Re: [rmcat] Proposed text on fairness for draft-ietf-rmcat-eval-criteria

[Colin Perkins <[e-mail]>]

Hi,

Thanks, Xiaoqing!  Along with the minutes, summarising the other decisions (https://datatracker.ietf.org/meeting/101/materials/minutes-101-rmcat-01 <https://datatracker.ietf.org/meeting/101/materials/minutes-101-rmcat-01>), it should now be possible to revise this draft. Jörg, do you have everything you need?

Cheers,
Colin




> On 26 Mar 2018, at 16:35, Xiaoqing Zhu (xiaoqzhu) <[e-mail]> wrote:
> 
> Hi,
> 
> This is to follow up on a discussion we had at the recent IETF-101 RMCAT WG, regarding how to address the notion of "fairness" in the eval-criteria draft.  
> 
> Brief recap for folks not at the meeting: There was consensus to remove the current open issue (#1) on using Jain's Fairness Index (JFI) as fairness metric since nobody used it in their tests. On the other hand, there were some debates/discussions regarding whether & how to address the issue of fairness in the eval-criteria draft.  I signed up to provide some updated text to cover this topic. 
> 
> As promised, please find below my proposed revision for #7 in Sec. 3, Metrics: 
> 
> -------------------------------------
> 7. Self-Fairness and Fairness with respect to cross traffic:  Experiments testing a given RMCAT proposal must report on relative ratios of the average throughput (measured at coarser time intervals) obtained by each RMCAT stream. In the presence of background cross-traffic such as TCP, the report must also include the relative ratio between average throughput of RMCAT streams and cross-traffic streams. 
> 
> During static periods of a test (i.e., when bottleneck bandwidth is constant and no arrival/departure of streams),  these report on relative ratios serve as an indicator of how fair the RMCAT streams share bandwidth amongst themselves and against cross-traffic streams. The throughput measurement interval can be set at a few values --- for example, at 1s, 5s, and 20s --- so as to measure fairness across different time scales. 
> 
> As a general guideline, the relative ratio between RMCAT flows with the same priority level and similar path RTT should be bounded between (0.333 and 3.) 
> --------------------------------------
> 
> The last part of this write-up spells a rough guide-rail on fairness. I've removed the other two criteria in the original text for the following reasons: 
> * On 1. Does not trigger the circuit breaker:  this is already stated in the intro and should always apply; 
> * On 3. RTT should not grow by a factor of 3 for the existing flows when a new flow is added: I understand this to be more a measure of low-latency instead of fairness.   Also I am not aware that any eval tests have used this in the past.  Am I missing something here? 
> 
> Additional feedback and input is welcome.  I'll leave it to the authors on how they want to incorporate this. 
> 
> Thanks,
> Xiaoqing


-- 
Colin Perkins
https://csperkins.org/