
>