Network Working Group | S. Vinapamula |
Internet-Draft | Juniper Networks |
Intended status: Standards Track | S. Sivakumar |
Expires: April 2, 2015 | Cisco Systems |
M. Boucadair | |
France Telecom | |
T. Reddy | |
Cisco | |
September 29, 2014 |
Application-Initiated Flow High Availability Awareness through PCP
draft-vinapamula-flow-ha-04
This document specifies a mechanism for a host to signal via PCP which connections should be protected against network failures. These connections will be elected to be subject to high availability mechanisms enabled at the network side.
This approach assumes that aplications/users have more visibility about sensitive connections rather than any heuristic that can be enabled at the network side to guess which connections should be secured.
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].
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 http://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 April 2, 2015.
Copyright (c) 2014 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 (http://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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Internet service continuity is critical in Service Providers' environment. To achieve this, most Service Providers deploy active-backup systems. This not only helps them in service continuity during failover, but also help in service continuity hitless upgrade or minimal hit upgrades of both software or hardware and achieve desired level of service continuity compliance.
For some of the network functions, a state would be maintained for every connection for processing subsequent packets of that connection. For service continuity of those connections on backup when active fail, that corresponding state had to be check-pointed on the backup. NAPT is one such network function, where a state is maintained for every connection.
Heuristic based on the protocol, mapping lifetime, etc are used in the network side to elect which connections are elected to High Availability (HA) means. This document advocates for an application-initiated approach that would allow applications/user to signal to the network which of their connections are critical.
PCP-initiated signalling is superior to heuristics deployed at the network side.
This document specifies how PCP can be extended to signal which connection should be subject to HA mechanism. This document does not make any assumption on the PCP-controlled device that will make use of the content of signals issued by PCP clients. These devices are likely to be flow-aware.
The proposed approach is aligned with the current networking trends advocating for open network APIs to interact with applications/services. Policy-decision making process at the network side will be enriched with information signaled by application using PCP for instance.
In a high availability (HA) deployment, it is expensive in terms of memory, CPU and other resources to checkpoint all connections state. Also check-pointing may not be required for all connections as all connections may not be critical. But, this leaves a challenge to identify what connections to checkpoint.
Typically, this is addressed by identifying long lived connections and check-pointing state of only those connections that lived long enough, to the backup for service continuity.
However, following are the issues with that approach:
This proposal is based on the assumption that an application or user is the best judge to decide which of its connections' are critical.
An application/user may indicate the desire for checkpoint through PCP client, using the CHECKPOINT_REQUIRED option as described in Figure 1.
The entry to be backuped is indicated by the content of a MAP or PEER message.
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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |Option Code=TBA| Reserved | Option Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Option Name: CHECKPOINT_REQUIRED Number: <TBA> Purpose: Indicate if an entry needs to be check pointed. Valid for Opcodes: MAP, PEER Length: 0. May appear in: request, response. Maximum occurrences: 1.
Figure 1: CHECKPOINT_REQUIRED PCP Option
The description of the fields is as follows:
It was tempting to include additional fields in the option but this would lead to a more complex design that is not justified, e.g.,:
An application or user can use this option to indicate that one or more of its connections are critical and disruption is not desired. Doing so will trigger check-pointing of state to the backup.
Communication between application/user and PCP client is implementation-specific.
Support for the CHECKPOINT_REQUIRED option by PCP servers and PCP clients is optional. This option (Code TBA; see Figure 1) MAY be included in a PCP MAP/PEER request to indicate a connection is to be protected against network failures.
The PCP client includes a CHECKPOINT_REQUIRED option in a MAP or PEER request to signal that the corresponding mapping is to be protected.
A PCP server MAY ignore the CHECKPOINT_REQUIRED option sent to it by a PCP client (e.g., if it does not support the option or if it is configured to ignore it). To signal that it has not accepted the option, a PCP server simply does not include the CHECKPOINT_REQUIRED option in the response. If the PCP client does not receive a CHECKPOINT_REQUIRED option in a response to a request enclosing a CHECKPOINT_REQUIRED option, this means the PCP server does not support the option or it is configured to ignore it.
If the CHECKPOINT_REQUIRED option is not included in the PCP client request, the PCP server does not include the CHECKPOINT_REQUIRED option in the associated response. This is mainly because there are not valid motivation that would justify a PCP server notify a PCP client about it reliability decision.
When the PCP server receives a CHECKPOINT_REQUIRED option, the PCP server checks if it can honor this request depending on whether resources are available for check-pointing. If there are no resources available for check-pointing, but there are resources available to honor the MAP/PEER request, a response is sent back to the PCP client without including the CHECKPOINT_REQUIRED option (i.e., the request is processed as any MAP/PEER request that does not convey a CHECKPOINT_REQUIRED option). If check-pointing resources are still available and the quota for this PCP client is not reached, the PCP server tags the corresponding entry as eligible to HA mechanism and sends back the CHECKPOINT_REQUIRED option in the positive answer to the PCP client.
To update the check-pointing behavior of a mapping maintained by the PCP server, the PCP client generates a PCP MAP/PEER renewal request that includes a CHECKPOINT_REQUIRED option to indicate this mapping has to be check-pointed or without including a CHECKPOINT_REQUIRED option to indicate this mapping need not be check-pointed anymore. Upon receipt of the PCP request, the PCP server proceeds to the same operations to validate a MAP/PEER request refreshing an existing mapping. If validation checks are successfully passed, the PCP server updates the check-point flag associated with that mapping accordingly (i.e., it is set if a CHECKPOINT_REQUIRED option was included in the refresh request or it is cleared if no CHECKPOINT_REQUIRED option was included) , and the PCP server returns the response to the PCP client accordingly.
What information to checkpoint and how to checkpoint is out of scope of this document, and is left for implementations. Also, interest to indicate check-pointing by users/applications in a PCP request, may be automatic, semi-automatic, or human intervened. This behavior is also left for application implementations.
It is RECOMMENDED to checkpoint state on backup for honored requests before a response is sent to the PCP client.
Below are provided some examples for illustration purposes:
In conjunction with NAT, other network functions that may maintain state for each connection such as stateful firewall may register to PCP server, and may be triggered for check-pointing respective state of that connection.
PCP-related security considerations are discussed in [RFC6887].
CHECKPOINT_REQUIRED option can be used by an attacker to identify critical flows. This issue is mitigated if the network on which the PCP messages are to be sent is fully trusted. Means to defend against attackers who can intercept packets between the PCP server and the PCP client should be enabled. In some deployments, access control lists (ACLs) can be installed on the PCP client, PCP server, and the network between them, so those ACLs allow only communications between trusted PCP elements. If the networking environment between the PCP client and PCP server is not secure, means to protect exposing the content of PCP messages (e.g., DTLS [RFC6347]) are recommended.
A network device can always override the end-user signalling, i.e., what is signaled by the PCP client, if the instructions are conflicting with the network policies.
There is a risk that every PCP client may wish to checkpoint every connection, which can potentially load the system. Administration SHOULD restrict the number of connections that can be elected to be backuped and the rate of check-pointing on per PCP client.
The following PCP Option Code is to be allocated in the optional-to- process range (the registry is maintained in http://www.iana.org/ assignments/pcp-parameters):
Thanks to Reinaldo Penno, Stuart Shechire, and Dave Thaler for their comments.
[RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
[RFC6887] | Wing, D., Cheshire, S., Boucadair, M., Penno, R. and P. Selkirk, "Port Control Protocol (PCP)", RFC 6887, April 2013. |
[RFC6347] | Rescorla, E. and N. Modadugu, "Datagram Transport Layer Security Version 1.2", RFC 6347, January 2012. |