Re: [xrblock] Burst Discard Count

-----邮件原件-----
发件人: Varun Singh [mailto:vsingh.ietf@gmail.com] 
发送时间: 2015年2月27日 19:36
收件人: Qin Wu
抄送: xrblock@ietf.org
主题: Re: [xrblock] Burst Discard Count

Hi Qin,

On Fri, Feb 27, 2015 at 4:05 AM, Qin Wu <bill.wu@huawei.com> wrote:
> The argument is correct, but when the burst count for loss and burst count for discard are corresponding to the same reported source, the value for them are probably same.

To confirm, the burstCount indicates the number of intervals where bursts occur, and not the packets in an individual bursts.

I think the implicit assumption is that losses and discards are correlated, and that discards may occur before losses. However, consider a congestion control algorithm that is sensitive to discards may correct before any losses occur. Then the burstCount measuring only losses would not increase  and hence burst discards count will not correspond to burst loss count.

[Qin]: Somebody questioned the usefulness of burst gap discard report. Therefore Combined loss/discard report was proposed to address this issue.
http://www.ietf.org/mail-archive/web/xrblock/current/msg00172.html
http://www.ietf.org/mail-archive/web/xrblock/current/msg00264.html

> Also burst gap discard is intended to be sent together with burst gap loss.
> See the section 1 of RFC7003:
> "
> This block is intended to be used in conjunction with [RFC7002], which 
> provides the total packets discarded and on which this block therefore depends.
> However, the metric in [RFC7002] may be used independently of the 
> metrics in this block.
> "
>
> -Qin
> -----邮件原件-----
> 发件人: xrblock [mailto:xrblock-bounces@ietf.org] 代表 Varun Singh
> 发送时间: 2015年2月26日 19:34
> 收件人: xrblock@ietf.org
> 主题: [xrblock] Burst Discard Count
>
> Hi all,
>
> While preparing the -01 version of the 
> draft-ietf-xrblock-rtcweb-rtcp-xr-metrics, I noticed that the
> RFC6958 - Burst/Gap Loss reports a "burst count" metric, which indicates the number of bursts  in an reporting interval (cumulative or arbitrary time interval). However, RFC7003 does not report a corresponding burst count metric for discarded packets.
>
> I forget if this was discussed before, but it appears to be an oversight, as arguments for including burst count for loss also apply to burst discards -- unless I am missing something or it is defined elsewhere. How do we fix this?
>
> This may be non-blocking for the rtcweb metrics as we could instead use the burstLossRate and burstDiscardRate which are defined in the summary statistics in RFC7004.
>
> Thanks and Regards,
> Varun
>
>
> --
> http://www.callstats.io
>
> _______________________________________________
> xrblock mailing list
> xrblock@ietf.org
> https://www.ietf.org/mailman/listinfo/xrblock

--
http://www.callstats.io