| Internet-Draft | PCPPS | September 2026 |
| Barth, et al. | Expires 13 March 2027 | [Page] |
This document introduces a Power Conserving Path Placement Strategy (PCPPS). During periods of low demand, PCPPS concentrates traffic onto a small set of network resources. This causes other network resources to become idle or nearly idle. When demand increases, PCPPS redistributes traffic as required.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 13 March 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This document introduces a Power Conserving Path Placement Strategy (PCPPS). During periods of low demand, PCPPS concentrates traffic onto a small set of network resources. This causes other network resources to become idle or nearly idle. When demand increases, PCPPS redistributes traffic as required.¶
PCPPS can be implemented in either a centralized or distributed traffic engineering model. This document focuses on distributed implementations.¶
An independent sleep management component is configured with power saving criteria. When it detects network resources that satisfy these criteria, it powers them down. When it receives a request from PCPPS, it powers them up again. The sleep management component is beyond the scope of this document.¶
Network operators can control the degree to which traffic is concentrated onto a small set of network resources. They can configure constraints that prevent traffic flows from being assigned to a path that does not satisfy their requirements. They can also configure the degree to which power conservation is prioritized in path placement.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
This document uses the following terms:¶
Path - An ordered set of links that connect a source node to a destination node. In a robust network, there are many paths from a particular source to a particular destination.¶
Traffic flow - A set of packets that have the same source and destination, and traverse the same path. Packets on an MPLS [RFC3031] Label Switched Path (LSP) are an example of a traffic flow.¶
Constraint - A rule that prevents a traffic flow from traversing some set of paths. For example, a constraint might prevent a particular traffic flow from traversing a path that contains low-speed links.¶
Optimization metric - The CSPF path placement algorithm uses the optimization metric to compute the best path between a source and a destination. An optimization metric is associated with the objective function [RFC4657] for which CSPF is optimizing (e.g., delay).¶
Sleep Status - An attribute of a hardware component. A component that is ASLEEP may consume power, but not enough to operate normally. A component that is AWAKE consumes enough power to operate normally. Sleep status is also an attribute of an IGP adjacency. See Sleeping Adjacency, below.¶
Power-sleep-capable - An interface attribute. If an interface is power-sleep-capable, it can be transitioned into the ASLEEP state. Otherwise, it cannot be transitioned into the ASLEEP state.¶
Power Savings Potential (PSP) - The PSP for a component is the difference between power consumption while AWAKE and power consumption while ASLEEP.¶
Sleeping Adjacency - An IGP adjacency whose endpoints are ASLEEP. A sleeping adjacency cannot pass traffic until it returns to the AWAKE state. It can be returned to the AWAKE state by waking its endpoints.¶
Sleeping bandwidth - An attribute of a LAG TE link that reflects lost capacity when LAG members are powered down.¶
PCPPS leverages Constraint-based Shortest Path Forwarding (CSPF). CSPF can be centralized or distributed. When it is centralized, a Path Computation Element (PCE) calculates a path for every flow in the network. When it is distributed, each node calculates a path for each flow that originates on it.¶
As stated in Section 3, many paths can connect a source node to a destination node. CSPF computes a path:¶
that does not violate any of the flow's constraints¶
whose links have sufficient unreserved bandwidth to support the flow¶
whose links have the lowest cumulative optimization metric¶
CSPF requires the following inputs:¶
Information regarding traffic flows (e.g., source, destination, required bandwidth, constraints)¶
The network topology (i.e., nodes, node attributes, links, and link attributes)¶
CSPF acquires this information from a Traffic Engineering Data Base (TED). Typically, an Interior Gateway Protocol (IGP) or the Border Gateway Protocol-Link State (BGP-LS) populates the TED.¶
As stated in Section 4, PCPPS leverages CSPF. When employing PCPPS the resulting TE path:¶
MUST NOT contain any sleeping adjacencies¶
MUST NOT violate constraints¶
MUST HAVE the lowest cumulative PCPPS metric.¶
The cumulative PCPPS metric is the sum of PCPPS metrics for each adjacency on the path. See Section 6 for details.¶
If CSPF cannot compute a path, it considers paths that include sleeping adjacencies. If such a path exists, it sends a request to the sleep management component. It then retries the path computation, using the newly awakened adjacency.¶
The following paragraphs describe an example algorithm that can be used to compute the PCPPS metric for an adjacency. Compliant implementations can use this algorithm or another algorithm to compute the PCPPS.¶
The PCPPS metric is the sum of the following:¶
The adjacency's optimization metric.¶
The cumulative PSP of each of the adjacency's endpoints (i.e., interfaces) adjusted by policy.¶
An interface's PSP is equal to 0 if any of the following conditions exist:¶
The interface is not power-sleep-capable¶
On point to point interfaces, if the interface on the other side of the adjacency is not power-sleep-capable¶
Otherwise, its cumulative PSP is equal to the interface's PSP plus the PSP of the interface's Parent Power Groups, recursively. Adjusting the cumulative PSP according to a policy produces an overall PCPPS metric that respects both power consumption and the optimization metric.¶
If two paths have identical cumulative PCPPS metrics, the tie can be broken based on a symmetric max-fill policy. This policy should cause traffic to consolidate rather than disperse.¶
To avoid re-use of links that are already idle prefer links with a utilization above a configured threshold.¶
Each TED interface entry includes a Power-Sleep Capability Bit. This bit determines whether the interface can be powered down when idle or nearly idle.¶
For interfaces on the local node, this bit is administratively assigned and advertised by an IGP. For interfaces on a remote node, this bit is learned from an IGP. See [I-D.many-lsr-power-group].¶
Each TED interface entry includes a PSP value, measured in milliwatts. This value represents the interface PSP. It does not include the PSPs of the Power Groups to which it is a member.¶
For interfaces on the local node, this value is administratively assigned or learned from hardware. It is advertised by an IGP. For interfaces on a remote node, this value is learned from an IGP. See [I-D.many-lsr-power-group].¶
Each TED interface entry includes zero or more references to a Power Group. See Section 8 for a description of Power Groups.¶
For Power Groups on the local node, this data is administratively assigned or learned from hardware. It is advertised by an IGP. For Power Groups on a remote node, this data is learned from an IGP. See [I-D.many-lsr-power-group].¶
If CSPF cannot compute a path, it considers paths that include sleeping adjacencies. If such a path exists, it sends a request to the sleep management component. It then retries the path computation, using the newly awakened adjacency.¶
The TED includes an entry that describes each sleeping adjacency. These entries are distinguishable from those that describe non-sleeping adjacencies.¶
Each TED interface entry includes a unidirectional sleeping bandwidth value, measured in bits per second. This value represents the sleeping bandwidth on a link. This is useful for IS-IS adjacencies formed over LAGs.¶
For interfaces on the local node, this value is administratively assigned or learned from hardware. It is advertised by an IGP. For interfaces on a remote node, this value is learned from an IGP. See [I-D.many-lsr-power-group].¶
Figure 1 depicts a line card (LC1). LC1 contains two forwarding engines (FE1 and FE2) and four interface complexes (INTCOMP1 through INTCOMP4). INTCOMP1 supports two interfaces (INT1 and INT2). Likewise, INTCOMP3 supports two interfaces (INT4 and INT5). INTCOMP2 and INTCOMP4 support one interface each (INT3 and INT6).¶
An interface complex includes PHY, MAC, encryption, gearbox, and other related circuitry. INTCOMP1 and INTCOMP3 also contain optics. INTCOMP2 and INTCOMP4 do not contain optics. Therefore, the interfaces that they support have their own optics.¶
INTCOMP1 and INTCOMP3 provide 400 Gbps of forwarding capacity each, while INTCOMP2 and INTCOMP4 provide 800 Gbps of forwarding capacity each.¶
Each hardware component has a PSP. LC1's PSP is 10 watts while FE1 and FE2 have PSPs of 300 watts. INTCOMP1 and INTCOMP3 have PSPs of 15 watts, while INTCOMP2 and INTCOMP4 have PSPs of 20 watts. INT3 and INT6 contain optics that have PSPs of 5 watts. INT1, INT2, INT4 and INT5 do not have separate optics. Therefore, they have a PSP of 0 watts.¶
INT1 and INT2 depend upon INTCOMP1. If INTCOMP1 fails, so do INT1 and INT2. Likewise, INT3 depends upon INTCOMP2. If INTCOMP2 fails, so does INT3.¶
INTCOMP1 and INTCOMP2 depend on FE1. If FE1 fails, so do INTCOMP1, INTCOMP2, INT1, INT2, and INT3. Likewise, INTCOMP3 and INTCOMP4 depend on FE2. If FE2 fails, so do INTCOMP3, INTCOMP4, INT4, INT5, and INT6.¶
FE1 and FE2 depend on LC1. If LC1 fails, so do all of the forwarding engines, interface complexes, and interfaces in the diagram.¶
In Figure 1, LC1, FE1, FE2, INTCOMP1, INTCOMP2, INTCOMP3, and INTCOMP4 are all Power Groups. Each Power Group, except for LC1, has exactly one parent. LC1 does not have a parent. Many Power Groups can have the same parent.¶
Each Power Group has one or more components and each component has a PSP. The PSP associated with a Power Group is equal to the sum of the PSPs associated with its components. A Power Group's PSP does not include the PSPs of its ancestors or its children.¶
The parent-child relationship reflects dependency. One Power Group is the child of another if any one of the child components depends upon any one of the parent components.¶
A network device's PSP characteristics can be described by any number of equivalent Power Group hierarchies. The paragraphs below demonstrate how two equivalent Power Group hierarchies can describe the PSP characteristics of the line card in Figure 1.¶
| Identifier | Parent | PSP | Hardware Components |
|---|---|---|---|
| 1 | None | 100 watts | LC1 |
| 2 | 1 | 300 watts | FE1 |
| 3 | 1 | 300 watts | FE2 |
| 4 | 2 | 15 watts | INTCOMP1 |
| 5 | 2 | 20 watts | INTCOMP2 |
| 6 | 3 | 15 watts | INTCOMP3 |
| 7 | 3 | 20 watts | INTCOMP4 |
Table 1 describes the PSP characteristics of the line card
in Figure 1 using a fine-grained Power Group hierarchy. We call it
fine-grained because each Power Group contains only one component.
Therefore, each Power
Group's PSP is equal to the PSP of its one and only component.¶
| Identifier | Parent | PSP | Hardware Components |
|---|---|---|---|
| 1 | None | 700 watts | LC1, FE1, FE2 |
| 2 | 1 | 15 watts | INTCOMP1 |
| 3 | 1 | 20 watts | INTCOMP2 |
| 4 | 1 | 15 watts | INTCOMP3 |
| 5 | 1 | 20 watts | INTCOMP4 |
Table 2 describes the PSP characteristics of the line card in Figure 1 using a less fine-grained Power Group hierarchy. We call it less fine-grained because Power Group 1 contains three components (LC1, FE1 and FE2). Its PSP is equal to the sum of the PSPs of LC1, FE1 and FE2 (i.e., 700 watts).¶
Section 8.4 describes how a network device's power-sleep capability determines which of the equivalent Power Group hierarchies it should advertise.¶
An interface is not part of a Power Group, even if it contains optics and consumes power. However, an interface can reference one or more Power Groups. When an interface references a Power Group, it MUST reference a Power Group that contains the hardware that supports it.¶
Interfaces that reference the same Power Group share common hardware dependencies. Therefore, CSPF may treat them as a group, either diverting traffic flows away from them all or routing traffic flows through them all.¶
A Link Aggregation Group (LAG) interface requires support from multiple interface complexes. Therefore a LAG interface references every Power Group that contains hardware that supports it.¶
A network device SHOULD advertise the least fine-grained Power Group hierarchy that can exercise its complete power-savings capability.¶
Assume that a network contains line cards that are power-sleep capable. Those line cards contain forwarding engines and interface complexes that are also power-sleep capable. This means that the line cards, forwarding engines and interface complexes can be powered on and off independently of the chassis.¶
In order to exercise its complete power savings capability, information regarding line card, forwarding engine and interface complex dependencies is required. Therefore, the line card must advertise the fine-grained Power Group hierarchy in Table 1.¶
Now assume that another network contains line cards that are power-sleep capable. Those line cards contain interface complexes that are also power-sleep capable. However, the forwarding engines are not power-sleep capable.¶
In order to exercise its complete power savings capability,
information regarding line card and interface complex
dependencies is required.
However, information regarding forwarding engine dependencies
is not required. Therefore, the line card could advertise
either the fine-grained Power Group hierarchy in Table 1 or the less
fine-grained Power Group hierarchy in Table 2.¶
Network operators must exercise care when configuring interfaces to be power-sleep capable.¶
The network must maintain sufficient bandwidth and redundancy, even when all power-sleep capable interfaces are ASLEEP.¶
Furthermore, interfaces that provide access to critical resources require special consideration. Critical resources include, but are not limited to, the following:¶
Putting an interface to sleep can cause protocol churn. For example, if an interface to the Designated Router (DR) on a multipoint interface is put to sleep, the backup DR becomes the DR.¶
This document requires no IANA actions.¶
Thanks to Les Ginsberg, Joel Halpern, Xiao Li, Zhang Li and Carlos Pignataro for their reviews and helpful comments.¶