ForCES Implementation Experience Draft
draft-haleplidis-forces-implementation-experience-01

Abstract

The forwarding and Control Element Separation (ForCES) protocol defines a standard communication and control mechanism through which a Control Element (CE) can control the behavior of a Forwarding Element (FE). This document captures the experience of implementing the ForCES protocol and model. Its aim is to help others by providing examples and possible strategies for implementing the ForCES protocol.

Status of this Memo

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 March 5, 2011.

Copyright Notice

Copyright (c) 2010 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.

Table of Contents

1.  Terminology and Conventions
    1.1.  Requirements Language
2.  Introduction
    2.1.  Document Goal
    2.2.  Definitions
3.  ForCES Architecture
    3.1.  Pre-association setup - Initial Configuration
    3.2.  TML
    3.3.  Model
        3.3.1.  Components
        3.3.2.  LFBs
    3.4.  Protocol
        3.4.1.  TLVs
        3.4.2.  Message Deserialization
        3.4.3.  Message Serialization
4.  Developement Platforms
5.  Acknowledgements
6.  IANA Considerations
7.  Security Considerations
8.  References
    8.1.  Normative References
    8.2.  Informative References
§  Authors' Addresses

1.  Terminology and Conventions

The terminology that is used is the same as in the FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810] and is not copied in this document.

1.1.  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 [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

2.  Introduction

Forwarding and Control Element Separation (ForCES) defines an architectural framework and associated protocols to standardize information exchange between the control plane and the forwarding plane in a ForCES Network Element (ForCES NE). [RFC3654] (Khosravi, H. and T. Anderson, “Requirements for Separation of IP Control and Forwarding,” November 2003.) has defined the ForCES requirements, and [RFC3746] (Yang, L., Dantu, R., Anderson, T., and R. Gopal, “Forwarding and Control Element Separation (ForCES) Framework,” April 2004.) has defined the ForCES framework.

The ForCES protocol works in a master-slave mode in which FEs are slaves and CEs are masters. The protocol includes commands for transport of Logical Function Block (LFB) configuration information, association setup, status, and event notifications, etc. The reader is encouraged to read FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810] for further information.

The FE-MODEL (Halpern, J. and J. Hadi Salim, “Forwarding and Control Element Separation (ForCES) Forwarding Element Model,” March 2010.) [RFC5812] presents a formal way to define FE Logical Function Blocks (LFBs) using XML. LFB configuration components, capabilities, and associated events are defined when the LFB is formally created. The LFBs within the FE are accordingly controlled in a standardized way by the ForCES protocol.

The Transport Mapping Layer (TML) transports the protocol messages. The TML is where the issues of how to achieve transport level reliability, congestion control, multicast, ordering, etc. are handled. It is expected that more than one TML will be standardized. The various possible TMLs could vary their implementations based on the capabilities of underlying media and transport. However, since each TML is standardized, interoperability is guaranteed as long as both endpoints support the same TML. All ForCES Protocol Layer implementations MUST be portable across all TMLs. Although more than one TML may be standardized for the ForCES Protocol, all ForCES implementations MUST implement the SCTP-TML (Hadi Salim, J. and K. Ogawa, “SCTP-Based Transport Mapping Layer (TML) for the Forwarding and Control Element Separation (ForCES) Protocol,” March 2010.) [RFC5811].

The Applicability Statement (Crouch, A., Khosravi, H., Doria, A., Wang, X., and K. Ogawa, “ForCES Applicability Statement,” June 2010.) [I‑D.ietf‑forces‑applicability] captures the applicable areas in which ForCES can be used.

2.1.  Document Goal

This document captures the experience of implementing the ForCES protocol and model and its main goal is not to tell others how to implement, but to provide alternatives, ideas and proposals as how it can be implemented.

Also, this document mentions possible problems and potential choices that can be made, in an attempt to help implementors develop their own products.

Additionally this document takes into account that the reader has become familiar with the three main ForCES RFCs, the FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810], the FE-MODEL (Halpern, J. and J. Hadi Salim, “Forwarding and Control Element Separation (ForCES) Forwarding Element Model,” March 2010.) [RFC5812] and the SCTP-TML (Hadi Salim, J. and K. Ogawa, “SCTP-Based Transport Mapping Layer (TML) for the Forwarding and Control Element Separation (ForCES) Protocol,” March 2010.) [RFC5811].

2.2.  Definitions

This document follows the terminology defined by the ForCES Requirements in [RFC3654] (Khosravi, H. and T. Anderson, “Requirements for Separation of IP Control and Forwarding,” November 2003.) and by the ForCES framework in [RFC3746] (Yang, L., Dantu, R., Anderson, T., and R. Gopal, “Forwarding and Control Element Separation (ForCES) Framework,” April 2004.).

3.  ForCES Architecture

This section discusses the ForCES architecture, implementation challenges and how to overcome them.

3.1.  Pre-association setup - Initial Configuration

The initial configuration of the FE and the CE is done respectively by the FE Manager and the CE Manager. These entities have not as yet been standardized.

The simplest solution, are static configuration files, which play the role of the Managers and are read by FEs and CEs.

For more dynamic solutions however, it is expected that the Managers will be entities that will talk to each other and exchange details regarding the associations. Any developer can create any Manager, but they should at least be able to exchange the following details:

From the FE Manager side:

1. FEIDs
2. FE IPs, if the FEs and CEs will be communicating via network.
3. TML. The TML that will be used. If this is omitted, then SCTP MUST be chosen as default.
4. TML Priority ports. If this is omitted as well, then the CE MUST use the default values from the respective TML RFC.

From the CE Manager side:

1. CEIDs
2. CE IPs, if the FEs and CEs will be communicating via network.
3. TML. The TML that will be used. If this is omitted, then SCTP MUST be chosen as default.
4. TML Priority ports. If this is omitted as well, then the FE MUST use the default values from the respective TML RFC.

3.2.  TML

All ForCES implementations MUST support the SCTP as TML. Even if another TML will be chosen by the developer, SCTP is mandatory and MUST be supported.

There are several issues that should concern a developer for the TML.

1. Security. TML must be secure according to the respective RFC. For SCTP you have to use IPSec.
2. NAT issues. ForCES can be deployed everywhere and can run over SCTP/IP. In order for the FE and CE to work behind NATs you must ensure that the TML ports are forwarded and that the firewall allows SCTP through. This problem was identified during the first ForCES interoperability test and documented at the ForCES Implementation Report (Haleplidis, E., Ogawa, K., Wang, W., and J. Salim, “Implementation Report for ForCES,” June 2010.) [I‑D.ietf‑forces‑implementation‑report].

3.3.  Model

The model inherently is very dynamic. Using the basic atomic values that are specified, new datatypes can be built using atomic (single valued) and/or compound (structures and arrays).

The difficulty is to create something that is completely scalable so a develeper doesn't need to write the same code for new LFBs, or for new components etc. Just create code for the defined atomic values and then new components can be built based on already written code.

The model itself provides the key which is inheritance.

3.3.1.  Components

First, a basic component needs to be created as the mother of all the components with the basic parameters of all the components:

• The ID of the component.
• The access rights of that component.
• If it is an optional component.
• If it is of variable size.
• Minimum data size.
• Maximum data size.

If the data size of the component is not variable, then the size is either the minimum or the maximum size, as both should have the same value.

Next, some basic functions are in order:

• A common constructor.
• A common deconstructor.
• Retrieve Component ID.
• Retrieve access right property.
• Query if it is an optional component.
• Get Full Data.
• Set Full Data.
• Get Sparse Data.
• Set Sparse Data.
• Del Full Data.
• Del Sparse Data.
• Get Property
• Set Property
• Get Value.
• Set Value.
• Del Value.
• Get Data.
• Clone component.

The Get/Set/Del Full/Sparse Data and Get/Set Property functions handle the respective ForCES commands and return the respective TLV, for example the Set Full Data should return a Result TLV. The Get/Set/Del Value are called from the Get/Set/Del Full/Sparse Data respectively and provide the interface to the actual values in the hardware, separating the forces handling logic from the interface to the actual values.

The Get Data function should return the value of the data only, not in TLV format.

The last function seems out of place. That function MUST return a new component that has the exact same values and attributes. This function is useful in array components as described further.

The only requirement is to implement the base atomic data types. Any new atomic datatype can be built as a child of a base data type which will inherit all the functions and if necessary override them.

The struct component can then be built. A struct component is a component by itself, but consists of a number of atomic components. These atomic components create a static array within the struct. The ID of each atomic component is the array's index. The Clone function, for a struct component, MUST create and return an exact copy of the struct component with the same static array.

The most difficult component to be built is the array. The difficulty lie in the actual benefit of the model. You have absolute freedom over what you build. An array is an array of components. In all rows you have the exact same type of component either a single component or a struct. The struct can have multiple single components, or a combination of single components, structs and arrays and so on. So, the difficulty lies in how to create a new row, a new component by itself. This is where the Clone function is very useful. For the array a mother component that can spawn new components exactly like itself is needed. Once a set command is received, the mother component can spawn a new component, if the targeted row does not exists, and add it into the array, and with the set fulldata the value is set in the recently spawned component, as the spawned component knows how the data is created. In order to distinguish these spawned components from each other and their functionality, some kind of index is required that will also reflect on how the actual data of the specific component is stored on the hardware.

Once the basic constructors of all possible components are created, then a developer only has to create his LFB components or datatypes as a child of one of the already created components and the only thing the developer really needs to add, is the three functions of Get/Set/Del value of each component which is platform dependent. The rest stays the same.

3.3.2.  LFBs

The same architecture in the components can be used for the LFBs, allowing a developer to write LFB handling code only once. The parent LFB has some basic attributes:

• The LFB Class ID.
• The LFB Instance ID.
• An Array of Components.
• An Array of Capabilities.
• An Array of Events.

Then some common functions:

• Handle Configuration Command.
• Handle Query Command.
• Get Class ID.
• Get Instance ID.

Once these are created each LFB can inherit all these from the parent and the only thing it has to do is to add the components that have already been created.

An example can be seen in Figure 1 (Example code for creating part of FEProtocolLFB). The following code creates a part of FEProtocolLFB:

//FEID
cui = new Component_uInt(FEPO_FEID, ACCESS_READ_ONLY, FE_id);
Components[cui->get_ComponentId()]=cui; //Add component to array list

//Current FEHB Policy Value
cub = new Component_uByte(FEPO_FEHBPolicy, ACCESS_READ_WRITE, 0);
Components[cub->get_ComponentId()]=cub; //Add component to array list

//FEIDs for BackupCEs Array
cui = new Component_uInt(0, ACCESS_READ_WRITE, 0);
ca = new Component_Array(FEPO_BackupCEs, ACCESS_READ_WRITE);
ca->AddRow(cui, 1);
ca->AddMotherComponent(cui);
Components[ca->get_ComponentId()]=ca; //Add component to array list

The same concept can be applied to handling LFBs as one FE. An FE is a collection of LFBs. Thus all LFBs can be stored in an array based on the LFB's class id, version and instance. Then what is required is an LFBHandler that will handle the array of the LFBs. A specific LFB, for example, can be addressed using the following scheme:

LFBs[ClassID][Version][InstanceID].

Note: While an array can be used in components, capabilities and events, a hash table or a similar concept is better suited for storing LFBs using the component ID as the hash key with linked lists for collision handling, as the created array can have large gaps if the values of LFB Class ID vary greatly.

3.4.  Protocol

3.4.1.  TLVs

The goal, for protocol handling, is to create a general and scalable architecture that handles all protocol messages instead of something implementation specific. There are certain difficulties that have to be overcome first.

Since the model allows a developer to define any LFB required, the protocol has been thus created to cope with, allowing the user the freedom to configure and query any component whatever the underlying model. While this being a strong point for the protocol itself, one difficulty lies with the unkwown underlying model and the unlimited number of types of messages that can be created, making creating something generic a daunting task.

Another difficulty also arises from the batching capabilities of the protocol. You can have multiple Operations within a message, you can select more than one LFB to command, and more than one component to manipulate.

A possible solution is again provided by inheritance. There are two basic components in a protocol message.

1. The common header.
2. The rest of the packet.

The rest of the packet is divided in Type-Length-Value (TLV) packets, and in one case Index-Length-Value (ILV) packets.

The TLV hierarchy can be seen in the Figure 2 (ForCES TLV Hierrachy):

         Common Header
               |
   +-----------+----------+-----------+
   |           |          |           |
Redirect   LFBSelect   ASResult   ASTreason
  TLV         TLV        TLV         TLV
               |
               |
           Operation
              TLV
               |
               |            Optional
            PathData  ---> KeyInfo TLV
              TLV
               |
    +----------+---------+----------+
    |          |         |          |
SparseData   Result   FullData   PathData
   TLV        TLV       TLV        TLV

The above figure shows only the basic hierarchy level of TLVs and does not show batching. Also this figure does not show the recursion that can occur at the last level of the hierarchy. The figure shows one kind of recursion with PathData within PathData. FullData can be within FullData and SparseData. The possible combination of TLVs are described in detail in the FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810] as well as the data packing rules.

A TLV's main attributes are:

• Type
• Length
• Data
• An array of TLVs.

The array of TLVs is the next hierarchy level of TLVs nested in this TLV.

A TLVs common function could be:

• A basic constructor.
• A constructor using data from the wire.
• Add a new TLV for next level.
• Get the next TLV of next level.
• Get a specific TLV of next level.
• Replace a TLV of next level.
• Get the Data.
• Get the Length.
• Set the Data.
• Set the Length.
• Set the Type.
• Serialize the header.
• Serialize the TLV to be written on the wire.

Next all TLVs can inherit all these functions and attributes and either override them or create some new functions where required.

3.4.2.  Message Deserialization

What follows is a the algorithm for deserializing any protocol message:

1. Get the message header.
2. Read the length.
3. Check the message type to understand what kind of packet this is.
4. If the length is larger than the message header then there is data for this packet.
5. A check can be made here regarding the message type and the length of the packet

If the packet is a Query or Config type then for this level there are LFBSelector TLVs:

1. Read the next 2 shorts(type-length). If the type is an LFBSelector then the message is valid.
2. Read the necessary length for this LFBSelector and create the LFBSelector from the data of the wire.
3. Add this LFBSelector to the mainheader array of LFBSelectors
4. Do this until the rest of the packet has finished.

The next level of TLVs are Operation TLVs

1. Read the next 2 shorts(type-length). If the type is an OperationTLV then the message is valid.
2. Read the necessary length for this OperationTLV and create the OperationTLV from the data of the wire.
3. Add this OperationTLV to the LFBSelector array of TLVs.
4. Do this until the rest of the LFBSelector's Packet has finished.

1. Read the next 2 shorts(type-length). If the type is a PathData then the message is valid.
2. Read the necessary length for this PathDataTLV and create the PathDataTLV from the data of the wire.
3. Add this PathData TLV to the Operation TLV's array of TLVs.
4. Do this until the rest of the OperationTLV's Packet has finished.

Here it gets interesting, as the next level of PathDataTLVs can be either:

• PathData TLVs.
• FullData TLV.
• SparseData TLV.
• Result TLV.

The solution to this difficulty is recursion. If the next TLV is PathDataTLV then the PathDataTLV that is created uses the same kind of deserialisation until it reaches a FullDataTLV or SparseDataTLV. There can be only one FullDataTLV or SparseData within a PathData.

1. Read the next 2 shorts(type-length).
2. If the Type is a PathDataTLV then do again the previous algorithm but add the PathDataTLV to this PathDataTLV's array of TLVs.
3. Do this until the rest of the PathData's Packet has finished.
4. If the Type is a FullDataTLV then create the FullData TLV from the packet and add this to the PathData's array of TLVs.
5. If the Type is a SparseDataTLV then create the SparseData TLV from the packet and add this to the PathData's array of TLVs.
6. If the Type is a ResultTLV then create the Result TLV from the packet and add this to the PathData's array of TLVs.

If the message is a Query it MUST NOT have any kind of data inside the PathData.

If the message is a Query Response then it MUST either have a ResultTLV or a FullData TLV.

If the message is a Config it MUST have inside either a FullDataTLV or a SparseData TLV.

If the message is a Config Reponse, it MUST have inside a ResultTLV.

More details regarding message validation can be be read in Section 7 of the FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810].

Note: When deserializing, implementors must take care to ignore padding of TLVs as all must be 32-bit aligned. The length value in TLVs includes the Type and Length (4 bytes) but does not include padding.

3.4.3.  Message Serialization

The same concept can be applied in the message creation process. Having the TLVs ready, a developer can go bottom up. All that is required is the serialization function that will transform the TLV into bytes ready to be transfered on the network.

For example for the creation of a simple query from the CE to the FE, all the PathData are created. Then they will be serialized and inserted into an Operation TLV, which in turn will be serialized and inserted into an LFB Selector and in turn serialized and entered into the Common Header which will be passed to the TML to be transported to the FE.

Having an array of TLVs inside a TLV that are next in the TLV hierarchy, allows the developer to insert any number of next level TLVs thus creating any kind of message.

Note: When the TLV is serialized to be written on the wire, implementors must take care to include padding to TLVs as all must be 32-bit aligned.

4.  Developement Platforms

Any Developmens platform that can support the SCTP TML and the TML of the developer's choosing is available for use.

Figure 3 (SCTP Support on Operating Systems) provides an initial survey of SCTP support for C/C++ and Java at present time.

/-------------+-------------+-------------+-------------\
|\ Platform   |             |             |             |
| ----------\ |   Windows   |    Linux    |   Solaris   |
|  Language  \|             |             |             |
+-------------+-------------+-------------+-------------+
|             |             |             |             |
|    C/C++    |  Supported  |  Supported  |  Supported  |
|             |             |             |             |
+-------------+-------------+-------------+-------------+
|             |   Limited   |             |             |
|    Java     | Third Party |  Supported  |  Supported  |
|             | Not from SUN|             |             |
\-------------+-------------+-------------+-------------/

A developer should be aware of some limitations regarding Java implementations.

Java inherently does not support unsigned types. A workaround this can be found in the creation of classes that do the translation of unsigned to java types. The problem is that the unsigned long cannot be used as it is in the Java platform. The proposed set of classes can be found in (, “Classes that support unsigned primitive types for Java. All except the unsigned long,” .) [Java Unsigned Types].

5.  Acknowledgements

The authors would like to thank Adrian Farrel for sponsoring this document and Jamal Hadi Salim for discussions to make this document better.

6.  IANA Considerations

This memo includes no request to IANA.

7.  Security Considerations

Developers of ForCES FEs and CEs should take the security considerations of the ForCES framework in [RFC3746] (Yang, L., Dantu, R., Anderson, T., and R. Gopal, “Forwarding and Control Element Separation (ForCES) Framework,” April 2004.) and FE-protocol (Doria, A., Hadi Salim, J., Haas, R., Khosravi, H., Wang, W., Dong, L., Gopal, R., and J. Halpern, “Forwarding and Control Element Separation (ForCES) Protocol Specification,” March 2010.) [RFC5810] into account.

Also, as specified in the security considerations section of the SCTP-TML RFC (Hadi Salim, J. and K. Ogawa, “SCTP-Based Transport Mapping Layer (TML) for the Forwarding and Control Element Separation (ForCES) Protocol,” March 2010.) [RFC5811] the transport-level security, has to be ensured by IPSec.

8.  References

8.1. Normative References

8.2. Informative References

Authors' Addresses