| Internet-Draft | PAM for Multi-SLO | August 2023 |
| Mirsky, et al. | Expires 23 February 2024 | [Page] |
This document defines a set of metrics for networking services with performance requirements expressed as Service Level Objectives (SLO). These metrics, referred to as Precision Availability Metrics (PAM), are useful for defining and monitoring SLOs. For example, PAM can be used by providers and/or customers of an IETF Network Slice Service to assess whether the service is provided in compliance with its defined SLOs.¶
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 23 February 2024.¶
Copyright (c) 2023 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.¶
Service providers and users often need to assess the quality with which network services are being delivered. In particular, in cases where service level guarantees are documented (including their companion metrology) as part of a contract established between the customer and the service provider, and Service Level Objectives (SLOs) are defined, it is essential to provide means to verify that what has been delivered complies with what has been possibly negotiated and (contractually) defined between the customer and the service provider. Examples of SLOs would be target values for the maximum packet delay (one-way and/or round-trip) or maximum packet loss ratio that would be deemed acceptable.¶
More generally, SLOs can be used to characterize the ability of a particular set of nodes to communicate according to certain measurable expectations. Those expectations can include but are not limited to aspects such as latency, delay variation, loss, capacity/throughput, ordering, and fragmentation. Whatever SLO parameters are chosen and whichever way service level parameters are being measured, precision availability metrics indicate whether or not a given service has been available according to expectations at all times.¶
Several metrics (often documented in the IANA Registry of Performance Metrics [IANA-PM-Registry] according to [RFC8911] and [RFC8912]), can be used to characterize the service quality, expressing the perceived quality of delivered networking services versus their SLOs. Of concern is not so much the absolute service level (for example, actual latency experienced) but whether the service is provided in compliance with the negotiated and eventually contracted service levels. For instance, this may include whether the experienced packet delay falls within an acceptable range that has been contracted for the service. The specific quality of service depends on the SLO or a set of thereof for a given service that is in effect. A non-compliance to an SLO might result in the degradation of the quality of experience for gamers or even jeopardize the safety of a large geographical area.¶
The same service level may be deemed acceptable for one application, while unacceptable for another, depending on the needs of the application. Hence it is not sufficient to measure service levels per se over time, but to assess the quality of the service being contextually provided (e.g., with the applicable SLO in mind). However, at this point, there are no standard metrics that can be used to account for the quality with which services are delivered relative to their SLOs, and whether their SLOs are being met at all times. Such metrics and the instrumentation to support them are essential for various purposes, including monitoring (to ensure that networking services are performing according to their objectives) as well as accounting (to maintain a record of service levels delivered, which is important for the monetization of such services as well as for the triaging of problems).¶
The current state-of-the-art of metrics includes, for example, interface metrics, useful to obtain statistical data on traffic volume and behavior that can be observed at an interface [RFC2863] and [RFC8343]. However, they are agnostic of actual service levels and not specific to distinct flows. Flow records [RFC7011] and [RFC7012] maintain statistics about flows, including flow volume and flow duration, but again, contain very little information about end-to-end service levels, let alone whether the service levels delivered meet their respective targets, i.e., their associated SLOs.¶
This specification introduces a new set of metrics, Precision Availability Metrics (PAM), aimed at capturing end-to-end service levels for a flow, specifically the degree to which flows comply with the SLOs that are in effect. PAM can be used to assess whether a service is provided in compliance with its defined SLOs. This information can be used in multiple ways, for example, to optimize service delivery, take timely counteractions in the event of service degradation, or account for the quality of services being delivered.¶
Availability is discussed in Section 3.4 of [RFC7297]. In this document, the term "availability" reflects that a service that is characterized by its SLOs is considered unavailable whenever those SLOs are violated, even if basic connectivity is still working. "Precision" refers to the fact that services whose end-to-end service levels are governed by SLOs, and which must therefore be precisely delivered according to the associated quality and performance requirements. It should be noted that precision refers to what is being assessed, not the mechanism used to measure it. In other words, it does not refer to the precision of the mechanism with which actual service levels are measured. Furthermore, the precision, with respect to the delivery of an SLO, particularly applies when a metric value approaches the specified threshold levels in the SLO.¶
The specification and implementation of methods that provide for accurate measurements is a separate topic independent of the definition of the metrics in which the results of such measurements would be expressed. Likewise, Service Level Expectations (SLEs), as defined in Section 4.1 of [I-D.ietf-teas-ietf-network-slices], are outside the scope of this document.¶
In this document, SLA and SLO are used as defined in [RFC3198]. The reader may refer to Section 4.1 of [I-D.ietf-teas-ietf-network-slices] for an applicability example of these concepts in the context of IETF Network Slice Services.¶
PAM Precision Availability Metric¶
OAM Operations, Administration, and Maintenance¶
SLA Service Level Agreement¶
SLE Service Level Expectations¶
SLO Service Level Objective¶
VI Violated Interval¶
VIR Violated Interval Ratio¶
VPC Violated Packets Count¶
SVI Severely Violated Interval¶
SVIR Severely Violated Interval Ratio¶
SVPC Severely Violated Packets Count¶
VFI Violation-Free Interval¶
When analyzing the availability metrics of a service between two measurement points, a time interval as the unit of PAM needs to be selected. In [ITU.G.826], a time interval of one second is used. That is reasonable, but some services may require different granularity (e.g., second or decamillisecond). For that reason, the time interval in PAM is viewed as a variable parameter though constant for a particular measurement session. Furthermore, for the purpose of PAM, each time interval is classified either as Violated Interval (VI), Severely Violated Interval (SVI), or Violation-Free Interval (VFI ). These are defined as follows:¶
The monitoring performance parameters to determine the quality of an interval is performed between the elements of the network that are referred to for the SLO corresponding to the performance parameter. Mechanisms of setting levels of a threshold of an SLO are outside the scope of this document.¶
From these definitions, a set of basic metrics can be defined that count the numbers of time intervals that fall into each category:¶
These count metrics are essential in calculating respective ratios (see Section 3.2) that can be used to assess the instability of a service.¶
Beyond accounting for violated intervals, it is sometimes beneficial to maintain counts of packets for which a performance threshold is violated. For example, this allows to distinguish between cases in which violated intervals are caused by isolated violation occurrences (such as, a sporadic issue that may be caused by a temporary spike in a queue depth along the packet's path) or by broad violations across multiple packets (such as a problem with slow route convergence across the network or more foundational issues such as insufficient network resources). Maintaining such counts and comparing them with the overall amount of traffic also facilitates assessing compliance with statistical SLOs (see Section 4). For these reasons, the following additional metrics are defined:¶
A set of metrics can be created based on PAM introduced in Section 3. In this document, these metrics are referred to as "derived PAM". Some of these metrics are modeled after Mean Time Between Failure (MTBF) metrics - a "failure" in this context referring to a failure to deliver a service according to its SLO.¶
An analogous set of metrics can be produced for SVI:¶
To indicate a historic degree of precision availability, additional derived PAMs can be defined as follows:¶
It might be useful for a service provider to determine the current condition of the service for which PAMs are maintained. To facilitate this, it is conceivable to complement PAM with a state model. Such a state model can be used to indicate whether a service is currently considered as available or unavailable depending on the network's recent ability to provide service without incurring intervals during which violations occur. It is conceivable to define such a state model in which transitions occur per some predefined PAM settings.¶
While the definition of a service state model is outside the scope of this document, the following section provides some considerations for how such a state model and accompanying configuration settings could be defined.¶
For example, a state model could be defined by a Finite State Machine featuring two states, "available" and "unavailable". The initial state could be "available". A service could subsequently be deemed as "unavailable" based on the number of successive interval violations that have been experienced up to the particular observation time moment. To return to a state of "available", a number of intervals without violations would need to be observed.¶
The number of successive intervals with violations, as well as the number of successive intervals that are free of violations, required for a state to transition to another state is defined by a configuration setting. Specifically, the following configuration parameters are defined:¶
Additional configuration parameters could be defined to account for the severity of violations. Likewise, it is conceivable to define configuration settings that also take VIR and SVIR into account.¶
It should be noted that certain SLAs may be statistical, requiring the service levels of packets in a flow to adhere to specific distributions. For example, an SLA might state that any given SLO applies to at least a certain percentage of packets, allowing for a certain level of, for example, packet loss and/or exceeding packet delay threshold to take place. Each such event, in that case, does not necessarily constitute an SLO violation. However, it is still useful to maintain those statistics, as the number of out-of-SLO packets still matters when looked at in proportion to the total number of packets.¶
Along that vein, an SLA might establish a multi-tiered SLO of, say, end-to-end latency (from the lowest to highest tier) as follows:¶
In that case, any individual packet with a latency greater than 20 ms latency and lower than 30 ms cannot be considered an SLO violation in itself, but compliance with the SLO may need to be assessed after the fact.¶
To support statistical SLOs more directly requires additional metrics, such as metrics that represent histograms for service level parameters with buckets corresponding to individual service level objectives. For the example just given, a histogram for a particular flow could be maintained with four buckets: one containing the count of packets within 20 ms, a second with a count of packets between 20 and 25 ms (or simply all within 25 ms), a third with a count of packets between 25 and 30 ms (or merely all packets within 30 ms, and a fourth with a count of anything beyond (or simply a total count). Of course, the number of buckets and the boundaries between those buckets should correspond to the needs of the SLA associated with the application, i.e., to the specific guarantees and SLOs that were provided. The definition of histogram metrics is outside the scope of this document but could be considered for future work along with issues discussed in Section 6.¶
PAM provides several benefits with other, more conventional performance metrics. Without PAM, it would be possible to conduct ongoing measurements of service levels and maintain a time-series of service level records, then assess compliance with specific SLOs after the fact. However, doing so would require the collection of vast amounts of data that would need to be generated, exported, transmitted, collected, and stored. In addition, extensive postprocessing would be required to compare that data against SLOs and analyze its compliance. Being able to perform these tasks at scale and in real-time would present significant additional challenges.¶
Adding PAM allows for a more compact expression of service level compliance. In that sense, PAM does not simply represent raw data but expresses actionable information. In conjunction with proper instrumentation, PAM can thus help avoid expensive postprocessing.¶
The following is a list of items that are outside the scope of this specification, but which will be useful extensions and opportunities for future work:¶
This document has no IANA actions.¶
Instrumentation for metrics that are used to assess compliance with SLOs constitute an attractive target for an attacker. By interfering with the maintenance of such metrics, services could be falsely identified as complying (when they are not) or vice-versa (i.e., flagged as being non-compliant when indeed they are). While this document does not specify how networks should be instrumented to maintain the identified metrics, such instrumentation needs to be adequately secured to ensure accurate measurements and prohibit tampering with metrics being kept.¶
Where metrics are being defined relative to an SLO, the configuration of those SLOs needs to be adequately secured. Likewise, where SLOs can be adjusted, the correlation between any metric instance and a particular SLO must be unambiguous. The same service levels that constitute SLO violations for one flow that should be maintained as part of the "violated time units" and related metrics, may be compliant for another flow. In cases when it is impossible to tie together SLOs and PAM, it will be preferable to merely maintain statistics about service levels delivered (for example, overall histograms of end-to-end latency) without assessing which constitutes violations.¶
By the same token, where the definition of what constitutes a "severe" or a "significant" violation depends on configuration settings or context. The configuration of such settings or context needs to be specially secured. Also, the configuration must be bound to the metrics being maintained. Thus, it will be clear which configuration setting was in effect when those metrics were being assessed. An attacker that can tamper with such configuration settings will render the corresponding metrics useless (in the best case) or misleading (in the worst case).¶
The authors greatly appreciate review and comments by Bjørn Ivar Teigen and Christian Jacquenet.¶