PPV - Per Packet Value

What is the PPV?

PPV Simulator: go

The proposed framework

Defines resource sharing policies for all situations by throughput-Packet Value functions
Proposes a Packet Marker at the edge of the network, which marks a Packet Value on each packet, based on the throughput-PV function of that flow.
Within the network, Resource Nodes only need maximize the total transmitted Packet Value. This maximization results in implementing the policies, without the need for any flow awareness.

Per Packet Value Concept

The Packet Value represents the gain of the operator when the packet is delivered. It expresses the relative importance of one packet to another. [Value/bit]
The aim of the network, and every Resource Node within is to maximize the total Packet Value of the transmitted packets.
The Congestion Threshold Value separates the packets with PVs that get transmitted from the ones that get dropped (at a Resource Node)
- resulting from the combination of available capacity, amount of offered traffic and the Packet Value composition of the offered traffic
- increases as the congestion increases
The framework complements E2E Congestion Control
- PPV enforces fairness, only low controlled loss has to be provided by E2E Congestion Control
- Even incompatible CCs can coexist in a network implementing PPV

Packet Marker

Requirement on packet marking:
- if all packets below a Congestion Threshold Value are dropped
- then the throughput of the remaining packets shall be as defined by the throughput-PV function at this threshold
An example implementation (per flow)
- Quantize the throughput-PV functions
- Associate a token bucket to each quantized region (PV as defined, incoming token rate is the throughput of this region, size is token rate * typical buffer size of Resource Nodes)
- When a packet of a flow arrives, select the token bucket with the highest PV, where there are enough tokens (packet size)

Resource Node

Task is to maximize the total transmitted Packet Value
The simplest algorithm is to always serve packets with highest PV first
- Non-practical for typical flows as it results in out of order delivery
A simple FIFO implementation is possible,
- when the queue becomes full drop the packet with the smallest Packet Value first
If different packets have different resource demands that can be taken to account when maximizing the total transmitted Packet Value
- Packet Values cannot be directly compared, but must be normalized by the cost of their transmission (r)
- We define the Effective Packet Value of a packet as its Packet V alue divided by its transmission cost (r)

Videos

	Laki, S., Gombos, G., Nádas, S., & Turányi, Z. (2017, July). "Take your own share of the PIE", In Proceedings of the Applied Networking Research Workshop (pp. 27-32). ACM. (video presentation)
	Sándor Laki, Gergő Gombos, Péter Hudoba, Szilveszter Nádas, Zoltán Kiss, Gergely Pongrácz, Csaba Keszei: "Scalable Per Subscriber QoS with Core-Stateless Scheduling", (2018), ACM SIGCOMM 2018 Industrial Demos, (pdf, demo video)
	Ferenc Fejes, Sándor Laki, Gergő Gombos, Szilveszter Nádas, Zoltán Kiss: "Decoupling Delay and Resource Sharing Targets with Efficient Core-Stateless AQM", (2019), ACM SIGCOMM 2019 Demos, (pdf)
	Szilveszter Nádas,Gergő Gombos, Ferenc Fejes, Sándor Laki: "Towards Core-Stateless Fairness on Multiple Timescales", (2019), ANRW (pdf)
	Ferenc Fejes, Gergő Gombos, Sándor Laki, Szilveszter Nádas: "On the Incompatibility of Scalable Congestion Controls over the Internet", (2020), FIT
	Szilveszter Nádas, Gergő Gombos, Ferenc Fejes, Sándor Laki: "A Congestion Control Independent L4S Scheduler", (2020), ANRW
	Szilveszter Nádas, Balázs Varga, Ferenc Fejes, Gergő Gombos, Sándor Laki: "On Congestion Control (un)fairness", (2020), IETF109 ICCRG
	Ferenc Fejes, Szilveszter Nádas, Gergő Gombos, Sándor Laki: "A Core-Stateless L4S Scheduler for P4-enabled hardware switches with emulated HQoS", (2021), INFOCOM Demo 2021 (Best Demo Award)
	Dávid Kis, Gergő Gombos, Sándor Laki, Szilveszter Nádas: "Resource Sharing Beyond FQ: 35K Users at 100Gbps", (2022), SIGCOMM Demo 2022

News
2021.05

May 2021 - Our demo paper entitled "A Core-Stateless L4S Scheduler for P4-enabled Hardware Switches with Emulated HQoS” received the INFOCOM 2021 Best Demo Award.
2020.05

Our paper on A Congestion Control Independent L4S Scheduler has been accepted for publication at the ACM/IRTF Applied Networking Research Workshop 2020 (ANRW’20) that will be held together with the IETF-108 meeting.
2020.05

Our paper On the incompatibility of scalable congestion controls over the Internet has been accepted for publication at the 2nd Workshop on the Future of Internet Transport held in co-location with IFIP Networking 2020.
2019.11

Our paper entitled "Who will Save the Internet from the Congestion Control Revolution?" has been accepted for publication at the Workshop on Buffer Sizing, Stanford University, CA, USA. The workshop program seems very exciting.
2019.08

Our demo (Decoupling Delay and Resource Sharing Targets with Efficient Core-Stateless AQM) was presented at ACM SIGCOMM 2019, Beijing, China.
2019.06

Our paper on Towards Core-Stateless Fairness on Multiple Timescales has been accepted for publication at ACM/IRTF ANRW 2019.
2018.07

The IETF-102 meeting has just started with more than 1000 participants in Montreal, QC, Canada. Our paper is being presented at ANRW'18, a satelite conference of the IETF meeting tomorrow. (presentation)
2018.06

Our industrial demo ("Scalable Per Subscriber QoS with Core-Stateless Scheduling") submitted together with Ericsson has been accepted for presentation at SIGCOMM 2018.

The Team

Szilveszter Nádas

Ericsson Research, Hungary