<- RFC Index (8501..8600)
RFC 8592
Independent Submission R. Browne
Request for Comments: 8592 A. Chilikin
Category: Informational Intel
ISSN: 2070-1721 T. Mizrahi
Huawei Network.IO Innovation Lab
May 2019
Key Performance Indicator (KPI) Stamping
for the Network Service Header (NSH)
Abstract
This document describes methods of carrying Key Performance
Indicators (KPIs) using the Network Service Header (NSH). These
methods may be used, for example, to monitor latency and QoS marking
to identify problems on some links or service functions.
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for informational purposes.
This is a contribution to the RFC Series, independently of any other
RFC stream. The RFC Editor has chosen to publish this document at
its discretion and makes no statement about its value for
implementation or deployment. Documents approved for publication by
the RFC Editor are not candidates for any level of Internet Standard;
see Section 2 of RFC 7841.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
https://www.rfc-editor.org/info/rfc8592.
Copyright Notice
Copyright (c) 2019 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.
Browne, et al. Informational [Page 1]
RFC 8592 KPI Timestamping May 2019
Table of Contents
1. Introduction ....................................................2
2. Terminology .....................................................3
2.1. Requirements Language ......................................3
2.2. Definition of Terms ........................................3
2.2.1. Terms Defined in This Document ......................4
2.3. Abbreviations ..............................................5
3. NSH KPI Stamping: An Overview ...................................6
3.1. Prerequisites ..............................................7
3.2. Operation ..................................................9
3.2.1. Flow Selection ......................................9
3.2.2. SCP Interface ......................................10
3.3. Performance Considerations ................................11
4. NSH KPI-Stamping Encapsulation .................................12
4.1. KPI-Stamping Extended Encapsulation .......................13
4.1.1. NSH Timestamping Encapsulation (Extended Mode) .....15
4.1.2. NSH QoS-Stamping Encapsulation (Extended Mode) .....17
4.2. KPI-Stamping Encapsulation (Detection Mode) ...............20
5. Hybrid Models ..................................................22
5.1. Targeted VNF Stamping .....................................23
6. Fragmentation Considerations ...................................23
7. Security Considerations ........................................24
8. IANA Considerations ............................................24
9. References .....................................................25
9.1. Normative References ......................................25
9.2. Informative References ....................................25
Acknowledgments ...................................................27
Contributors ......................................................27
Authors' Addresses ................................................27
1. Introduction
The Network Service Header (NSH), as defined by [RFC8300], specifies
a method for steering traffic among an ordered set of Service
Functions (SFs) using an extensible service header. This allows for
flexibility and programmability in the forwarding plane to invoke the
appropriate SFs for specific flows.
The NSH promises a compelling vista of operational flexibility.
However, many service providers are concerned about service and
configuration visibility. This concern increases when considering
that many service providers wish to run their networks seamlessly in
"hybrid mode", whereby they wish to mix physical and virtual SFs and
run services seamlessly between the two domains.
Browne, et al. Informational [Page 2]
RFC 8592 KPI Timestamping May 2019
This document describes generic methods to monitor and debug Service
Function Chains (SFCs) in terms of latency and QoS marking of the
flows within an SFC. These are referred to as "detection mode" and
"extended mode" and are explained in Section 4.
The methods described in this document are compliant with hybrid
architectures in which Virtual Network Functions (VNFs) and Physical
Network Functions (PNFs) are freely mixed in the SFC. These methods
also provide flexibility for monitoring the performance and
configuration of an entire chain or parts thereof as desired. These
methods are extensible to monitoring other Key Performance Indicators
(KPIs). Please refer to [RFC7665] for an architectural context for
this document.
The methods described in this document are not Operations,
Administration, and Maintenance (OAM) protocols such as [Y.1731]. As
such, they do not define new OAM packet types or operations. Rather,
they monitor the SFC's performance and configuration for subscriber
payloads and indicate subscriber QoE rather than out-of-band
infrastructure metrics. This document differs from [In-Situ-OAM] in
the sense that it is specifically tied to NSH operations and is not
generic in nature.
2. Terminology
2.1. Requirements Language
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
BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
2.2. Definition of Terms
This section presents the main terms used in this document. This
document also makes use of the terms defined in [RFC7665] and
[RFC8300].
Browne, et al. Informational [Page 3]
RFC 8592 KPI Timestamping May 2019
2.2.1. Terms Defined in This Document
First Stamping Node (FSN): The first node along an SFC that stamps
packets using KPI stamping. The FSN matches each packet with a
Stamping Controller (SC) flow based on (but not limited to) a
stamping classification criterion such as transport 5-tuple
coordinates.
Last Stamping Node (LSN): The last node along an SFC that stamps
packets using KPI stamping. From a forwarding point of view, the
LSN removes the NSH and forwards the raw IP packet to the next
hop. From a control-plane point of view, the LSN reads all the
metadata (MD) and exports it to a system performance statistics
agent or repository. The LSN should use the NSH Service Index
(SI) to indicate if an SF was at the end of the chain. The LSN
may change the Service Path Identifier (SPI) to a preconfigured
value so that the network underlay forwards the MD back directly
to the KPI database (KPIDB) based on this value.
Key Performance Indicator Database (KPIDB): Denotes the external
storage of MD for reporting, trend analysis, etc.
KPI stamping: The insertion of latency-related and/or QoS-related
information into a packet using NSH MD.
Flow ID: A unique 16-bit identifier written into the header by the
classifier. This allows 65536 flows to be concurrently stamped on
any given NSH service chain.
QoS stamping: The insertion of QoS-related information into a packet
using NSH MD.
Stamping Controller (SC): The central logic that decides what
packets to stamp and how to stamp them. The SC instructs the
classifier on how to build the parts of the NSH that are specific
to KPI stamping.
Stamping Control Plane (SCP): The control plane between the FSN and
the SC.
Browne, et al. Informational [Page 4]
RFC 8592 KPI Timestamping May 2019
2.3. Abbreviations
DEI Drop Eligible Indicator
DSCP Differentiated Services Code Point
FSN First Stamping Node
KPI Key Performance Indicator
KPIDB Key Performance Indicator Database
LSN Last Stamping Node
MD Metadata
NFV Network Function Virtualization
NSH Network Service Header
OAM Operations, Administration, and Maintenance
PCP Priority Code Point
PNF Physical Network Function
PNFN Physical Network Function Node
QoE Quality of Experience
QoS Quality of Service
RSP Rendered Service Path
SC Stamping Controller
SCL Service Classifier
SCP Stamping Control Plane
SF Service Function
SFC Service Function Chain
SI Service Index
SSI Stamp Service Index
Browne, et al. Informational [Page 5]
RFC 8592 KPI Timestamping May 2019
TS Timestamp
VLAN Virtual Local Area Network
VNF Virtual Network Function
3. NSH KPI Stamping: An Overview
A typical KPI-stamping architecture is presented in Figure 1.
Stamping
Controller
| KPIDB
| SCP Interface |
,---. ,---. ,---. ,---.
/ \ / \ / \ / \
( SCL )-------->( SF1 )--------->( SF2 )--------->( SFn )
\ FSN / \ / \ / \ LSN /
`---' `---' `---' `---'
Figure 1: Logical Roles in NSH KPI Stamping
The SC will be part of the SFC control-plane architecture, but it is
described separately in this document for clarity.
The SC is responsible for initiating start/stop stamp requests to the
SCL or FSN and also for distributing the NSH-stamping policy into the
service chain via the SCP interface.
The FSN will typically be part of the SCL but is called out as a
separate logical entity for clarity.
The FSN is responsible for marking NSH MD fields; this tells nodes in
the service chain how to behave in terms of stamping at the SF
ingress, the SF egress, or both, or ignoring the stamp NSH MD
completely.
The FSN also writes the Reference Time value, a (possibly inaccurate)
estimate of the current time of day, into the header, allowing the
"SPI:Flow ID" performance to be compared to previous samples for
offline analysis.
The FSN should return an error to the SC if not synchronized to the
current time of day and forward the packet along the service chain
unchanged. The code and format of the error are specific to the
protocol used between the FSN and SC; these considerations are out of
scope.
Browne, et al. Informational [Page 6]
RFC 8592 KPI Timestamping May 2019
SF1 and SF2 stamp the packets as dictated by the FSN and process the
payload as per normal.
Note 1: The exact location of the stamp creation may not be in the SF
itself and may be applied by a hardware device -- for
example, as discussed in Section 3.3.
Note 2: Special cases exist where some of the SFs are NSH unaware.
This is covered in Section 5.
The LSN should strip the entire NSH and forward the raw packet to the
IP next hop as per [RFC8300]. The LSN also exports NSH-stamping
information to the KPIDB for offline analysis; the LSN may export the
stamping information of either (1) all packets or (2) a subset based
on packet sampling.
In fully virtualized environments, the LSN is likely to be co-located
with the SF that decrements the NSH SI to zero. Corner cases exist
where this is not the case; see Section 5.
3.1. Prerequisites
Timestamping has its own set of prerequisites; however, these
prerequisites are not required for QoS stamping. In order to
guarantee MD accuracy, all servers hosting VNFs should be
synchronized from a centralized stable clock. As it is assumed that
PNFs do not timestamp (as this would involve a software change and a
probable impact on throughput performance), there is no need for them
to synchronize. There are two possible levels of synchronization:
Level A: Low-accuracy time-of-day synchronization, based on NTP
[RFC5905].
Level B: High-accuracy synchronization (typically on the order of
microseconds), based on [IEEE1588].
Each SF SHOULD have Level A synchronization and MAY have Level B
synchronization.
Level A requires each platform (including the SC) to synchronize its
system real-time clock to an NTP server. This is used to mark the MD
in the chain, using the Reference Time field in the NSH KPI stamp
header (Section 4.1). This timestamp is inserted into the NSH by the
first SF in the chain. NTP accuracy can vary by several milliseconds
between locations. This is not an issue, as the Reference Time is
merely being used as a time-of-day reference inserted into the KPIDB
for performance monitoring and MD retrieval.
Browne, et al. Informational [Page 7]
RFC 8592 KPI Timestamping May 2019
Level B synchronization requires each platform to be synchronized to
a Primary Reference Clock (PRC) using the Precision Time Protocol
(PTP) [IEEE1588]. A platform MAY also use Synchronous Ethernet
[G.8261] [G.8262] [G.8264], allowing more accurate frequency
synchronization.
If an SF is not synchronized at the moment of timestamping, it should
indicate its synchronization status in the NSH. This is described in
more detail in Section 4.
By synchronizing the network in this way, the timestamping operation
is independent of the current RSP. Indeed, the timestamp MD can
indicate where a chain has been moved due to a resource starvation
event as indicated in Figure 2, between VNF3 and VNF4 at time B.
Delay
| v
| v
| x
| x x = Reference Time A
| xv v = Reference Time B
| xv
| xv
|______|______|______|______|______|_____
VNF1 VNF2 VNF3 VNF4 VNF5
Figure 2: Flow Performance in a Service Chain
For QoS stamping, it is desired that the SCL or FSN be synchronized
in order to provide a Reference Time for offline analysis, but this
is not a hard requirement (they may be in holdover or free-run state,
for example). Other SFs in the service chain do not need to be
synchronized for QoS-stamping operations, as described below.
QoS stamping can be used to check the consistency of configuration
across the entire chain or parts thereof. By adding all potential
Layer 2 and Layer 3 QoS fields into a QoS sum at the SF ingress or
egress, this allows quick identification of QoS mismatches across
multiple Layer 2 / Layer 3 fields, which otherwise is a manual,
expert-led consuming process.
Browne, et al. Informational [Page 8]
RFC 8592 KPI Timestamping May 2019
|
|
| xy
| xy x = ingress QoS sum
| xv v = egress QoS sum
| xv y = egress QoS sum mismatch
| xv
|______|______|______|______|______|_____
SF1 SF2 SF3 SF4 SF5
Figure 3: Flow QoS Consistency in a Service Chain
Referring to Figure 3, x, v, and y are notional sum values of the QoS
marking configuration of the flow within a given chain. As the
encapsulation of the flow can change from hop to hop in terms of VLAN
header(s), MPLS labels, or DSCP(s), these values are used to compare
the consistency of configuration from, for example, payload DSCP
through overlay and underlay QoS settings in VLAN IEEE 802.1Q bits,
MPLS bits, and infrastructure DSCPs.
Figure 3 indicates that, at SF4 in the chain, the egress QoS marking
is inconsistent. That is, the ingress QoS settings do not match the
egress. The method described here will indicate which QoS field(s)
is inconsistent and whether this is ingress (where the underlay has
incorrectly marked and queued the packet) or egress (where the SF has
incorrectly marked and queued the packet.
Note that the SC must be aware of cases when an SF re-marks QoS
fields deliberately and thus does not flag an issue for desired
behavior.
3.2. Operation
KPI-stamping detection mode uses MD Type 2 as defined in [RFC8300].
This involves the SFC classifier stamping the flow at the chain
ingress and no subsequent stamps being applied; rather, each upstream
SF can compare its local condition with the ingress value and take
appropriate action. Therefore, detection mode is very efficient in
terms of header size that does not grow after the classification.
This is further explained in Section 4.2.
3.2.1. Flow Selection
The SC should maintain a list of flows within each service chain to
be monitored. This flow table should be in the format "SPI:Flow ID".
The SC should map these pairs to unique values presented as Flow IDs
per service chain within the NSH TLV specified in this document (see
Section 4). The SC should instruct the FSN to initiate timestamping
Browne, et al. Informational [Page 9]
RFC 8592 KPI Timestamping May 2019
on flow table match. The SC may also tell the classifier the
duration of the timestamping operation, by either the number of
packets in the flow or a certain time duration.
In this way, the system can monitor the performance of all en-route
traffic, an individual subscriber in a chain, or just a specific
application or QoS class that is used in the network.
The SC should write the list of monitored flows into the KPIDB for
correlation of performance and configuration data. Thus, when the
KPIDB receives data from the LSN, it understands to which flow the
data pertains.
The association of a source IP address with a subscriber identity is
outside the scope of this document and will vary by network
application. For example, the method of association of a source IP
address with an International Mobile Subscriber Identity (IMSI) will
be different from how a Customer Premises Equipment (CPE) entity with
a Network Address Translation (NAT) function may be chained in an
enterprise NFV application.
3.2.2. SCP Interface
An SCP interface is required between the SC and the FSN or
classifier. This interface is used to:
o Query the SFC classifier for a list of active chains and flows.
o Communicate which chains and flows to stamp. This can be a
specific "SPI:Flow ID" combination or can include wildcards for
monitoring subscribers across multiple chains or multiple flows
within one chain.
o Instruct how the stamp should be applied (ingress, egress, both
ingress and egress, or specific).
o Indicate when to stop stamping (after either a certain number of
packets or a certain time duration).
Typically, SCP timestamps flows for a certain duration for trend
analysis but only stamps one packet of each QoS class in a chain
periodically (perhaps once per day or after a network change).
Therefore, timestamping is generally applied to a much larger set of
packets than QoS stamping.
The exact specification of SCP is left for further study.
Browne, et al. Informational [Page 10]
RFC 8592 KPI Timestamping May 2019
3.3. Performance Considerations
This document does not mandate a specific stamping implementation
method; thus, NSH KPI stamping can be performed by either hardware
mechanisms or software.
If software-based stamping is used, applying and operating on the
stamps themselves incur an additional small delay in the service
chain. However, it can be assumed that these additional delays are
all relative for the flow in question. This is only pertinent for
timestamping mode, and not for QoS-stamping mode. Thus, whilst the
absolute timestamps may not be fully accurate for normal
non-timestamped traffic, they can be assumed to be relative.
It is assumed that the methods described in this document would only
operate on a small percentage of user flows.
The service provider may choose a flexible policy in the SC to
timestamp a selection of a user plane every minute -- for example, to
highlight any performance issues. Alternatively, the LSN may
selectively export a subset of the KPI stamps it receives, based on a
predefined sampling method. Of course, the SC can stress-test an
individual flow or chain should a deeper analysis be required. We
can expect that this type of deep analysis will have an impact on the
performance of the chain itself whilst under investigation. This
impact will be dependent on vendor implementations and is outside the
scope of this document.
For QoS stamping, the methods described here are even less intrusive,
as typically packets are only QoS stamped periodically (perhaps once
per day) to check service chain configuration per QoS class.
Browne, et al. Informational [Page 11]
RFC 8592 KPI Timestamping May 2019
4. NSH KPI-Stamping Encapsulation
KPI stamping uses NSH MD Type 0x2 for detection of anomalies and
extended mode for root-cause analysis of KPI violations. These are
further explained in this section.
The generic NSH MD Type 2 TLV for KPI stamping is shown below.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Ver|O|U| TTL | Length |U|U|U|U|Type=2 | Next Protocol |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Service Path Identifier | Service Index |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type |U| Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length KPI Metadata header and TLV(s) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 4: Generic NSH KPI Encapsulation
Relevant fields in the header that the FSN must implement are as
follows:
o The O bit must not be set.
o The MD type must be set to 0x2.
o The Metadata Class must be set to a value from the experimental
range 0xfff6 to 0xfffe according to an agreement by all parties to
the experiment.
o Unassigned bits: All fields marked "U" are unassigned and
available for future use [RFC8300].
o The Type field may have one of the following values; the content
of the Variable Length KPI Metadata header and TLV(s) field
depends on the Type value:
* Type = 0x01 (Det): Detection
* Type = 0x02 (TS): Timestamp Extended
* Type = 0x03 (QoS): QoS stamp Extended
The Type field determines the type of KPI-stamping format. The
supported formats are presented in the following subsections.
Browne, et al. Informational [Page 12]
RFC 8592 KPI Timestamping May 2019
4.1. KPI-Stamping Extended Encapsulation
The generic NSH MD Type 2 KPI-stamping header (extended mode) is
shown in Figure 5. This is the format for performance monitoring of
service chain issues with respect to QoS configuration and latency.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Ver|O|U| TTL | Length |U|U|U|U|Type=2 | Next Protocol |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Service Path Identifier | Service Index |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type |U| Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length KPI Configuration Header |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length KPI Value (LSN) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
\ \
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length KPI Value (FSN) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 5: Generic KPI Encapsulation (Extended Mode)
As mentioned above, two types are defined under the experimental MD
class to indicate the extended KPI MD: a timestamp type and a
QoS-stamp type.
The KPI Encapsulation Configuration Header format is shown below.
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|K|K|T|K|K|K|K|K| Stamping SI | Flow ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Reference Time |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 6: KPI Encapsulation Configuration Header
Browne, et al. Informational [Page 13]
RFC 8592 KPI Timestamping May 2019
The bits marked "K" are reserved for specific KPI type use and are
described in the subsections below.
The T bit should be set if Reference Time follows the KPI
Encapsulation Configuration Header.
The SSI (Stamping SI) contains the SI used for KPI stamping and is
described in the subsections below.
The Flow ID is a unique 16-bit identifier written into the header by
the classifier. This allows 65536 flows to be concurrently stamped
on any given NSH service chain (SPI). Flow IDs are not written by
subsequent SFs in the chain. The FSN may export monitored Flow IDs
to the KPIDB for correlation.
Reference Time is the wall clock of the FSN and may be used for
historical comparison of SC performance. If the FSN is not Level A
synchronized (see Section 3.1), it should inform the SC over the SCP
interface. The Reference Time is represented in 64-bit NTP format
[RFC5905], as presented in Figure 7:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Seconds |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Fraction |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 7: NTP 64-Bit Timestamp Format (RFC 5905)
Browne, et al. Informational [Page 14]
RFC 8592 KPI Timestamping May 2019
4.1.1. NSH Timestamping Encapsulation (Extended Mode)
The NSH timestamping extended encapsulation is shown below.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Ver|O|C|U|U|U|U|U|U| Length |U|U|U|U|Type=2 | NextProto |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Service Path ID | Service Index |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type=TS(2) |U| Len |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|I|E|T|U|U|U|SSI| Stamping SI | Flow ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| Reference Time (T bit is set) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|I|E|U|U|U| SYN | Stamping SI | Unassigned |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| Ingress Timestamp (I bit is set) (LSN) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Egress Timestamp (E bit is set) (LSN) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
. .
. .
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|I|E|U|U|U| SYN | Stamping SI | Unassigned |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| Ingress Timestamp (I bit is set) (FSN) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Egress Timestamp (E bit is set) (FSN) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 8: NSH Timestamp Encapsulation (Extended Mode)
The FSN KPI stamp MD starts with the Stamping Configuration Header.
This header contains the I, E, and T bits, and the SSI.
The I bit should be set if the Ingress stamp is requested.
The E bit should be set if the Egress stamp is requested.
Browne, et al. Informational [Page 15]
RFC 8592 KPI Timestamping May 2019
The SSI field must be set to one of the following values:
o 0x0: KPI stamp mode. No SI is specified in the Stamping SI field.
o 0x1: KPI stamp hybrid mode is selected. The Stamping SI field
contains the LSN SI. This is used when PNFs or NSH-unaware SFs
are used at the tail of the chain. If SSI=0x1, then the value in
the Type field informs the chain regarding which SF should act as
the LSN.
o 0x2: KPI stamp Specific mode is selected. The Stamping SI field
contains the targeted SI. In this case, the Stamping SI field
indicates which SF is to be stamped. Both Ingress stamps and
Egress stamps are performed when the SI=SSI in the chain. For
timestamping mode, the FSN will also apply the Reference Time and
Ingress Timestamp. This will indicate the delay along the entire
service chain to the targeted SF. This method may also be used as
a light implementation to monitor end-to-end service chain
performance whereby the targeted SF is the LSN. This is not
applicable to QoS-stamping mode.
Each stamping node adds stamp MD that consists of the Stamping
Reporting Header and timestamps.
The E bit should be set if the Egress stamp is reported.
The I bit should be set if the Ingress stamp is reported.
With respect to timestamping mode, the SYN bits are an indication of
the synchronization status of the node performing the timestamp and
must be set to one of the following values:
o In synch: 0x00
o In holdover: 0x01
o In free run: 0x02
o Out of synch: 0x03
If the platform hosting the SF is out of synch or in free run, no
timestamp is applied by the node, and the packet is processed
normally.
Browne, et al. Informational [Page 16]
RFC 8592 KPI Timestamping May 2019
If the FSN is out of synch or in free run, the timestamp request is
rejected and is not propagated through the chain. In such an event,
the FSN should inform the SC over the SCP interface. Similarly, if
the KPIDB receives timestamps that are out of order (i.e., a
timestamp of an "N+1" SF is prior to the timestamp of an "N" SF), it
should notify the SC of this condition over the SCP interface.
The outer SI value is copied into the stamp MD as the Stamping SI to
help cater to hybrid chains that are a mix of VNFs and PNFs or
through NSH-unaware SFs. Thus, if a flow transits through a PNF or
an NSH-unaware node, the delta in the inner SI between timestamps
will indicate this.
The Ingress Timestamp and Egress Timestamp are represented in 64-bit
NTP format. The corresponding bits (I and E) are reported in the
Stamping Reporting Header of the node's MD.
4.1.2. NSH QoS-Stamping Encapsulation (Extended Mode)
Packets have a variable QoS stack. For example, the same payload IP
can have a very different stack in the access part of the network
than the core. This is most apparent in mobile networks where, for
example, in an access circuit we would have an infrastructure IP
header (DSCP) composed of two layers -- one based on transport and
the other based on IPsec -- in addition to multiple MPLS and VLAN
tags. The same packet, as it leaves the Packet Data Network (PDN)
Gateway Gi egress interface, may be very much simplified in terms of
overhead and related QoS fields.
Because of this variability, we need to build extra meaning into the
QoS headers. They are not, for example, all PTP timestamps of a
fixed length, as in the case of timestamping; rather, they are of
variable lengths and types. Also, they can be changed on the
underlay at any time without the knowledge of the SFC system.
Therefore, each SF must be able to ascertain and record its ingress
and egress QoS configuration on the fly.
Browne, et al. Informational [Page 17]
RFC 8592 KPI Timestamping May 2019
The suggested QoS Type (QT) and lengths are listed below.
QoS Type Value Length Comment
----------------------------------------------------------
IVLAN 0x01 4 Bits Ingress VLAN (PCP + DEI)
EVLAN 0x02 4 Bits Egress VLAN
IQINQ 0x03 8 Bits Ingress QinQ (2x (PCP + DEI))
EQINQ 0x04 8 Bits Egress QinQ
IMPLS 0x05 3 Bits Ingress Label
EMPLS 0x06 3 Bits Egress Label
IMPLS 0x07 6 Bits Two Ingress Labels (2x EXP)
EMPLS 0x08 6 Bits Two Egress Labels
IDSCP 0x09 8 Bits Ingress DSCP
EDSCP 0x0A 8 Bits Egress DSCP
For stacked headers such as MPLS and 802.1ad, we extract the relevant
QoS data from the header and insert it into one QoS value in order to
be more efficient in terms of packet size. Thus, for MPLS, we
represent both experimental bits (EXP) fields in one QoS value, and
both 802.1p priority and drop precedence in one QoS value, as
indicated above.
For stack types not listed here (for example, three or more MPLS
tags), the SF would insert IMPLS/EMPLS several times, with each layer
in the stack indicating EXP QoS for that layer.
Browne, et al. Informational [Page 18]
RFC 8592 KPI Timestamping May 2019
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Ver|O|C|U|U|U|U|U|U| Length |U|U|U|U|Type=2 | NextProto=0x0 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Service Path ID | Service Index |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type=QoS(3) |U| Len |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|U|U|T|U|U|U|SSI| Stamping SI | Flow ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| Reference Time (T bit is set) |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|U|U|U|U|U|U|U|U| Stamping SI | Unassigned |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| QT | QoS Value |U|U|U|E| QT | QoS Value |U|U|U|E|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
. .
. .
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|U|U|U|U|U|U|U|U| Stamping SI | Unassigned |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-|
| QT | QoS Value |U|U|U|E| QT | QoS Value |U|U|U|E|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 9: NSH QoS Configuration Encapsulation (Extended Mode)
The encapsulation in Figure 9 is very similar to the encapsulation
detailed in Section 4.1.1, with the following exceptions:
o I and E bits are not required, as we wish to examine the full QoS
stack at the ingress and egress at every SF.
o SYN status bits are not required.
o The QT and QoS values are as outlined in the list above.
o The E bit at the tail of each QoS context field indicates if this
is the last egress QoS stamp for a given SF. This should coincide
with SI=0 at the LSN, whereby the packet is truncated, the NSH MD
is sent to the KPIDB, and the subscriber's raw IP packet is
forwarded to the underlay next hop.
Browne, et al. Informational [Page 19]
RFC 8592 KPI Timestamping May 2019
Note: It is possible to compress the frame structure to better
utilize the header, but this would come at the expense of crossing
byte boundaries. For ease of implementation, and so that
QoS stamping is applied on an extremely small subset of user-plane
traffic, we believe that the above structure is a pragmatic
compromise between header efficiency and ease of implementation.
4.2. KPI-Stamping Encapsulation (Detection Mode)
The format of the NSH MD Type 2 KPI-stamping TLV (detection mode) is
shown in Figure 10.
This TLV is used for KPI anomaly detection. Upon detecting a problem
or an anomaly, it will be possible to enable the use of KPI-stamping
extended encapsulations, which will provide more detailed analysis.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Ver|O|U| TTL | Length |U|U|U|U|Type=2 | Next Protocol |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Service Path Identifier | Service Index |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type=Det(1) |U| Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| KPI Type | Stamping SI | Flow ID |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Threshold KPI Value |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Ingress KPI stamp |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 10: Generic NSH KPI Encapsulation (Detection Mode)
The following fields are defined in the KPIDB MD:
o KPI Type: This field determines the type of KPI stamp that is
included in this MD. If a receiver along the path does not
understand the KPI type, it will pass the packet on transparently
and will not drop it. The supported values of KPI Type are:
* 0x0: Timestamp
* 0x1: QoS stamp
Browne, et al. Informational [Page 20]
RFC 8592 KPI Timestamping May 2019
o Threshold KPI Value: In the first header, the SFC classifier may
program a KPI threshold value. This is a value that, when
exceeded, requires the SF to insert the current SI value into the
SI field. The KPI type is the type of KPI stamp inserted into the
header as per Figure 10.
o Stamping SI: This is the Service Identifier of the SF when the
above threshold value is exceeded.
o Flow ID: The Flow ID is inserted into the header by the SFC
classifier in order to correlate flow data in the KPIDB for
offline analysis.
o Ingress KPI stamp: The last 8 octets are reserved for the
KPI stamp. This is the KPI value at the chain ingress at the SFC
classifier. Depending on the KPI type, the KPI stamp includes
either a timestamp or a QoS stamp. If the KPI type is Timestamp,
then the Ingress KPI stamp field contains a timestamp in 64-bit
NTP timestamp format. If the KPI type is QoS stamp, then the
format of the 64-bit Ingress KPI stamp is as follows.
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| QT | QoS Value | Unassigned |
+-+-+-+-+-+-+-+-+-+-+-+-+ +
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 11: QoS-Stamp Format (Detection Mode)
As an example operation, let's say we are using KPI type 0x01
(Timestamp). When an SF (say SFn) receives the packet, it can
compare the current local timestamp (it first checks that it is
synchronized to the network's PRC) with the chain Ingress Timestamp
to calculate the latency in the chain. If this value exceeds the
timestamp threshold, it then inserts its SI and returns the NSH to
the KPIDB. This effectively tells the system that at SFn the packet
violated the KPI threshold. Please refer to Figure 8 for the
timestamp format.
When this occurs, the SFC control-plane system would then invoke the
KPI extended mode, which uses a more sophisticated (and intrusive)
method to isolate the root cause of the KPI violation, as described
below.
Browne, et al. Informational [Page 21]
RFC 8592 KPI Timestamping May 2019
Note: Whilst detection mode is a valuable tool for latency actions,
the authors feel that building the logic into the KPI system for QoS
configuration is not justified. As QoS stamping is done infrequently
and on a tiny percentage of the user plane, it is more practical to
use extended mode only for service chain QoS verification.
5. Hybrid Models
A hybrid chain may be defined as a chain whereby there is a mix of
NSH-aware and NSH-unaware SFs.
Figure 12 shows an example of a hybrid chain with a PNF in the
middle.
Stamping
Controller
| KPIDB
| SCP Interface |
,---. ,---. ,---. ,---.
/ \ / \ / \ / \
( SCL )-------->( SF1 )--------->( SF2 )--------->( SFn )
\ FSN / \ / \ PNF1/ \ LSN /
`---' `---' `---' `---'
Figure 12: Hybrid Chain with PNF in Middle
In this example, the FSN begins its operation and sets the SI to 3.
SF1 decrements the SI to 2 and passes the packet to an SFC proxy
(not shown).
The SFC proxy strips the NSH and passes the packet to the PNF. On
receipt back from the PNF, the proxy decrements the SI and passes the
packet to the LSN with SI=1.
After the LSN processes the traffic, it knows from the SI value that
it is the last node in the chain, and it exports the entire NSH and
all MD to the KPIDB. The payload is forwarded to the next hop on the
underlay minus the NSH. The stamping information packet may be given
a new SPI to act as a homing tag to transport the stamp data back to
the KPIDB.
Browne, et al. Informational [Page 22]
RFC 8592 KPI Timestamping May 2019
Figure 13 shows an example of a hybrid chain with a PNF at the end.
Stamping
Controller
| KPIDB
| SCP Interface |
,---. ,---. ,---. ,---.
/ \ / \ / \ / \
( SCL )-------->( SF1 )--------->( SF2 )--------->( PNFN )
\ FSN / \ / \ LSN / \ /
`---' `---' `---' `---'
Figure 13: Hybrid Chain with PNF at End
In this example, the FSN begins its operation and sets the SI to 3.
The SSI field is set to 0x1, and the type is set to 1. Thus, when
SF2 receives the packet with SI=1, it understands that it is expected
to take on the role of the LSN, as it is the last NSH-aware node in
the chain.
5.1. Targeted VNF Stamping
For the majority of flows within the service chain, stamps (Ingress
stamps, Egress stamps, or both) will be carried out at each hop until
the SI decrements to zero and the NSH and stamp MD are exported to
the KPIDB. However, the need to just test a particular VNF may exist
(perhaps after a scale-out operation, software upgrade, or underlay
change, for example). In this case, the FSN should mark the NSH as
follows:
o The SSI field is set to 0x2.
o Type is set to the expected SI at the SF in question.
o When the outer SI is equal to the SSI, stamps are applied at the
SF ingress and egress, and the NSH and MD are exported to the
KPIDB.
6. Fragmentation Considerations
The methods described in this document do not support fragmentation.
The SC should return an error should a stamping request from an
external system exceed MTU limits and require fragmentation.
Depending on the length of the payload and the type of KPI stamp and
chain length, this will vary for each packet.
Browne, et al. Informational [Page 23]
RFC 8592 KPI Timestamping May 2019
In most service provider architectures, we would expect SI << 10,
which may include some PNFs in the chain that do not add overhead.
Thus, for typical Internet Mix (IMIX) packet sizes [RFC6985], we
expect to be able to perform timestamping on the vast majority of
flows without fragmentation. Thus, the classifier can apply a simple
rule that only allows KPI stamping on packet sizes less than 1200
bytes, for example.
7. Security Considerations
The security considerations for the NSH in general are discussed in
[RFC8300].
In-band timestamping, as defined in this document, can be used as a
means for network reconnaissance. By passively eavesdropping on
timestamped traffic, an attacker can gather information about network
delays and performance bottlenecks.
The NSH timestamp is intended to be used by various applications to
monitor network performance and to detect anomalies. Thus, a
man-in-the-middle attacker can maliciously modify timestamps in order
to attack applications that use the timestamp values. For example,
an attacker could manipulate the SFC classifier operation, such that
it forwards traffic through "better-behaved" chains. Furthermore, if
timestamping is performed on a fraction of the traffic, an attacker
can selectively induce synthetic delay only to timestamped packets
and can systematically trigger measurement errors.
Similarly, if an attacker can modify QoS stamps, erroneous values may
be imported into the KPIDB, resulting in further misconfiguration and
subscriber QoE impairment.
An attacker that gains access to the SCP can enable timestamping and
QoS stamping for all subscriber flows, thereby causing performance
bottlenecks, fragmentation, or outages.
As discussed in previous sections, NSH timestamping relies on an
underlying time synchronization protocol. Thus, by attacking the
time protocol, an attacker can potentially compromise the integrity
of the NSH timestamp. A detailed discussion about the threats
against time protocols and how to mitigate them is presented in
[RFC7384].
8. IANA Considerations
This document has no IANA actions.
Browne, et al. Informational [Page 24]
RFC 8592 KPI Timestamping May 2019
9. References
9.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC7665] Halpern, J., Ed. and C. Pignataro, Ed., "Service Function
Chaining (SFC) Architecture", RFC 7665,
DOI 10.17487/RFC7665, October 2015,
<https://www.rfc-editor.org/info/rfc7665>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in
RFC 2119 Key Words", BCP 14, RFC 8174,
DOI 10.17487/RFC8174, May 2017,
<https://www.rfc-editor.org/info/rfc8174>.
[RFC8300] Quinn, P., Ed., Elzur, U., Ed., and C. Pignataro, Ed.,
"Network Service Header (NSH)", RFC 8300,
DOI 10.17487/RFC8300, January 2018,
<https://www.rfc-editor.org/info/rfc8300>.
9.2. Informative References
[IEEE1588]
IEEE, "IEEE Standard for a Precision Clock Synchronization
Protocol for Networked Measurement and Control Systems",
IEEE Standard 1588,
<https://standards.ieee.org/standard/1588-2008.html>.
[RFC5905] Mills, D., Martin, J., Ed., Burbank, J., and W. Kasch,
"Network Time Protocol Version 4: Protocol and Algorithms
Specification", RFC 5905, DOI 10.17487/RFC5905, June 2010,
<https://www.rfc-editor.org/info/rfc5905>.
[RFC7384] Mizrahi, T., "Security Requirements of Time Protocols in
Packet Switched Networks", RFC 7384, DOI 10.17487/RFC7384,
October 2014, <https://www.rfc-editor.org/info/rfc7384>.
[RFC6985] Morton, A., "IMIX Genome: Specification of Variable Packet
Sizes for Additional Testing", RFC 6985,
DOI 10.17487/RFC6985, July 2013,
<https://www.rfc-editor.org/info/rfc6985>.
Browne, et al. Informational [Page 25]
RFC 8592 KPI Timestamping May 2019
[Y.1731] ITU-T Recommendation G.8013/Y.1731, "Operations,
administration and maintenance (OAM) functions and
mechanisms for Ethernet-based networks", August 2015,
<https://www.itu.int/rec/T-REC-G.8013/en>.
[G.8261] ITU-T Recommendation G.8261/Y.1361, "Timing and
synchronization aspects in packet networks", August 2013,
<https://www.itu.int/rec/T-REC-G.8261>.
[G.8262] ITU-T Recommendation G.8262/Y.1362, "Timing
characteristics of a synchronous Ethernet equipment slave
clock", November 2018,
<https://www.itu.int/rec/T-REC-G.8262>.
[G.8264] ITU-T Recommendation G.8264/Y.1364, "Distribution of
timing information through packet networks", August 2017,
<https://www.itu.int/rec/T-REC-G.8264>.
[In-Situ-OAM]
Brockners, F., Bhandari, S., Pignataro, C., Gredler, H.,
Leddy, J., Youell, S., Mizrahi, T., Mozes, D., Lapukhov,
P., Chang, R., Bernier, D., and J. Lemon, "Data Fields for
In-situ OAM", Work in Progress,
draft-ietf-ippm-ioam-data-05, March 2019.
Browne, et al. Informational [Page 26]
RFC 8592 KPI Timestamping May 2019
Acknowledgments
The authors gratefully acknowledge Mohamed Boucadair, Martin
Vigoureux, and Adrian Farrel for their thorough reviews and helpful
comments.
Contributors
This document originated as draft-browne-sfc-nsh-timestamp-00; the
following people were coauthors of that draft. We would like to
thank them and recognize them for their contributions.
Yoram Moses
Technion
Email: moses@ee.technion.ac.il
Brendan Ryan
Intel Corporation
Email: brendan.ryan@intel.com
Authors' Addresses
Rory Browne
Intel
Dromore House
Shannon
Co. Clare
Ireland
Email: rorybrowne@yahoo.com
Andrey Chilikin
Intel
Dromore House
Shannon
Co. Clare
Ireland
Email: andrey.chilikin@intel.com
Tal Mizrahi
Huawei Network.IO Innovation Lab
Israel
Email: tal.mizrahi.phd@gmail.com
Browne, et al. Informational [Page 27]