RFC 3127
Authentication, Authorization, and Accounting: Protocol Evaluation
Pages: 84
Informational
Informational
Part 2 of 3 – Pages 21 to 55
Appendix C - Position Briefs
C.1 SNMP PRO Evaluation
Evaluation of SNMP AAA Requirements PRO Evaluation Evaluator - Stuart Barkley Ref [1] is "Comparison of SNMPv3 Against AAA Network Access Requirements", aka 'the document' Ref [2] is the aaa eval criteria as modified by us, aka 'the requirements' The document uses T to indicate total compliance, P to indicate partial compliance and F to indicate no compliance. For each section I've indicated my grade for the section. If there is a change, I've indicated that and the grade given by the authors. 1 Per item discussion 1.1 General Requirements 1.1.1 Scalability - Grade T The document indicates that SNMP can adequately handle that scale from the requirements document. Since most current uses are ppp connections and SNMP is already capable of handling the interface table and other per session tables it is clear that basic capacity exists. Additions to support other tables and variables scales in a simple linear fashion with the number of additional variables and protocol interactions. Regardless of the final selected protocol handling the scaling required is not a trivial undertaking. SNMP can draw upon existing network management practices to assist in this implementation. 1.1.2 Fail-over - Grade T SNMP is of vital importance to the operation of most networks. Existing infrastructures can handle required failover or other redundant operations. 1.1.3 Mutual Authentication - Grade T The use of shared secrets described in the document is a well understood method of integrity control. Although shared secrets don't necessarily provide full authentication since other parties may also have the same secrets, the level of authentication is sufficient for the task at hand. In many cases the SNMP infrastructure will
already exist and shared secrets should already be properly managed on an operational network. A failure of the SNMP shared secret approach regardless of the AAA protocol will likely leave equipment and systems open to substantial misuse bypassing any more elaborate AAA authentication. 1.1.4 Transmission Level Security - Grade T SNMPv3 provides many additional security options which were not available or were more controversial in previous SNMP versions. 1.1.5 Data Object Confidentiality - New Grade P (from T) The document discusses SNMPv3 which can provide data confidentially for data passing over the wire. There is substantial implied AAA architecture (brokers and proxies) in the requirements that full conformance is difficult to determine. In particular, the evaluator has difficulty with the concept of "the target AAA entity for whom the data is ultimately destined", but will concede that the desired requirement is only partially met (most especially with the transfer of a PAP password). 1.1.6 Data Object Integrity - New Grade T (from P) SNMP has full capabilities that allow the authentication of the data. Brokers, proxies or other intermediaries in the data chain can verify the source of the information and determine that the data has not been tampered with. The document downgrades the grade to P because of confusion over the integrity checking role of intermediaries. 1.1.7 Certificate Transport - Grade T The requirements require the capability of transporting certificates but do not have any specific use for the certificates. The requirements make assumptions that the protocol selected will be dependent upon certificates, but this is not necessarily true. SNMP can transport arbitrary objects and can transport certificates if necessary. The document indicates some issues with size of certificates and current maximum practical data sizes, however if the compact encoding requirement extends to the internal certificate information this should be less of an issue. 1.1.8 Reliable AAA Transport - New Grade T (from P) The requirements is stated rather strongly and makes substantial assumptions of AAA protocol architecture and based upon current protocols and their failings. SNMP allows for great flexibility in retransmission schemes depending upon the importance of the data.
1.1.9 Run over IPv4 - Grade T SNMP has operated in this mode for many years. 1.1.10 Run over IPv6 - New Grade T (from P) SNMP must support IPv6 for many other systems so support for this should be possible by the time the requirement becomes effective. The document indicates that experimental versions satisfying this requirement are already in existence. 1.1.11 Support Proxy and Routing Brokers - New Grade T (from P) The requirements make significant assumptions about the final architecture. It is well within the capabilities of SNMP to provide intermediaries which channel data flows between multiple parties. The document downgrades SNMPs compliance with this requirement due to issues which are covered more specifically under "Data Object Confidentially" which the evaluator has downgraded to P. 1.1.12 Auditability - New Grade T (from F) Data flows inside SNMP are easily auditable by having secondary data flows established which provide copies of all information to auxiliary servers. The document grades this as a failure, but this support is only minor additions within a more fully fleshed out set of data flows. 1.1.13 Shared Secret Not Required - Grade T Shared secrets are not required by SNMP. They are desirable in many instances where a lower level does not provide the necessary capabilities. The document supplies pointers to various security modes available. 1.1.14 Ability to Carry Service Specific Attributes - Grade T SNMP has long had the ability for other parties to create new unambiguous attributes. 1.2 Authentication Requirements 1.2.1 NAI Support - Grade T SNMP easily supports this. NAIs were defined to be easily carried in existing protocols.
1.2.2 CHAP Support - Grade T SNMP can easily provide objects to pass the necessary information for CHAP operation. 1.2.3 EAP Support - New Grade T (from P) SNMP can easily provide objects to pass the necessary information for EAP operation. As with CHAP or PAP MIB objects can be created to control this operation thus the upgrade from the document grade. 1.2.4 PAP/Clear-text Passwords - New Grade P (from T) SNMP can easily provide objects to pass the necessary information for PAP operation. The requirement about non-disclosure of clear text passwords make assumptions about the protocol implementation. The choice to use clear text passwords is inherently insecure and forced protocol architecture don't really cover this. This requirement grade is downgraded to P (partial) because the document does not really address the confidentially of the data at application proxies. 1.2.5 Reauthorization on demand - Grade T SNMP can easily provide objects to control this operation. 1.2.6 Authorization w/o Authentication - New Grade T (from T) The document makes an incorrect interpretation of this requirement. However, SNMP makes no restriction which prevents to desired requirements. No actual change of grade is necessary, since both the actual requirements and the incorrect interpretation are satisfied by SNMP. 1.3 Authorization Requirements 1.3.1 Static and Dynamic IP Addr Assignment - Grade T SNMP can easily provide objects to control this operation. 1.3.2 RADIUS Gateway Capability - Grade T As the document describes, with the addition of any necessary compatibility variables SNMP can be gatewayed to RADIUS applications.
1.3.3 Reject Capability - Grade T Any of the active components in the SNMP based structure could decide to reject and authentication request for any reason. Due to mixing different levels of requirements the document doesn't attempt to directly address this, instead indicating that a higher level application can cause this operation. 1.3.4 Preclude Layer 2 Tunneling - New Grade T (from ?) Nothing in SNMP explicitly interacts with the selection of any tunneling mechanisms the client may select. The document author was unclear about the needs here. 1.3.5 Reauth on Demand - Grade T SNMP can easily provide objects to control this operation. 1.3.6 Support for ACLs - Grade T The document indicates that should it be desired SNMP can provide objects to control these operations. In addition, active components can apply substantial further configurable access controls. 1.3.7 State Reconciliation - Grade T The requirements describe an over broad set of required capabilities. The document indicates concern over incompatibilities in the requirements, however SNMP can provide methods to allow active components to reacquire lost state information. These capabilities directly interact with scalability concerns and care needs to be taken when expecting this requirement to be met at the same time as the scalability requirements. 1.3.8 Unsolicited Disconnect - Grade T The document indicates that SNMP can easily provide objects to control this operation. 1.4 Accounting Requirements 1.4.1 Real Time Accounting - Grade T SNMP can provide this mode of operation. The document outlines methods both fully within SNMP and using SNMP to interface with other transfer methods. Many providers already use SNMP for real time
notification of other network events. This capability can directly interact with scalability concerns and implementation care needs to be taken to make this properly interact is large scale environments. 1.4.2 Mandatory Compact Encoding - Grade T The document indicates the possibility of controlling external protocols to handle data transmissions where the BER encoding of SNMP objects would be considered excessive. SNMP BER encoded protocol elements are generally in a fairly compact encoding form compared with text based forms (as used in some existing radius log file implementations). This interacts with the general requirement for carrying service specific attributes and the accounting requirement for extensibility. With careful MIB design and future work on SNMP payload compression the SNMP coding overhead can be comparable with other less extensible protocols. 1.4.3 Accounting Record Extensibility - Grade T SNMP has a strong tradition of allowing vendor specific data objects to be transferred. 1.4.4 Batch Accounting - Grade T There are many methods which a SNMP based system could use for batch accounting. The document discusses SNMP parameters to control the batching process and indicates that certain existing MIBs contain examples of implementation strategies. SNMP log tables can provide accounting information which can be obtained in many methods not directly related to real time capabilities. The underlying system buffering requirements are similar regardless of the protocol used to transport the information. 1.4.5 Guaranteed Delivery - Grade T SNMP is very amenable to providing guaranteed delivery. Particularly in a pull model (versus the often assumed push model) the data gatherer can absolutely know that all data has been transfered. In the common push model the data receiver does not know if the originator of the data is having problems delivering the data. 1.4.6 Accounting Timestamps - Grade T Timestamps are used for many SNMP based operations. The document points at the DateAndTime textual convention which is available for use. As with all environments the timestamps accuracy needs evaluation before the information should be relied upon.
1.4.7 Dynamic Accounting - Grade T As long as there is some way to relate multiple records together there are no problems resolving multiple records for the same session. This interacts with the scalability requirement and care must be taken when implementing a system with both of these requirements. 1.5 MOBILE IP Requirements 1.5.1 Encoding of MOBILE IP Registration Messages - Grade T SNMP can easily provide objects to transfer this information. 1.5.2 Firewall Friendly - New Grade T (from P) SNMP is already deployed in many operational networks. SNMPv3 addresses most concerns people had with the operation of previous versions. True SNMPv3 proxies (as opposed to AAA proxies) should become commonplace components in firewalls for those organizations which require firewalls. 1.5.3 Allocation of Local Home Agent - New Grade T (from ?) SNMP is not concerned with the LHA. This can be under control of the Local network to meet its needs. 2. Summary Discussion SNMP appears to meet most stated requirements. The areas where the SNMP proposal falls short are areas where specific AAA architectures are envisioned and requirements based upon that architecture are specified. Scaling of the protocol family is vital to success of a AAA suite. The SNMP protocol has proved scalable in existing network management and other high volume data transfer operations. Care needs to be taken in the design of a large scale system to ensure meeting the desired level of service, but this is true of any large scale project. 3. General Requirements SNMP is well understood and already supported in many ISP and other operational environments. Trust models already exist in many cases and can be adapted to provide the necessary access controls needed by the AAA protocols. Important issues with previous versions of SNMP have been corrected in the current SNMPv3 specification.
The SNMP proposal is silent on the specific data variables and message types to be implemented. This is largely due to the requirements not specifying the necessary data elements and the time constraints in extracting that information from the base document set. Such a data model is necessary regardless of the ultimate protocol selected. 4. Summary Recommendation SNMP appears to fully meet all necessary requirements for the full AAA protocol family.C.2 SNMP CON Evaluation
Evaluation of SNMP AAA Requirements CON Evaluation Evaluator - Michael StJohns Ref [1] is "Comparison of SNMPv3 Against AAA Network Access Requirements", aka 'the document' Ref [2] is the aaa eval criteria as modified by us. The document uses T to indicate total compliance, P to indicate partial compliance and F to indicate no compliance. For each section I've indicated my grade for the section. If there is no change, I've indicated that and the grade given by the authors. Section 1 - Per item discussion 1.1 General Requirements 1.1.1 Scalability - Although the document indicates compliance with the requirement, its unclear how SNMP actually meets those requirements. The document neither discusses how SNMP will scale, nor provides applicable references. The argument that there is an existence proof given the deployed SNMP systems appears to assume that one manager contacting many agents maps to many agents (running AAA) contacting one AAA server. A server driven system has substantially different scaling properties than a client driven system and SNMP is most definitely a server (manager) driven system. Eval - F 1.1.2 Fail-over - The document indicates the use of application level time outs to provide this mechanism, rather than the mechanism being a characteristic of the proposed protocol. The protocol provides only partial compliance with the requirement. Eval - P
1.1.3 Mutual Authentication - There is some slight handwaving here, but the protocol's USM mode should be able to support this requirement. Eval - No Change (T) 1.1.4 Transmission Level Security - The authors should elaborate on the specific use of the SNMPv3 modes to support these requirements, but the text is minimally acceptable. Eval - No Change (T) 1.1.5 Data Object Confidentiality - The authors describe a mechanism which does not appear to completely meet the requirement. VACM is a mechanism for an end system (agent) to control access to its data based on manager characteristics. This mechanism does not appear to map well to this requirement. Eval - P 1.1.6 Data Object Integrity - There appears to be some handwaving going on here. Again, SNMP does not appear to be a good match to this requirement due to at least in part a lack of a proxy intermediary concept within SNMP. Eval - F 1.1.7 Certificate Transport - The document does indicate compliance, but notes that optimization might argue for use of specialized protocols. Eval - No Change (T) 1.1.8 Reliable AAA Transport - The document indicates some confusion with the exact extent of this requirement. Given the modifications suggested by the eval group to the explanatory text in [2] for the related annotation, the point by point explanatory text is not required. The document does indicate that the use of SNMP is irrespective of the underlying transport and the support of this requirement is related at least partially to the choice of transport. However, SNMP over UDP - the most common mode for SNMP - does not meet this requirement. Eval - No Change (P) 1.1.9 Run over IPv4 - While the evaluator agrees that SNMPv3 runs over V4, the authors need to point to some sort of reference. Eval - No Change (T) 1.1.10 Run over IPv6 - The document indicates both experimental implementations and future standardization of SNMPv3 over IPv6. Eval - No Change (P) 1.1.11 Support Proxy and Routing Brokers - The section of the document (5.5.3) that, by title, should have the discussion of SNMP proxy is marked as TBD. The section notes that the inability to completely comply with the data object confidentiality and integrity requirements might affect the compliance of this section and the evaluator agrees. Eval - F
1.1.12 Auditability - The document indicates no compliance with this requirement. Eval - No Change (F) 1.1.13 Shared Secret Not Required - Slight handwaving here, but SNMPv3 does not necessarily require use of its security services if other security services are available. However, the interaction with VACM in the absence of USM is not fully described and may not have good characteristics related to this requirement. Eval - P 1.1.14 Ability to Carry Service Specific Attributes - SNMP complies via the use of MIBs. Eval - No Change (T) 1.2 Authentication Requirements 1.2.1 NAI Support - The document indicates that MIB objects can be created to meet this requirement, but gives no further information. Eval - P 1.2.2 CHAP Support - The document indicates that MIB objects can be created to meet this requirement, but gives no further information. Given the normal CHAP model, its unclear exactly how this would work. Eval - F 1.2.3 EAP Support - The document notes that EAP payloads can be carried as specific MIB objects, but also notes that further design work would be needed to fully incorporate EAP. Eval - No Change (P) 1.2.4 PAP/Clear-text Passwords - The document notes the use of MIB objects to carry the clear text passwords and the protection of those objects under normal SNMPv3 security mechanisms. Eval - No Change (T) 1.2.5 Reauthorization on demand - While there's some handwaving here, its clear that the specific applications can generate the signals to trigger reauthorization under SNMP. Eval - No Change (T) 1.2.6 Authorization w/o Authentication - The author appears to be confusing the AAA protocol authorization with the AAA user authorization and seems to be over generalizing the ability of SNMP to deal with general AAA user authorization. Eval - F 1.3 Authorization Requirements 1.3.1 Static and Dynamic IP Addr Assignment - The reference to MIB objects without more definite references or descriptions continues to be a negative. While the evaluator agrees that MIB objects can represent addresses, the document needs to at least lead the reader in the proper direction. Eval - F
1.3.2 RADIUS Gateway Capability - The transport and manipulation of Radius objects appears to be only a part of what is required. Eval - P 1.3.3 Reject Capability - Again, a clarification of how SNMP might accomplish this requirement would be helpful. The overall document lacks a theory of operation for SNMP in an AAA role that might have clarified the various approaches. Eval - F 1.3.4 Preclude Layer 2 Tunneling - Document indicates lack of understanding of this requirement. Eval - F 1.3.5 Reauth on Demand - See response in 1.3.3 above. None of the text responding to this requirement, nor any other text in the document, nor any of the references describes the appropriate framework and theory. Eval - F 1.3.6 Support for ACLs - The response text again references MIB objects that can be defined to do this job. There is additional engineering and design needed before this is a done deal. Eval - P 1.3.7 State Reconciliation - The text fails to address the basic question of how to get the various parts of the AAA system back in sync. Eval - F 1.3.8 Unsolicited Disconnect - Assuming that the NAS is an SNMP agent for an AAA server acting as an SNMP manager the evaluator concurs. Eval - No Change (T). 1.4 Accounting Requirements 1.4.1 Real Time Accounting - SNMP Informs could accomplish the requirements. Eval - No Change (T) 1.4.2 Mandatory Compact Encoding - This is a good and reasonable response. SNMP can vary the style and type of reported objects to meet specific needs. Eval - No Change (T). 1.4.3 Accounting Record Extensibility - MIBs are extensible. Eval - No Change (T) 1.4.4 Batch Accounting - MIBs provide data collection at various times. Eval - No Change (T) 1.4.5 Guaranteed Delivery - There's some weasel wording here with respect to what guaranteed means, but the description of mechanisms does appear to meet the requirements. Eval - No Change (T)
1.4.6 Accounting Timestamps - Accounting records can use the DateAndTime Textual Convention to mark their times. Eval - No Change (T) 1.4.7 Dynamic Accounting - The author may have partially missed the point on this requirement. While the number of records per session is not of great interest, the delivery may be. The author should go a little more into depth on this requirement. Eval - No Change (T) 1.5 MOBILE IP Requirements 1.5.1 Encoding of MOBILE IP Registration Messages - Registration messages can probably be encoded as SNMP messages. Eval - No Change (T) 1.5.2 Firewall Friendly - There's a chicken and egg problem with the response to the requirement in that the author hopes that SNMP as an AAA protocol will encourage Firewall vendors to make SNMP a firewall friendly protocol. Eval - F 1.5.3 Allocation of Local Home Agent - The author disclaims an understanding of this requirement. Eval - F 2. Summary Discussion The documents evaluation score was substantially affected by a lack of any document, reference or text which described a theory of operation for SNMP in AAA mode. Of substantial concern are the items relating to the AAA server to server modes and AAA client to server modes and the lack of a map to the SNMP protocol for those modes. The evaluator also notes that the scaling issues of SNMP in SNMP agent/manager mode are in no way indicative of SNMP in AAA client/server mode. This has a possibility to substantially impair SNMPs use in an AAA role. However, SNMP may have a reasonable role in the Accounting space. SNMP appears to map well with existing technology, and with the requirements. 3. General Requirements SNMP appears to meet the general requirements of an IP capable protocol, but may not have a proper field of use for all specific requirements.
4. Summary Recommendation Recommended in Part. SNMP is NOT RECOMMENDED for use as either an authentication or authorization protocol, but IS RECOMMENDED for use as an accounting protocol.C.3 RADIUS+ PRO Evaluation
Evaluation of RADIUS AAA Requirements PRO Evaluation Evaluator - Mark Stevens Ref [1] is "Comparison of RADIUS Against AAA Network Access Requirements" Ref [2] is "Framework for the extension of the RADIUS(v2) protocol" Ref [3] is the aaa eval criteria as modified by us. The documents uses T to indicate total compliance, P to indicate partial compliance and F to indicate no compliance. For each section I've indicated my grade for the section. I have indicated whether or not my evaluation differs from the statements made with respect to RADIUS++. The evaluation ratings as given below may differ from the evaluations codified in the document referred to as, "Comparison of RADIUS Against AAA Network Access Requirements" without any indication. 1.1 General Requirements 1.1.1 [a] Scalability - In as much as a protocol's scalability can be measured, the protocol seems to transmit information in a fairly efficient manner.So, in that the protocol appears not to consume an inordinate amount of bandwidth relative to the data it is transmitting, this protocol could be considered scalable. However, the protocol has a limit in the number of concurrent sessions it can support between endpoints. Work arounds exist and are in use. Eval - P (no change) 1.1.2 [b] Fail-over - The document indicates the use of application level time outs to provide this mechanism, rather than the mechanism being a characteristic of the proposed protocol. The fail-over requirement indicates that the protocol must provide the mechanism rather than the application. The implication is that the application need not be aware that the fail-over and subsequent correction when it happens. The application using the RADIUS++ protocol will be involved in fail-over recovery activities. The protocol layer of the software does not appear to have the capability built-in. Given the wording of the requirement: Eval - P (changed from T)
1.1.3 [c] Mutual Authentication - The RADIUS++ protocol provides shared-secret as a built-in facility for mutual authentication. The authors of the document suggest the use of IPSec to obtain mutual authentication functions. The RADIUS++ protocol provides no road blocks to obtaining mutual authentication between instances of AAA applications, however the protocol provides no facilities for doing so. 1.1.4 [d] Transmission Level Security - The RADIUS++ protocol provides no transmission level security features, nor does it preclude the use of IPSec to obtain transmission level security. Eval - P (no change) 1.1.5 [e] Data Object Confidentiality - The document describes a RAIDUS++ message designed to server as an envelope in which encrypted RADIUS messages (attributes) may be enclosed. Eval - T (no change) 1.1.6 [f] Data Object Integrity - Using visible signatures, the RADIUS++ protocol appears to meet this requirement. Eval - T (no change) 1.1.7 [g] Certificate Transport - The document indicates compliance through the use of the CMS-Data Radius Attribute (message). Eval - T (no change) 1.1.8 [h] Reliable AAA Transport - The document points out that RADIUS++ can be considered a reliable transport when augmented with Layer 2 Tunneling Protocol. The protocol itself does not provide reliability features. Reliability remains the responsibility of the application or a augmenting protocol. Eval - P (no change) 1.1.9 [i] Run over IPv4 - Eval - T (no change) 1.1.10 [j] Run over IPv6 - an IPv6 Address data type must be defined. Eval - T (no change) 1.1.11 [k] Support Proxy and Routing Brokers - There is no mechanism for rerouting requests, but an extension can be made to do so. Eval - T (no change) 1.1.12 [l] Auditability - The document indicates no compliance with this requirement. Eval - F (no change) 1.1.13 [m] Shared Secret Not Required - RADIUS++ can be configured to run with empty shared secret values. Eval - T (no change)
1.1.14 [n] Ability to Carry Service Specific Attributes - Vendor escape mechanism can be used for this purpose.. Eval - T (no change) 1.2 Authentication Requirements 1.2.1 [a] NAI Support - Eval - T (no change) 1.2.2 [b] CHAP Support - Subject to dictionary attacks. Eval - P (changed from T) 1.2.3 [c] EAP Support - Eval - T (no change) 1.2.4 [d] PAP/Clear-text Passwords - No end-to-end security, but potential for encapsulation exists within current paradigm of the protocol. - Eval -T (no change) 1.2.5 [e] Reauthentication on demand - The RADIUS protocol supports re-authentication. In case re-authentication is initiated by the user or AAA client, the AAA client can send a new authentication request. Re-authentication can be initiated from the visited or home AAA server by sending a challenge message to the AAA client. Eval - T (no change) 1.2.6 [f] Authorization w/o Authentication - A new message type can be created to enable RADIUS++ to support Aw/oA . Eval - T (no change) 1.3 Authorization Requirements 1.3.1[a] Static and Dynamic IP Addr Assignment - Both supported. IPv6 would require the definition of a new address data type. Eval - P (no change) 1.3.2 [b] RADIUS Gateway Capability - The transport and manipulation of RADIUS objects appears to be only a part of what is required. Requirement seems to be worded to preclude RADIUS. Eval - P (changed from T) 1.3.3 [c] Reject Capability - Eval -T 1.3.4 [d] Preclude Layer 2 Tunneling - I do not see a definition in the AAA eval criteria document. Eval - ?
1.3.5 [e] Reauthorization on Demand - Implementation in the field demonstrate that extensions to RADIUS can support the desired behavior. Re-authentication is currently coupled to re- authorization. Eval - P (no change) 1.3.6 [f] Support for ACLs - Currently done in the applications behind the RADIUS end points, not the within the protocol. RADIUS++ could define additional message types to deal with expanded access control within new service areas. Eval - P (no change) 1.3.7 [g] State Reconciliation - Eval - F (no change) 1.3.8 [h] Unsolicited Disconnect - RADIUS++ extensions to support. Eval - T. (no change) 1.4 Accounting Requirements 1.4.1 [a] Real Time Accounting - Eval - T (no change) 1.4.2 [b] Mandatory Compact Encoding - Eval - T (no change) 1.4.3 [c] Accounting Record Extensibility - Eval - T (no change) 1.4.4 [d] Batch Accounting - RADIUS++ offers no new features to support batch accounting. Eval - F No change) 1.4.5 [e] Guaranteed Delivery - Retransmission algorithm employed. Eval - T (no change) 1.4.6 [f] Accounting Timestamps - RADIUS++ extensions support timestamps. Eval - T (no change) 1.4.7 [g] Dynamic Accounting - RADIUS++ extensions to support. Eval - T (no change) 1.5 MOBILE IP Requirements 1.5.1 [a] Encoding of MOBILE IP Registration Messages - RADIUS++ extensions can be made to include registration messages as an opaque payload. Eval - T (no change) 1.5.2 [b] Firewall Friendly - RADIUS is known to be operational in environments where firewalls acting as a proxy are active. Eval - T (no change) 1.5.3 [c] Allocation of Local Home Agent -Requirement statement needs some clarification and refinement. Eval - F (no change)
2. Summary Discussion The RADIUS protocol, and its associated extensions, is presently not fully compliant with the AAA Network Access requirements. However, it is possible with a small effort to extend present procedures to meet the requirements as listed in, while maintaining a high level of interoperability with the wide deployment and installed base of RADIUS clients and servers. 3. General Requirements RADIUS++ the protocol and the application meet the majority of the requirements and can be extended to meet the requirements where necessary. 4. Summary Recommendation RADIUS++ as it could be developed would provide a level of backward compatibility that other protocols cannot achieve. By extending RADIUS in the simple ways described in the documents listed above, the transition from existing RADIUS-based installations to RADIUS++ installations would be easier. Although accounting continues to be weaker than other approaches, the protocol remains a strong contender for continued use in the areas of Authorization and Authentication.C.4 RADIUS+ CON Evaluation
Evaluation of RADIUS++ (sic) AAA Requirements CON Evaluation Evaluator - David Nelson Ref [1] is "Comparison of RADIUS Against AAA Network Access Requirements", a.k.a. 'the document' Ref [2] is "Framework for the extension of the RADIUS(v2) protocol", a.k.a. 'the protocol' Ref [3] is the AAA evaluation criteria as modified by us. Ref [4] is RFC 2869. Ref [5] is an expired work in progress "RADIUS X.509 Certificate Extensions". Ref [6] is RFC 2868 The document uses T to indicate total compliance, P to indicate partial compliance and F to indicate no compliance. Evaluator's Note: The document [1] pre-dates the protocol [2]. It is clear from reading [2], that some of the issues identified as short comings in [1] are now addressed in [2]. The evaluator has attempted to take note of these exceptions, where they occur.
Section 1 - Per item discussion 1.1 General Requirements 1.1.1 Scalability - The document [1] indicates partial compliance, largely in deference to the "tens of thousands of simultaneous requests" language in [3], that has been deprecated. The issue of simultaneous requests from a single AAA client is addressed in [1], indicating that the apparent limitation of 256 uniquely identifiable outstanding request can be worked around using well known techniques, such as the source UDP port number of the request. The document claims "P", and the evaluator concurs. 1.1.2 Fail-over - The document [1] indicates the use of application level time outs to provide the fail-over mechanism. Since the AAA protocol is indeed an application-layer protocol, this seems appropriate. There are significant issues of how to handle fail- over in a proxy-chain environment that have not been well addressed, however. The document claims "T", and the evaluator awards "P". 1.1.3 Mutual Authentication - The document [1] indicates that mutual authentication exists in the presence of a User-Password or CHAP- Password attribute in an Access-Request packet or the Message- Authenticator [4] in any packet. Once again, this addresses hop-by- hop authentication of RADIUS "peers", but does not fully address proxy-chain environments, in which trust models would need to be established. The document further indicates that strong mutual authentication could be achieved using the facilities of IPsec. This claim would apply equally to all potential AAA protocols, and cannot be fairly said to be a property of the protocol itself. The document claims "T", and the evaluator awards "F". 1.1.4 Transmission Level Security - The document [1] indicates that transmission layer security, as defined in [3], is provided in the protocol, using the mechanisms described in section 1.1.3. It should be noted that this requirement is now a SHOULD in [3]. The document claims "P", and the evaluator concurs. 1.1.5 Data Object Confidentiality - The document [1] indicates that end-to-end confidentiality is not available in RADIUS, but goes on to say that it could be added. The protocol [2] actually makes an attempt to specify how this is to be done, in section 4.3.2.2 of [2], using a CMS-data attribute, based in large part upon RFC 2630. The evaluator has not, at this time, investigated the applicability of RFC 2630 to the AAA work. The document claims "F", but in light of the specifics of the protocol [2], the evaluator awards "P".
1.1.6 Data Object Integrity - The document [1] indicates that end- to-end integrity is not available in RADIUS, but goes on to say that it could be added. The protocol [2] actually makes an attempt to specify how this is to be done, in section 4.3.2.1 of [2], using a CMS-data attribute, based in large part upon RFC 2630. The evaluator has not, at this time, investigated the applicability of RFC 2630 to the AAA work. The document claims "F", but in light of the specifics of the protocol [2], the evaluator awards "P". 1.1.7 Certificate Transport - The document [1] indicates that certificate transport is not available in RADIUS, but goes on to say that it could be added. The protocol [2] actually makes an attempt to specify how this is to be done, in section 4.3.2.3 of [2], using a CMS-data attribute, based in large part upon RFC 2630. The evaluator has not, at this time, investigated the applicability of RFC 2630 to the AAA work. Other relevant work in the area of certificate support in RADIUS may be found in an expired work in progress, "RADIUS X.509 Certificate Extensions" [5]. The document claims "F", but in light of the specifics of the protocol [2], the evaluator awards "P". 1.1.8 Reliable AAA Transport - The document [1] indicates that RADIUS provides partial compliance with the requirements of the original AAA requirements document. However, in [3], the requirement has been simplified to "resilience against packet loss". Once again, the evaluator finds that the protocol [2] meets this criteria on a hop- by-hop basis, but fails to effectively address these issues in a proxy-chain environment. The document claims "P", and the evaluator awards "F". 1.1.9 Run over IPv4 - RADIUS is widely deployed over IPv4. The document claims "T", and the evaluator concurs. 1.1.10 Run over IPv6 - The document [1] indicates that adoption of a limited number of new RADIUS attributes to support IPv6 is straightforward. Such discussion has transpired on the RADIUS WG mailing list, although that WG is in the process of shutting down. The document claims "P", and the evaluator concurs. 1.1.11 Support Proxy and Routing Brokers - The document [1] indicates that RADIUS is widely deployed in proxy-chains of RADIUS servers. This is equivalent to the Proxy Broker case, but the Routing Broker case is a different requirement. The protocol [2] does not describe any detail of how a Routing Broker might be accommodated, although it opens the door by indicating that the RADIUS++ protocol is peer-to- peer, rather than client/server. The document claims "P", and the evaluator awards "F".
1.1.12 Auditability - The document [1] indicates no compliance with this requirement. The document claims "F", and the evaluator concurs. 1.1.13 Shared Secret Not Required - The document [1] indicates that RADIUS may effectively skirt the requirement of application-layer security by using a value of "zero" for the pre-shared secret. While this is a bit creative, it does seem to meet the requirement. The document claims "T" and the evaluator concurs. 1.1.14 Ability to Carry Service Specific Attributes - RADIUS has a well defined Vendor-Specific Attribute, which, when properly used, does indeed provide for the ability to transport service-specific attributes. The document claims "T", and the evaluator concurs. 1.2 Authentication Requirements 1.2.1 NAI Support - The document [1] indicates that RADIUS specifies the NAI as one of the suggested formats for the User-Name attribute. The document claims "T", and the evaluator agrees. 1.2.2 CHAP Support - CHAP support is widely deployed in RADIUS. The document claims [1] "T", and the evaluator concurs. 1.2.3 EAP Support - The document [1] indicates that EAP support in RADIUS is specified in [4]. The document claims [1] "T", and the evaluator concurs. 1.2.4 PAP/Clear-text Passwords - The document [1] indicates that RADIUS provides protection of clear-text passwords on a hop-by-hop basis. The protocol [2] indicates how additional data confidentiality may be obtained in section 4.3.2.2 of [2], using a CMS-data attribute, based in large part upon RFC 2630. The evaluator has not, at this time, investigated the applicability of RFC 2630 to the AAA work. The document claims [1] "F", but in light of the specifics of the protocol [2], the evaluator awards "P". 1.2.5 Reauthentication on demand - The document [1] indicates that RADIUS may accomplish re-authentication on demand by means of an Access-Challenge message sent from a server to a client. The evaluator disagrees that this is likely to work for a given session once an Access-Accept message has been received by the client. The document claims "T", and the evaluator awards "F". 1.2.6 Authorization w/o Authentication - This requirement, as applied to the protocol specification, mandates that non- necessary authentication credentials not be required in a request for authorization. The actual decision to provide authorization in the
absence of any authentication resides in the application (e.g. AAA server). RADIUS does require some form of credential in request messages. The document [1] claims "F", and the evaluator concurs. 1.3 Authorization Requirements 1.3.1 Static and Dynamic IP Addr Assignment - The document [1] indicates that RADIUS can assign IPv4 addresses, and can easily be extended to assign IPv6 addresses (see section 1.1.10). Of greater concern, however, is the issue of static vs. dynamic addresses. If dynamic address has the same meaning as it does for DHCP, then there are issues of resource management that RADIUS has traditionally not addressed. The document claims "P", and the evaluator concurs. 1.3.2 RADIUS Gateway Capability - The document [1] maintains that a RADIUS++ to RADIUS gateway is pretty much a tautology. The document claims "T", and the evaluator concurs. 1.3.3 Reject Capability - The document [1] maintains that RADIUS Proxy Servers, and potentially RADIUS++ Routing Brokers, have the ability to reject requests based on local policy. The document claims "T" and the evaluator concurs. 1.3.4 Preclude Layer 2 Tunneling - The document [1] indicates that [6] defines support for layer two tunneling in RADIUS. The document claims "T", and the evaluator concurs. 1.3.5 Reauth on Demand - The document [1] indicates that RADIUS provides this feature by means of the Session-Timeout and Termination- Action attributes. While this may, in fact, be sufficient to provide periodic re-authorization, it would not provide re- authorization on demand. The protocol [2] does not address this further. The document claims "P", and the evaluator awards "F". 1.3.6 Support for ACLs - The document [1] describes the attributes in RADIUS that are used to convey the access controls described in [3]. Certain of these (e.g. QoS) are not currently defined in RADIUS, but could easily be defined as new RADIUS attributes. The document claims "P", and the evaluator concurs. 1.3.7 State Reconciliation - The document [1] addresses each of the sub- items, as listed in the original AAA requirements document. In reviewing the document against the modified requirements of [3], there is still an issue with server-initiated state reconciliation messages. While the protocol [2] makes provision for such messages, as servers are allowed to initiate protocol dialogs, no detailed
message formats are provided. This is an area that has traditionally been a short coming of RADIUS. The document claims "P", and the evaluator awards "F". 1.3.8 Unsolicited Disconnect - Much of the discussion from the previous section applies to this section. The document [1] claims "F", and the evaluator concurs. 1.4 Accounting Requirements 1.4.1 Real Time Accounting - RADIUS Accounting is widely deployed and functions within the definition of real time contained in [3]. The document [1] claims "T", and the evaluator concurs. 1.4.2 Mandatory Compact Encoding - RADIUS Accounting contains TLVs for relevant accounting information, each of which is fairly compact. Note that the term "bloated" in [3] is somewhat subjective. The document [1] claims "T", and the evaluator concurs. 1.4.3 Accounting Record Extensibility - RADIUS Accounting may be extended by means of new attributes or by using the Vendor-Specific attribute. While it has been argued that the existing attribute number space is too small for the required expansion capabilities, the protocol [2] addresses this problem in section 3.0, and its subsections, of [2]. The document [1] claims "T", and the evaluator concurs. 1.4.4 Batch Accounting - RADIUS has no explicit provisions for batch accounting, nor does the protocol [2] address how this feature might be accomplished. The document [1] claims "F", and the evaluator concurs. 1.4.5 Guaranteed Delivery - RADIUS Accounting is widely deployed and provides guaranteed delivery within the context of the required application-level acknowledgment. The document [1] claims "T", and the evaluator concurs. 1.4.6 Accounting Timestamps - The document [1] indicates that this feature is specified in [4] as the Event-Timestamp attribute. The document claims [1] "T", and the evaluator concurs. 1.4.7 Dynamic Accounting - The document [1] indicates that this requirement is partially met using the accounting interim update message as specified in [4]. In addition, there was work in the RADIUS WG regarding session accounting extensions that has not been included in [4], i.e., some expired works in progress. The document claims [1] "P", and the evaluator concurs.
1.5 MOBILE IP Requirements 1.5.1 Encoding of MOBILE IP Registration Messages - The document [1] claims "F", and the evaluator concurs. 1.5.2 Firewall Friendly - The document [1] indicates that RADIUS deployment is know to have occurred in fire-walled environments. The document claims "T", and the evaluator concurs. 1.5.3 Allocation of Local Home Agent - The document [1] claims "F", and the evaluator concurs. 2. Summary Discussion The document [1] and the protocol [2] suffer from having been written in a short time frame. While the protocol does provide specific guidance on certain issues, citing other relevant documents, it is not a polished protocol specification, with detailed packet format diagrams. There is a pool of prior work upon which the RADIUS++ protocol may draw, in that many of the concepts of Diameter were first postulated as works in progress within the RADIUS WG, in an attempt to "improve" the RADIUS protocol. All of these works in progress have long since expired, however. 3. General Requirements RADIUS++ meets many of the requirements of an AAA protocol, as it is the current de facto and de jure standard for AAA. There are long- standing deficiencies in RADIUS, which have been well documented in the RADIUS and NASREQ WG proceedings. It is technically possible to revamp RADIUS to solve these problems. One question that will be asked, however, is: "What significant differences would there be between a finished RADIUS++ protocol and the Diameter protocol?". 4. Summary Recommendation Recommended in part. What may possibly be learned from this submission is that it is feasible to have a more RADIUS-compliant RADIUS-compatibility mode in Diameter.
C.5 Diameter PRO Evaluation
Evaluation of Diameter against the AAA Requirements PRO Evaluation Evaluator - Basavaraj Patil Ref [1] is "Diameter Framework Document". Ref [2] is "Diameter NASREQ Extensions". Ref [3] is the AAA evaluation criteria as modified by us. Ref [4] is "Diameter Accounting Extensions". Ref [5] is "Diameter Mobile IP Extensions". Ref [6] is "Diameter Base Protocol". Ref [7] is "Diameter Strong Security Extension". Ref [8] is "Comparison of Diameter Against AAA Network Access Requirements". The document uses T to indicate total compliance, P to indicate partial compliance and F to indicate no compliance. Evaluator's note : The Diameter compliance document [8] claims Total "T" compliance with all the requirements except : - 1.2.5 - 1.5.2 Section 1 - Per item discussion 1.1 General Requirements 1.1.1 Scalability Diameter is an evolution of RADIUS and has taken into consideration all the lessons learned over many years that RADIUS has been in service. The use of SCTP as the transport protocol reduces the need for multiple proxy servers (Sec 3.1.1 Proxy Support of [1]) as well as removing the need for application level acks. The use and support of forwarding and redirect brokers enhances scalability. Evaluator concurs with the "T" compliance on this requirement. 1.1.2 Fail-over Again with the use of SCTP, Diameter is able to detect disconnect indications upon which it switches to an alternate server (Sec 4.0 [6]). Also Requests and Responses do not have to follow the same path and this increases the reliability. Evaluator concurs with the "T" compliance on this requirement.
1.1.3 Mutual Authentication The compliance document quotes the use of symmetric transforms for mutual authentication between the client and server (Sec 7.1 of [6]). The use of IPSec as an underlying security mechanism and thereby use the characteristics of IPSec itself to satisfy this requirement is also quoted. Evaluator concurs with the "T" compliance on this requirement. 1.1.4 Transmission Level Security Although this requirement has been deprecated by the AAA evaluation team the document complies with it based on the definition (referring to hop-by-hop security). Section 7.1 of [6] provides the details of how this is accomplished in Diameter. Evaluator concurs with the "T" compliance on this requirement. 1.1.5 Data Object Confidentiality This requirement seems to have come from Diameter. Ref [7] explains in detail the use of Cryptographic Message Syntax (CMS) to achieve data object confidentiality. A CMS-Data AVP is defined in [7]. Evaluator concurs with the "T" compliance on this requirement. 1.1.6 Data Object Integrity Using the same argument as above and the hop-by-hop security feature in the protocol this requirement is completely met by Diameter. Evaluator concurs with the "T" compliance on this requirement. 1.1.7 Certificate Transport Again with the use of the CMS-Data AVP, objects defined as these types of attributes allow the transport of certificates. Evaluator concurs with the "T" compliance on this requirement. 1.1.8 Reliable AAA Transport Diameter recommends that the protocol be run over SCTP. SCTP provides the features described for a reliable AAA transport. Although the compliance is not a perfect fit for the definition of this tag item, it is close enough and the functionality achieved by using SCTP is the same. Evaluator concurs with the "T" compliance on this requirement.
1.1.9 Run over IPv4 Is an application layer protocol and does not depend on the underlying version of IP. Evaluator concurs with the "T" compliance on this requirement. 1.1.10 Run over IPv6 Is an application layer protocol and does not depend on the underlying version of IP. Evaluator concurs with the "T" compliance on this requirement. 1.1.11 Support Proxy and Routing Brokers Section 3.1.1/2 of the framework document [1] provides an explanation of how Diameter supports proxy and routing brokers. In fact it almost appears as though the requirement for a routing broker came from Diameter. Evaluator concurs with the "T" compliance on this requirement. 1.1.12 Auditability With the use of CMS-Data AVP [7] a trail is created when proxies are involved in the transaction. This trail can provide auditability. Evaluator concurs with the "T" compliance on this requirement. 1.1.13 Shared Secret Not Required With the use of IPSec as the underlying security mechanism, Diameter does not require the use of shared secrets for message authentication. Evaluator concurs with the "T" compliance on this requirement. 1.1.14 Ability to Carry Service Specific Attributes The base protocol [6] is defined by Diameter and any one else can define specific extensions on top of it. Other WGs in the IETF can design an extension on the base protocol with specific attributes and have them registered by IANA. Evaluator concurs with the "T" compliance on this requirement.
1.2 Authentication Requirements 1.2.1 NAI Support The base protocol [6] defines an AVP that can be used to support NAIs. Diameter goes one step further by doing Message forwarding based on destination NAI AVPs. Evaluator concurs with the "T" compliance on this requirement. 1.2.2 CHAP Support Reference [2] section 3.0 describes the support for CHAP. Evaluator concurs with the "T" compliance on this requirement. 1.2.3 EAP Support Reference [2] section 4.0 describes the support for EAP. Evaluator concurs with the "T" compliance on this requirement. 1.2.4 PAP/Clear-text Passwords Reference [2] section 3.1.1.1 describes the support for PAP. Evaluator concurs with the "T" compliance on this requirement. 1.2.5 Reauthentication on demand The use of Session-Timeout AVP as the mechanism for reauthentication is claimed by the compliance document. However no direct references explaining this in the base protocol [6] document were found. Evaluator deprecates the compliance on this to a "P" Note: However this is a trivial issue. 1.2.6 Authorization w/o Authentication Diameter allows requests to be sent without having any authentication information included. A Request-type AVP is defined in [2] and it can specify authorization only without containing any authentication. Evaluator concurs with the "T" compliance on this requirement.
1.3 Authorization Requirements 1.3.1 Static and Dynamic IP Addr Assignment The base protocol includes an AVP for carrying the address. References [6.2.2 of 2] and [4.5 of 5] provide detailed explanations of how this can be done. Evaluator concurs with the "T" compliance on this requirement. 1.3.2 RADIUS Gateway Capability One of the basic facets of Diameter is to support backward compatibility and act as a RADIUS gateway in certain environments. Evaluator concurs with the "T" compliance on this requirement. 1.3.3 Reject Capability Based on the explanation provided in the compliance document for this requirement evaluator concurs with the "T" compliance on this requirement. 1.3.4 Preclude Layer 2 Tunneling Ref [2] defines AVPs supporting L2 tunnels Evaluator concurs with the "T" compliance on this requirement. 1.3.5 Reauth on Demand A session timer defined in [6] is used for reauthorization. However Diameter allows reauthorization at any time. Since this is a peer- to-peer type of protocol any entity can initiate a reauthorization request. Evaluator concurs with the "T" compliance on this requirement. 1.3.6 Support for ACLs Diameter defines two methods. One that supports backward compatibility for RADIUS and another one with the use of a standard AVP with the filters encoded in it. Evaluator concurs with the "T" compliance on this requirement. 1.3.7 State Reconciliation A long explanation on each of the points defined for this tag item in the requirements document. Evaluator concurs with the "T" compliance for this requirement.
1.3.8 Unsolicited Disconnect The base protocol [6] defines a set of session termination messages which can be used for unsolicited disconnects. Evaluator concurs with the "T" compliance on this requirement. 1.4 Accounting Requirements 1.4.1 Real Time Accounting Evaluator concurs with the "T" compliance based on explanations in [4]. 1.4.2 Mandatory Compact Encoding Use of Accounting Data Interchange Format (ADIF)-Record-AVP for compact encoding of accounting data. Evaluator concurs with the "T" compliance. 1.4.3 Accounting Record Extensibility ADIF can be extended. Evaluator concurs with the "T" compliance. 1.4.4 Batch Accounting Sec 1.2 of [4] provides support for batch accounting. 1.4.5 Guaranteed Delivery Sections 2.1/2 of [4] describe messages that are used to guarantee delivery of accounting records. Evaluator concurs with the "T" compliance. 1.4.6 Accounting Timestamps Timestamp AVP [6] is present in all accounting messages. Evaluator concurs with the "T" compliance. 1.4.7 Dynamic Accounting Interim accounting records equivalent to a call-in-progress can be sent periodically. Evaluator concurs with the "T" compliance.
1.5 MOBILE IP Requirements 1.5.1 Encoding of MOBILE IP Registration Messages Ref [5] provides details of how Diameter can encode MIP messages. Evaluator concurs with the "T" compliance. 1.5.2 Firewall Friendly Some handwaving here and a possible way of solving the firewall problem with a Diameter proxy server. Document claims "T", evaluator deprecates it to a "P" 1.5.3 Allocation of Local Home Agent Diameter can assign a local home agent in a visited network in conjunction with the FA in that network. Evaluator concurs with the "T" Summary Recommendation Diameter is strongly recommended as the AAA protocol. The experience gained from RADIUS deployments has been put to good use in the design of this protocol. It has also been designed with extensibility in mind thereby allowing different WGs to develop their own specific extension to satisfy their requirements. With the use of SCTP as the transport protocol, reliability is built in. Security has been addressed in the design of the protocol and issues that were discovered in RADIUS have been fixed. Diameter also is a session based protocol which makes it more scalable. The support for forwarding and redirect brokers is well defined and this greatly improves the scalability aspect of the protocol. Lastly the protocol has been implemented by at least a few people and interop testing done. This in itself is a significant step and a positive point for Diameter to be the AAA protocol.C.6 Diameter CON Evaluation
Evaluation of Diameter against the AAA Requirements CON Brief Evaluator: Barney Wolff
Section 1 - Per item discussion
1.1 General Requirements
1.1.1 Scalability - P (was T) The evaluator is concerned with
scalability to the small, not to the large. Diameter/SCTP may prove
difficult to retrofit to existing NAS equipment.
1.1.2 Fail-over - P (was T) SCTP gives an indication of peer
failure, but nothing in any Diameter or SCTP document the evaluator
was able to find even mentions how or when to switch back to a
primary server to which communication was lost. After a failure, the
state machines end in a CLOSED state and nothing seems to trigger
exit from that state. It was not clear whether a server, on
rebooting, would initiate an SCTP connection to all its configured
clients. If not, and in any case when the communication failure was
in the network rather than in the server, the client must itself,
after some interval, attempt to re-establish communication. But no
such guidance is given.
Of course, the requirement itself fails to mention the notion of
returning to a recovered primary. That is a defect in the
requirement. The evaluator has had unfortunate experience with a
vendor's RADIUS implementation that had exactly the defect that it
often failed to notice recovery of the primary.
1.1.3 Mutual Authentication - T
1.1.4 Transmission Level Security - T
1.1.5 Data Object Confidentiality - P (was T). Yes, the CMS data
type is supported. But the work in progress, "Diameter Strong
Security Extension", says:
Given that asymmetric transform operations are expensive, Diameter
servers MAY wish to use them only when dealing with inter-domain
servers, as shown in Figure 3. This configuration is normally
desirable since Diameter entities within a given administrative
domain MAY inherently trust each other. Further, it is desirable
to move this functionality to the edges, since NASes do not
necessarily have the CPU power to perform expensive cryptographic
operations.
Given all the fuss that has been made about "end-to-end"
confidentiality (which really means "NAS-to-home_server"), the
evaluator finds it absurd that the proposed solution is acknowledged
to be unsuited to the NAS.
1.1.6 Data Object Integrity - P (was T). See above. 1.1.7 Certificate Transport - T 1.1.8 Reliable AAA Transport - T 1.1.9 Run over IPv4 - T 1.1.10 Run over IPv6 - T 1.1.11 Support Proxy and Routing Brokers - T 1.1.12 Auditability - T (based on our interpretation as non- repudiation, rather than the definition given in reqts) 1.1.13 Shared Secret Not Required - T 1.1.14 Ability to Carry Service Specific Attributes - T 1.2 Authentication Requirements 1.2.1 NAI Support - T 1.2.2 CHAP Support - T 1.2.3 EAP Support - T 1.2.4 PAP/Clear-text Passwords - T 1.2.5 Reauthentication on demand - P (was T). No mechanism was evident for the server to demand a reauthentication, based for example on detection of suspicious behavior by the user. Session- timeout is not sufficient, as it must be specified at the start. 1.2.6 Authorization w/o Authentication - T 1.3 Authorization Requirements 1.3.1 Static and Dynamic IP Addr Assignment - T 1.3.2 RADIUS Gateway Capability - P (was T). RADIUS has evolved from the version on which Diameter was based. EAP is a notable case where the convention that the Diameter attribute number duplicates the RADIUS one is violated. No protocol, not even RADIUS++, can claim a T on this. 1.3.3 Reject Capability - T (The evaluator fails to understand how any AAA protocol could rate anything other than T on this.)
1.3.4 Preclude Layer 2 Tunneling - T 1.3.5 Reauth on Demand - P (was T). As with reauthentication, there is no evident mechanism for the server to initiate this based on conditions subsequent to the start of the session. 1.3.6 Support for ACLs - P (was T). The evaluator finds the Filter- Rule AVP laughably inadequate to describe filters. For example, how would it deal with restricting SMTP to a given server, unless all IP options are forbidden so the IP header length is known? No real NAS could have such an impoverished filter capability, or it would not survive as a product. 1.3.7 State Reconciliation - P (was T). It is difficult for the evaluator to understand how this is to work in a multi-administration situation, or indeed in any proxy situation. Furthermore, SRQ with no session-id is defined to ask for info on all sessions, not just those "owned" by the requester. 1.3.8 Unsolicited Disconnect - T 1.4 Accounting Requirements 1.4.1 Real Time Accounting - T 1.4.2 Mandatory Compact Encoding - T 1.4.3 Accounting Record Extensibility - T 1.4.4 Batch Accounting - P (was T). The evaluator suspects that simply sending multiple accounting records in a single request is not how batch accounting should or will be done. 1.4.5 Guaranteed Delivery - T 1.4.6 Accounting Timestamps - T (The evaluator notes with amusement that NTP time cycles in 2036, not 2038 as claimed in the Diameter drafts. It's Unix time that will set the sign bit in 2038.) 1.4.7 Dynamic Accounting - T 1.5 MOBILE IP Requirements 1.5.1 Encoding of MOBILE IP Registration Messages - T 1.5.2 Firewall Friendly - F (was T). Until such time as firewalls are extended to know about or proxy SCTP, it is very unlikely that SCTP will be passed. Even then, the convenient feature of being able
to send a request from any port, and get the reply back to that port, means that a simple port filter will not be sufficient, and statefulness will be required. Real friendship would require that both source and dest ports be 1812. 1.5.3 Allocation of Local Home Agent - T 2. Summary Discussion In some areas, Diameter is not completely thought through. In general, real effort has gone into satisfying a stupendous range of requirements. 3. General Requirements Diameter certainly fails the KISS test. With SCTP, the drafts add up to 382 pages - well over double the size of RADIUS even with extensions. The evaluator sympathizes with the political instinct when faced with a new requirement no matter how bizarre, to say "we can do that" and add another piece of filigree. But the major places where Diameter claims advantage over RADIUS, namely "end-to-end" confidentiality and resource management, are just the places where some hard work remains, if the problems are not indeed intractable. More specifically, the evaluator sees no indication that specifying the separate transport protocol provided any advantage to defray the large increase in complexity. Application acks are still required, and no benefit from the transport acks was evident to the evaluator. Nor was there any obvious discussion of why "sequenced in-order" delivery is required, when AAA requests are typically independent. SCTP offers out-of-order delivery, but Diameter seems to have chosen not to use that feature. Whether TLV encoding or ASN.1/BER is superior is a religious question, but Diameter manages to require both, if the "strong" extension is implemented. The evaluator has a pet peeve with length fields that include the header, making small length values invalid, but that is a minor point. Finally, interoperability would be greatly aided by defining a standard "dictionary" format by which an implementation could adopt wholesale a set of attributes, perhaps from another vendor, and at least know how to display them. That is one of the advantages of MIBs.
4. Summary Recommendation Diameter is clearly close enough to meeting the myriad requirements that it is an acceptable candidate, though needing some polishing. Whether the vast increase in complexity is worth the increase in functionality over RADIUS is debatable.
(page 55 continued on part 3)