Enterprise Profile for the Precision Time Protocol With Mixed Multicast and Unicast Messages

Abstract

This document describes a profile for the use of the Precision Time Protocol in an IPV4 or IPv6 Enterprise information system environment. The profile uses the End to End Delay Measurement Mechanism, allows both multicast and unicast Delay Request and Delay Response Messages.¶

1. Introduction

The Precision Time Protocol ("PTP"), standardized in IEEE 1588, has been designed in its first version (IEEE 1588-2002) with the goal to minimize configuration on the participating nodes. Network communication was based solely on multicast messages, which unlike NTP did not require that a receiving node ("slave clock") in IEEE 1588-2008 [IEEE1588] needs to know the identity of the time sources in the network (the Master Clocks).¶

The "Best Master Clock Algorithm" (IEEE 1588-2008 [IEEE1588] Subclause 9.3), a mechanism that all participating PTP nodes must follow, set up strict rules for all members of a PTP domain to determine which node shall be the active sending time source (Master Clock). Although the multicast communication model has advantages in smaller networks, it complicated the application of PTP in larger networks, for example in environments like IP based telecommunication networks or financial data centers. It is considered inefficient that, even if the content of a message applies only to one receiver, it is forwarded by the underlying network (IP) to all nodes, requiring them to spend network bandwidth and other resources, such as CPU cycles, to drop the message.¶

The second revision of the standard (IEEE 1588-2008) is the current version (also known as PTPv2) and introduced the possibility to use unicast communication between the PTP nodes in order to overcome the limitation of using multicast messages for the bi-directional information exchange between PTP nodes. The unicast approach avoided that, in PTP domains with a lot of nodes, devices had to throw away more than 99% of the received multicast messages because they carried information for some other node. PTPv2 also introduced PTP profiles (IEEE 1588-2008 [IEEE1588] subclause 19.3). This construct allows organizations to specify selections of attribute values and optional features, simplifying the configuration of PTP nodes for a specific application. Instead of having to go through all possible parameters and configuration options and individually set them up, selecting a profile on a PTP node will set all the parameters that are specified in the profile to a defined value. If a PTP profile definition allows multiple values for a parameter, selection of the profile will set the profile-specific default value for this parameter. Parameters not allowing multiple values are set to the value defined in the PTP profile. Many PTP features and functions are optional, and a profile should also define which optional features of PTP are required, permitted, or prohibited. It is possible to extend the PTP standard with a PTP profile by using the TLV mechanism of PTP (see IEEE 1588-2008 [IEEE1588] subclause 13.4), defining an optional Best Master Clock Algorithm and a few other ways. PTP has its own management protocol (defined in IEEE 1588-2008 [IEEE1588] subclause 15.2) but allows a PTP profile specify an alternative management mechanism, for example SNMP.¶

2. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].¶

4. Problem Statement

This document describes a version of PTP intended to work in large enterprise networks. Such networks are deployed, for example, in financial corporations. It is becoming increasingly common in such networks to perform distributed time tagged measurements, such as one-way packet latencies and cumulative delays on software systems spread across multiple computers. Furthermore, there is often a desire to check the age of information time tagged by a different machine. To perform these measurements, it is necessary to deliver a common precise time to multiple devices on a network. Accuracy currently required in the Financial Industry range from 100 microseconds to 100 nanoseconds to the Grandmaster. This profile does not specify timing performance requirements, but such requirements explain why the needs cannot always be met by NTP, as commonly implemented. Such accuracy cannot usually be achieved with a traditional time transfer such as NTP, without adding non-standard customizations such as hardware time stamping, and on path support. These features are currently part of PTP, or are allowed by it. Because PTP has a complex range of features and options it is necessary to create a profile for enterprise networks to achieve interoperability between equipment manufactured by different vendors.¶

Although enterprise networks can be large, it is becoming increasingly common to deploy multicast protocols, even across multiple subnets. For this reason, it is desired to make use of multicast whenever the information going to many destinations is the same. It is also advantageous to send information which is unique to one device as a unicast message. The latter can be essential as the number of PTP slaves becomes hundreds or thousands.¶

PTP devices operating in these networks need to be robust. This includes the ability to ignore PTP messages which can be identified as improper, and to have redundant sources of time.¶

Interoperability among independent implementations of this PTP profile has been demonstrated at the ISPCS Plugfest ISPCS [ISPCS].¶

5. Network Technology

This PTP profile SHALL operate only in networks characterized by UDP RFC 768 [RFC0768] over either IPv4 RFC 791 [RFC0791] or IPv6 RFC 8200 [RFC8200], as described by Annexes D and E in IEEE 1588 [IEEE1588] respectively. If a network contains both IPv4 and IPv6, then they SHALL be treated as separate communication paths. Clocks which communicate using IPv4 can interact with clocks using IPv6 if there is an intermediary device which simultaneously communicates with both IP versions. A Boundary Clock might perform this function, for example. A PTP domain SHALL use either IPv4 or IPv6 over a communication path, but not both. The PTP system MAY include switches and routers. These devices MAY be Transparent Clocks, boundary Clocks, or neither, in any combination. PTP Clocks MAY be Preferred Masters, Ordinary Clocks, or Boundary Clocks. The Ordinary Clocks may be Slave Only Clocks, or be master capable.¶

Note that clocks SHOULD always be identified by their clock ID and not the IP or Layer 2 address. This is important in IPv6 networks since Transparent Clocks are required to change the source address of any packet which they alter. In IPv4 networks some clocks might be hidden behind a NAT, which hides their IP addresses from the rest of the network. Note also that the use of NATs may place limitations on the topology of PTP networks, depending on the port forwarding scheme employed. Details of implementing PTP with NATs are out of scope of this document.¶

PTP, like NTP, assumes that the one-way network delay for Sync Messages and Delay Response Messages are the same. When this is not true it can cause errors in the transfer of time from the Master to the Slave. It is up to the system integrator to design the network so that such effects do not prevent the PTP system from meeting the timing requirements. The details of network asymmetry are outside the scope of this document. See for example, ITU-T G.8271 [G8271].¶

6. Time Transfer and Delay Measurement

Master Clocks, Transparent Clocks and Boundary Clocks MAY be either one-step clocks or two-step clocks. Slave clocks MUST support both behaviors. The End to End Delay Measurement Method MUST be used.¶

Note that, in IP networks, Sync messages and Delay Request messages exchanged between a master and slave do not necessarily traverse the same physical path. Thus, wherever possible, the network SHOULD be traffic engineered so that the forward and reverse routes traverse the same physical path. Traffic engineering techniques for path consistency are out of scope of this document.¶

Sync messages MUST be sent as PTP event multicast messages (UDP port 319) to the PTP primary IP address. Two step clocks SHALL send Follow-up messages as PTP general messages (UDP port 320). Announce messages MUST be sent as multicast messages (UDP port 320) to the PTP primary address. The PTP primary IP address is 224.0.1.129 for IPv4 and FF0X:0:0:0:0:0:0:181 for Ipv6, where X can be a value between 0x0 and 0xF, see IEEE 1588 [IEEE1588] Annex E, Section E.3.¶

Delay Request Messages MAY be sent as either multicast or unicast PTP event messages. Master Clocks SHALL respond to multicast Delay Request messages with multicast Delay Response PTP general messages. Master Clocks SHALL respond to unicast Delay Request PTP event messages with unicast Delay Response PTP general messages. This allow for the use of Ordinary Clocks which do not support the Enterprise Profile, if they are slave Only Clocks.¶

Clocks SHOULD include support for multiple domains. The purpose is to support multiple simultaneous masters for redundancy. Leaf devices (non-forwarding devices) can use timing information from multiple masters by combining information from multiple instantiations of a PTP stack, each operating in a different domain. Redundant sources of timing can be ensembled, and/or compared to check for faulty Master Clocks. The use of multiple simultaneous masters will help mitigate faulty masters reporting as healthy, network delay asymmetry, and security problems. Security problems include man-in-the-middle attacks such as delay attacks, packet interception / manipulation attacks. Assuming the path to each master is different, failures malicious or otherwise would have to happen at more than one path simultaneously. Whenever feasible, the underlying network transport technology SHOULD be configured so that timing messages in different domains traverse different network paths.¶

7. Default Message Rates

The Sync, Announce and Delay Request default message rates SHALL each be once per second. The Sync and Delay Request message rates MAY be set to other values, but not less than once every 128 seconds, and not more than 128 messages per second. The Announce message rate SHALL NOT be changed from the default value. The Announce Receipt Timeout Interval SHALL be three Announce Intervals for Preferred Masters, and four Announce Intervals for all other masters.¶

The logMessageInterval carried in the unicast Delay Response message MAY be set to correspond to the master ports preferred message period, rather than 7F, which indicates message periods are to be negotiated. Note that negotiated message periods are not allowed, see forbidden PTP options (Section 13).¶

8. Requirements for Master Clocks

Master Clocks SHALL obey the standard Best Master Clock Algorithm from IEEE 1588 [IEEE1588]. PTP systems using this profile MAY support multiple simultaneous Grandmasters if each active Grandmaster is operating in a different PTP domain.¶

A port of a clock SHALL NOT be in the master state unless the clock has a current value for the number of UTC leap seconds.¶

If a unicast negotiation signaling message is received it SHALL be ignored.¶

9. Requirements for Slave Clocks

Slave clocks MUST be able to operate properly in a network which contains multiple Masters in multiple domains. Slaves SHOULD make use of information from the all Masters in their clock control subsystems. Slave Clocks MUST be able to operate properly in the presence of a Rogue Master. Slaves SHOULD NOT Synchronize to a Master which is not the Best Master in its domain. Slaves will continue to recognize a Best Master for the duration of the Announce Time Out Interval. Slaves MAY use an Acceptable Master Table. If a Master is not an Acceptable Master, then the Slave MUST NOT synchronize to it. Note that IEEE 1588-2008 requires slave clocks to support both two-step or one-step Master clocks. See IEEE 1588 [IEEE1588], subClause 11.2.¶

Since Announce messages are sent as multicast messages slaves can obtain the IP addresses of a master from the Announce messages. Note that the IP source addresses of Sync and Follow-up messages may have been replaced by the source addresses of a Transparent Clock, so, slaves MUST send Delay Request messages to the IP address in the Announce message. Sync and Follow-up messages can be correlated with the Announce message using the clock ID, which is never altered by Transparent Clocks in this profile.¶

10. Requirements for Transparent Clocks

Transparent Clocks SHALL NOT change the transmission mode of an Enterprise Profile PTP message. For example, a Transparent Clock SHALL NOT change a unicast message to a multicast message. Transparent Clocks SHOULD support multiple domains. Transparent Clocks which syntonize to the master clock will need to maintain separate clock rate offsets for each of the supported domains.¶

11. Requirements for Boundary Clocks

Boundary Clocks SHOULD support multiple simultaneous PTP domains. This will require them to maintain servo loops for each of the domains supported, at least in software. Boundary Clocks MUST NOT combine timing information from different domains.¶

12. Management and Signaling Messages

PTP Management messages MAY be used. Management messages intended for a specific clock, i.e. the IEEE 1588 [IEEE1588] defined attribute targetPortIdentity.clockIdentity is not set to All 1s, MUST be sent as a unicast message. Similarly, if any signaling messages are used they MUST also be sent as unicast messages whenever the message is intended for a specific clock.¶

14. Interoperation with IEEE 1588 Default Profile

Clocks operating in the Enterprise Profile will interoperate with clocks operating in the Default Profile described in IEEE 1588 [IEEE1588] Annex J.3. This variant of the Default Profile uses the End to End Delay Measurement Mechanism. In addition, the Default Profile would have to operate over IPv4 or IPv6 networks, and use management messages in unicast when those messages are directed at a specific clock. If either of these requirements are not met than Enterprise Profile clocks will not interoperate with Annex J.3 Default Profile Clocks. The Enterprise Profile will not interoperate with the Annex J.4 variant of the Default Profile which requires use of the Peer to Peer Delay Measurement Mechanism.¶

Enterprise Profile Clocks will interoperate with clocks operating in other profiles if the clocks in the other profiles obey the rules of the Enterprise Profile. These rules MUST NOT be changed to achieve interoperability with other profiles.¶

15. Profile Identification

The IEEE 1588 standard requires that all profiles provide the following identifying information.¶

          PTP Profile:
          Enterprise Profile
          Version: 1.0
          Profile identifier: 00-00-5E-00-01-00

          This profile was specified by the IETF

          A copy may be obtained at
          https://datatracker.ietf.org/wg/tictoc/documents

19. References