Re: [Detnet] About the E2E latency of C-SCORE

Hi Jinoo,

I agree, a flow may allocate a service rate that is larger than its actual consumed bandwidth during the admission check, to achieve the worst-case per-hop latency that is smaller then the packet interval.

Regards,
PSF

Original

From: JinooJoung <jjoung@smu.ac.kr>
To: 彭少富10053815;
Cc: detnet@ietf.org <detnet@ietf.org>;
Date: 2024年03月25日 16:47
Subject: Re: [Detnet] About the E2E latency of C-SCORE

Hello Shaofu,

As I answered to you during the 119 meeting session,
the latencies can be differentiated by allocating different service rates (r) to the flows.

Best,
Jinoo

On Mon, Mar 25, 2024 at 11:13 AM <peng.shaofu@zte.com.cn> wrote:

Hi Jinoo,

Thanks for your confirmation.
We have reached the consensus for this point, i.e.,  worst-case per-hop latency is L/r.

Based on that, perhaps I can further understand worst-case per-hop latency as packet interval.
For example, considering a flow with TSpec: packet size L and sending a packet per time duration d.
It will consume the bandwidth: r = L/d. 
Thus, L/r = d, that is the flow's packet interval and the delay level treatment enjoyed by that flow. 
Here, d seems not be explicitly affected or determined by the network. In other words, it may be challenge for the network to explicitly provide different delay level treatment for different flow, e.g, n flows each with the same TSpec: packet size L and rate r, where n*r = C, and C is the link capacity. In this example, C-SCORE will provide the same worst-case latency L/r = n*L/C for each flow, but that treatment may not be the expected RSpec of the individual flow. 
However, C-SCORE may meet the case that application has not the requirement of customized E2E delay.  : )

Regards,
PSF

Original

From: JinooJoung <jjoung@smu.ac.kr>
To: 彭少富10053815;
Cc: detnet@ietf.org <detnet@ietf.org>;
Date: 2024年03月24日 09:41
Subject: Re: [Detnet] About the E2E latency of C-SCORE

_______________________________________________
detnet mailing list
detnet@ietf.org
https://www.ietf.org/mailman/listinfo/detnet

Hello Shaofu,
Yes, you are correct.
The SLi is the decisive factor in node i.

However,
SLi = Lmax_i/Ri + L/r.

And L/r is much larger than Lmax_i/Ri.
Thus, according to the definition of "per-hope dominant factor", which is the largest sum term,
the per-hop dominant factor is L/r.

Best,
Jinoo

On Fri, Mar 22, 2024 at 4:35 PM <peng.shaofu@zte.com.cn> wrote:

Hi Jinoo,

Thank you for patiently explaining the latency of C-SCORE during the meeting. 
I just check the E2E latency equation provided in draft-joung-detnet-stateless-fair-queuing-02, it is:
    Dh(p) <= (B-L)/r + SL0 + SL1 + ... + SLh,     (5)

Now I realize that we are talking about different things.
The PBOO (pay bursts only once) you mentioned is actually (B-L)/r, right ?
But I (and maybe there are other people who also pay more attention to the worst-case latency per-hop), am more concerned about SL0 + SL1 + ... + SLh, that is actually an evaluation formula based on the worst-case latency per-hop multiplied by the number of hops.

IMO, (B-L)/r is negligible, especially when the flow is policing on the network entrance node and B = L.
That is, the worst-case per-hop latency SLi on each node i is the dominator factor.

Please correct me if I have any misunderstanding.

Regards,
PSF