Network working group Dacheng Zhang Internet Draft Xiaohu Xu Intended status: Experimental Huawei Technologies Co.,Ltd Created: May 24, 2010 Jiankang Yao Expires: November 2010 CNNIC Investigation in HIP Proxies draft-irtf-hiprg-proxies-00 Status of this Memo This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79. This document may contain material from IETF Documents or IETF Contributions published or made publicly available before November 10, 2008. The person(s) controlling the copyright in some of this material may not have granted the IETF Trust the right to allow modifications of such material outside the IETF Standards Process. Without obtaining an adequate license from the person(s) controlling the copyright in such materials, this document may not be modified outside the IETF Standards Process, and derivative works of it may not be created outside the IETF Standards Process, except to format it for publication as an RFC or to translate it into languages other than English. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. 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." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html This Internet-Draft will expire on November 24, 2010. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 1] Internet-Draft Investigation in HIP Proxies May 2010 Copyright Notice Copyright (c) 2009 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. Abstract HIP proxies play an important role in the transition from the current Internet architecture to the HIP architecture. A core objective of a HIP proxy is to facilitate the communications between legacy (or Non-HIP) hosts and HIP hosts while not modifying their protocol stacks. In this document, the legacy hosts served by proxies are referred to as the Made-up Legacy (ML) hosts. Currently, various designing solutions of HIP proxies have been proposed. These solutions may be applicable in different working circumstances. In this document, we attempt to investigate these solutions in detail and compare their performances in different scenarios. Conventions used in this document 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]. Table of Contents 1. Introduction...................................................3 2. Terminologies..................................................4 3. HIP Proxies....................................................4 3.1. Essential Operations of HIP Proxies.......................4 3.2. A Taxonomy of HIP Proxies.................................5 3.3. DI Proxies................................................5 3.4. N-DI Proxies..............................................7 4. Issues with LBMs in Supporting ML Hosts to Initiate Communication ..................................................................9 4.1. An Issues Caused by Intercepting DNS Lookups..............9 Zhang Xu and Jiankang. Expires November 24, 2010 [Page 2] Internet-Draft Investigation in HIP Proxies May 2010 4.2. Issues with LBMs in Capturing and Processing Replies from HIP hosts.....................................................11 5. Issues with LBMs which also Support HIP Hosts to Initiate Communication....................................................12 5.1. DNS Resource Records for ML Hosts........................12 5.2. An Asymmetric Path Issue.................................13 6. Issues with LBMs in supporting dynamic load balance and redundancy.......................................................15 6.1. Application of DI1 proxies in supporting dynamic load balance and redundancy........................................16 6.2. Application of DI2 proxies in supporting dynamic load balance and redundancy........................................17 6.3. Application of DI3 proxies in supporting dynamic load balance and redundancy........................................17 7. Security Consideration........................................17 8. Conclusions...................................................18 9. IANA Considerations...........................................18 10. Acknowledgments..............................................18 11. References...................................................18 Authors' Addresses...............................................20 1. Introduction As core components of HIP extensional solutions, HIP proxies have attracted increasing attention from both the industry and the academia. Currently, multiple research work is engaged in the design and the performance assessment of HIP proxies. In this document, we attempt to investigate several important designing solutions and compare their effectiveness in different scenarios. Actually, there has been a detailed discussion of HIP proxies in [SAL05]. This document can be regarded as a complement of that paper. Some new topics (e.g., the asymmetric path issues occurred in the load- balancing mechanisms for HIP proxies and the necessary of extending the HIP RR for HIP proxies) are discussed in the draft. Theoretically, ML hosts and the HIP hosts they intend to communicate with can be located anywhere in the network. However, in this document, without mentioned otherwise, legacy hosts are located within a private network and HIP hosts are located in the public network, as this is the most important scenario that HIP proxies are expected to support [SAL05]. The remainder of this document is organized as follows. Section 2 lists the key terminologies used in this document. In section 3, we indicate the essential functions of HIP proxies and provide a classification. In section 4 we analyze the issues that HIP proxies have to face in constructing a Load Balancing Mechanism (LBM) which Zhang Xu and Jiankang. Expires November 24, 2010 [Page 3] Internet-Draft Investigation in HIP Proxies May 2010 facilitates communications initiated by ML hosts. Section 5 analyzes the issues that HIP proxies in a LBM have to face if they also need to support communications initiated by HIP hosts. In section 6, we investigate the issues that HIP proxies have to deal with in supporting dynamic load balancing and redundancy. Section 7 provides a brief discussion about the influence brought by DNSsec to the deployment of HIP proxies. Section 8 is the conclusion of the entire document. 2. Terminologies BEX: HIP Base Exchange DI Proxy: DNS Inspecting Proxy HA: HIP association LBM: Load Balancing Mechanism N-DI Proxy: Non-DNS Inspecting Proxy 3. HIP Proxies 3.1. Essential Operations of HIP Proxies A primary function of HIP proxies is to exchange messages with HIP hosts on the performance of legacy hosts, using standard HIP protocols. In order to achieve this, a HIP proxy needs to intercept the packets transported between legacy and HIP hosts before they arrive at their destinations. Upon capturing such a packet, a HIP proxy needs to transfer it into the format which can be recognized by the host which the packet is destined for. Assume a proxy captures a packet sent out by a ML host. If the packet is destined to a HIP host, the proxy first checks whether there is an appropriate HIP association (HA) in its local database which can be used to process the packet. If such a HA is located, the proxy then find the proper key maintained in the HA and use it to encrypt the payload in the packet. The packet is then forwarded to the HIP host. However, if there is no such an HA or it has expired, the proxy needs to use the HI and HIT assigned to the ML host to carry out a HIP Base Exchange (BEX) and generate a new HA with the HIP host. The newly generated HA is then maintained in the local database. After capturing packet from a HIP host, the proxy also needs to use the keying material in the associated HA to decrypt the packet, transfer it into an ordinary IP packet, and forwards the IP packet to the legacy host. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 4] Internet-Draft Investigation in HIP Proxies May 2010 3.2. A Taxonomy of HIP Proxies In practice, there are various design solutions for HIP proxies. These solutions are based on different presumptions and supposed to execute in different circumstances. To benefit the analysis, we classify HIP proxies into DNS lookups Intercepting Proxies (DI proxies) and Non-DNS lookups Intercepting Proxies (N-DI proxies). As indicated by the name, a DI proxy needs to intercept DNS lookups in order to correctly process the follow-up communication between legacy hosts and HIP hosts, while N-DI proxies do not have to. To avoid confusion, in the remainder of this document we use the terms "lookup" and "answer" in specific ways. A lookup refers to the entire process of translating a domain name for a legacy host. The answer of a lookup is the response from a resolution server which terminates the lookup. 3.3. DI Proxies Usually, before a legacy host communicates with a remote host, it needs to query DNS servers to obtain the IP address of its destination. On this premise, a DI proxy can effectively identify the hosts which legacy hosts may contact in near future by intercepting DNS lookups. In practice, it is common to deploy one more multiple local DNS servers (resolvers) for a private network. Therefore, the hosts in the network can send their queries to the resolver instead of communicating with authoritative DNS servers directly. If the resolver does not cache the inquired RRs, it will try to obtain them from authoritative DNS servers. The resolver may have to contact multiple authoritative DNS servers to get the IP address of the authoritative DNS server which actually contains desired DNS RRs. If the resolver is located out of the private network, a HIP proxy located at the border of the network can intercept an initial DNS query from a legacy host and then use the FQDN obtained from the query to initiate a new DNS lookup with the resolver to inquire about the desired information (AAAA RRs, HIP RRs, and etc.). If the host that the legacy host intends to communicate with is HIP enabled, the DNS resolver will hand out a HIP RR associated with an AAAA RR to the proxy. After maintaining the needed information (e.g., HITs, HIs, IPs addresses) in the local database, the proxy returns an answer with an AAAA RR to the legacy host. When the resolver is located inside the private network, conditions are a little more complex. If a proxy can be located on the path between ML hosts and the resolver, it can work exactly as same as Zhang Xu and Jiankang. Expires November 24, 2010 [Page 5] Internet-Draft Investigation in HIP Proxies May 2010 what is illustrated above. The proxies which can be deployed in this way are introduced in the remainder of this sub-section. However, if a proxy is located at the border of the network, it has to spend more efforts to intercept and modify the DNS lookups between the resolver and authoritative DNS servers, because the resolver may have to contact multiple authoritative DNS servers to get a desired answer. In this case, to be more efficient, the proxy can only inspect the responses from DNS services and find out DNS answers. Because the answer of a DNS lookup does not contain any NS RR, it can be easily distinguished from the intermediate responses. After identifying a DNS answer, a DI proxy can locate the DNS sever maintaining the desired RRs from the packet header and identify the FQDN of the inquired host from the packet payload. Then, the proxy initiates an independent lookup to the DNS server to check whether the host is HIP enabled. If it is, the proxy maintains the information of the host for future usage and returns an answer with an AAAA RR to the resolver. Besides intercepting DNS lookups, some kinds of DI proxies also modify the contents of the AAAA RRs in DNS answers for resolvers or ML hosts in order to benefit their following up operations. For instance, [RFC5338] indicates that a HIP proxy can returns HITs rather than IP addresses in DNS answers to ML hosts. Consequently, the data packets from ML hosts to HIP hosts will use the HITs of the HIP hosts as destination addresses. [PAT07] also proposes a proxy solution which requires a HIP proxy to maintain an IP address pool. When sending a DNS answer to a ML host, the proxy selects an IP address from its pool and inserts it in the answer. The legacy host will regard this IP address as the IP address of the peer it intends to communicate with. In the subsequent communication, when the host sends a packet for the remote HIP host, it will use the selected IP address as the destination address. In the remainder of this document, these two types of proxies are referred to as DI1 proxies and DI2 proxies respectively, and the proxies which do not modify the contents of DNS answers (i.e., return the IP addresses of HIP hosts in answers) are referred to as DI3 proxies. Different modifications on DNS answers introduce different influences on the performances of DI proxies and impose different restrictions on their locations. Compared with DI1 and DI2 proxies, DI3 proxies show their limitations in many aspects. For instance, it is a practical solution for a ML host to publish the IP address of its proxy in its DNS AAAA RR so that the packets for the host will be directly forwarded to the proxy. Therefore, when a ML host served by a DI3 Zhang Xu and Jiankang. Expires November 24, 2010 [Page 6] Internet-Draft Investigation in HIP Proxies May 2010 proxy attempts to communicate with two remote ML hosts served by a same HIP proxy, it is difficult for the host to distinguish one remote host from the other as they both use the same IP address. In addition, DI3 proxies cannot work properly in the circumstance where HIP hosts renumber their IP addresses during the communication due to, e.g., mobility or re-homing. For DI1 or DI2 proxies, these issues can be largely mitigated as the IP addresses of HIP hosts will never be used by DI1 or DI2 proxies to identity hosts. Moreover, it is difficult for DI3 proxies to advertise routing information to attract the packets they needs to process (i.e., the packets transported between ML hosts and remote HIP hosts). Consequently, they can be only deployed at the borders of private networks. DI1 (or DI2) proxies, however, can attract the packets for HIP hosts to themselves by advertising associated routes, because the packets destined to HIP hosts use HITs (or the IP addresses selected from pools) as their destination addresses. Hence, they can locate inside the networks. Therefore, in a private network whose resolver are located inside, a DI1 or a DI2 proxy can be deployed on the path between the resolver and legacy hosts, and it needs only to intercept, modify and forward the queries from legacy host to the resolver. Compared with deploying HIP proxies at the border of the network, this deploying solution can reduce the overhead on the proxy imposed by intercepting DNS lookups. 3.4. N-DI Proxies Unlike DI proxies, an N-DI proxy does not attempt to find out the HIP hosts a legacy host may contact in advance by intercepting DNS lookups transported between legacy hosts (or resolvers) and DNS servers. Instead, it identifies whether the receiver of a packet is HIP enabled when it capture the packet. Typically, an N-DI proxy achieves this by inspecting the destination address of the packet. When the HIP hosts that ML hosts intend to contact are predicable and the number of the HIP hosts is finite, an N-DI proxy can maintain a list of mapping information between HITs and IP addresses[SAL05]. After intercepting a packet from a legacy host, the proxy can ensure the packet is for a HIP host, if the destination address of the packet is maintained in the list. Obviously, this solution is infeasible in the circumstances where it is difficult to identify the HIP hosts that ML hosts intend to contact in advance. On such occasions, an N-DI proxy has to find out whether a packet from a ML host is destined to a HIP host and or a legacy host. It is infeasible for an N-DI proxy to consult resolution systems to find out whether an IP address belongs to a HIP host or a legacy host. Therefore, the N-DI proxy has to maintain Zhang Xu and Jiankang. Expires November 24, 2010 [Page 7] Internet-Draft Investigation in HIP Proxies May 2010 to list of IP addresses. One is for HIP hosts, and the other is for legacy hosts. When intercepting a packet, the N-DI can compare the destination address of the packet against the addresses in the lists to find out whether the packet is destined to a HIP host. If no address is matched, the proxy has to consult resolution systems and maintain the address in the associated list according the answer from resolution systems. Obviously, an N-DI proxy may have to maintain a large amount of state information, which makes it less efficient and scaleable than DI-proxies. Fortunately, this issue can be mitigated by inserting HITs instead of IPv6 addresses in the AAAA RRs of HIP hosts in DNS servers. When receiving an answer containing the AAAA RR of a HIP host (e.g., host B), a legacy host (e.g., host A) will regard the HIT in the answer as the IPv6 address of B. Afterwards, when A sends a data packet to B, it use the HIT of B as the destination address. Because HITs share a prefix which is different from those of ordinary IP addresses, when an N-DI proxy (e.g., proxy P) catches the packet, P can easily distinguish it from the packets for legacy hosts. In addition, P can advertise a route of the prefix of HITs within the private network so as to avoid dealing with the packets transported between legacy hosts. After processing the packet, P may need to get the associated IP address from resolution servers which provide ID to locator mapping information (e.g., DHT servers), using the HIT found in the packet header. Otherwise, P can try to send the packet to an overlay which supports HIT-based routing in the public network (e.g., HIP Bone). Compared with DI proxies, N-DI proxies can be deployed in a more flexible way. For instance, in order to facilitate the legacy hosts in the private networks without HIP proxies to communicate with HIP hosts, Internet services providers (ISPs) may deploy HIP proxies in transit networks. If DI proxies are adopted, they need to locate in the places where they can intercept all the packets transported inside the transit network to find out DNS lookups because the IP addresses of DNS servers are normally unknown in advance. The jobs of processing packets are cumbersome. In addition, such locations may be quite difficult to find out. In this case, N-DI proxies show their advantages; an N-DI proxy can advertise a route of the HIT prefix (or a sub-prefix of HIT) in the transit network and easily attract the desired packets to it. Therefore, they can be deployed in a more flexible way and have to process fewer packets. However, there is a realistic problem which may prevent N-DI proxies from being widely employed. It is predicable that, in the initial period of widely deploying HIP hosts, various HIP proxy solutions will be adopted by different organizations and the information of HIP hosts in DNS servers will organized in an ad hoc way. At least in this period, it is extremely difficult to guarantee that all the RRs of Zhang Xu and Jiankang. Expires November 24, 2010 [Page 8] Internet-Draft Investigation in HIP Proxies May 2010 HIP hosts are modified appropriately. This issue makes it difficult for N-DI proxies to effectively distinguish packets for HIP hosts from those for legacy packets. From this perspective, the capability of DI proxies in modifying DNS answers is desirable. 4. Issues with LBMs in Supporting ML Hosts to Initiate Communication If there is only a single HIP proxy deployed for a private network, the proxy may become the cause of a single-point-of-failure. In addition, when the number of the users increases, the overhead imposed on the proxy may overwhelm its capability, which makes it the bottleneck of the whole mechanism. A typical solution to mitigate this issue is to organize multiple proxies to construct a LBM. By sharing overheads amongst multiple HIP proxies, a LBM can be more scalable and capable to tolerate the failures of a sub-set of HIP proxies. However, a LBM is not just a collection of multiple HIP proxies. Lots of issues need to be carefully considered. Generally, there are two solutions to share communication between ML hosts and HIP hosts among different HIP proxies. The first solution is to divided the ML hosts in the private network into different groups (e.g., according to their IP addresses), and the ML hosts in different sections are taken in charge of by different HIP proxies. The second solution is to divide the HIP hosts in the Internet into multiple groups (e.g., according to their HITs or IP addresses), every HIP proxy serves all the ML hosts in the private network but only take in charge of the packets to and from the HIP hosts in a group. Abstractly, the two solutions are identical. However, the first solution actually attempts to divide a private network into multiple sub-networks, and each of them is served by a HIP proxy. This may introduce additional modification to the topology of the private network, which is not desired in many cases. Therefore, in the design of existing LBM solutions, the second type of solution is more preferred. In the remainder of this document, we mainly consider the second one. 4.1. An Issues Caused by Intercepting DNS Lookups Zhang Xu and Jiankang. Expires November 24, 2010 [Page 9] Internet-Draft Investigation in HIP Proxies May 2010 +--------------------+ +------------------+ | | | | | +---+-------+ | | | +-----------+ |HIP proxy 1+---+ +---------+ | | |Legacy Host| +---+-------+ | |HIP Host | | | +-----------+ | . | | (HH1) | | | | . | +---------+ | | +---+--------+ | | | |HIP proxy n +--+ | |Private Network +---+--------+ | Public Network | | | | | +--------------------+ +------------------+ Figure 1: An example of LBM Figure 1 illustrates a simple LBM. In this mechanism, n proxies are deployed at the border of a private network. If such proxies are DI1 proxies, in order to share the overheads of processing data packets, each proxy needs to advertise a route of the HIT section it takes in charge of. In addition, each proxy also needs to advise a route of a section of IP addresses (or a default route for the entire IP address namespace) inside the private network to intercept DNS lookups. A problem occurs when the DNS lookups and the data packets sent by a legacy host are intercepted by different proxies. In such a case, the proxy intercepting a data packet will lack essential knowledge to correctly process it. If the proxies adopted in Figure 1 are DI3 proxies, then each proxy only needs to advertise a route of a section of IP addresses which is adopted to intercept both DNS lookups and data packets. On this occasion, if a HIP host and the DNS server maintaining its RR fall into two different IP sections, the DI3 proxy intercepting the lookups for the HIP host will not be the one intercepting subsequent data packets. Therefore, DI3-proxy- based LBMs also suffer from a same problem with DI1-proxy-based LBMs. A candidate solution to the problem is to propagate the mapping information obtained from DNS lookups amongst HIP proxies. Therefore, after intercepting a DNS conversation, a proxy can forward the gained information to the proxy expected to process the subsequent data packets. Alternatively, a proxy can attempt to collect required information from resolution systems after intercepting a data packet. This approach, however, imposes addition overheads to DI-proxies in communicating with resolution servers. If the proxies in Figure 1 are DI2 proxies, the problem can be eliminated. In a DI2-proxy-based LBM, each DI2 proxy needs to advertise two routes, one of the IP addresses in the pool and one of Zhang Xu and Jiankang. Expires November 24, 2010 [Page 10] Internet-Draft Investigation in HIP Proxies May 2010 a section of IP addresses for intercepting DNS lookups. After intercepting a DNS lookup, a DI2 proxy will return an IP address within the pool in the answer to the requester (a ML host or a resolver), which can ensure the subsequent data packets will be transported to the same proxy. 4.2. Issues with LBMs in Capturing and Processing Replies from HIP hosts Theoretically, when representing a ML host to communicate with a HIP host in the public network, a HIP porxy can use either an IP address it possesses or the IP address of the ML host as the source address of the packets forwarded to the HIP host. However, in practice, the succeeding option may cause several issues. For instance, in the succeeding option, a Hip proxy must be placed on the path of the packets transferred between HIP hosts and the ML hosts it serves in order to capture the reply packets from HIP hosts. In addition, the succeeding solution may cause problems in the load balancing scenarios where multiple HIP proxies provide services for a same group of ML hosts. Assume there are two HIP proxies located at the border of a private network. If the proxies adopt the succeeding solution, they need to advertise the routes of the ML hosts in the public network respectively. As a result, it is difficult to guarantee the packets transported between a legacy host and a HIP host are stuck to a same HIP proxy, and thus after a proxy intercepts a packet it may lack the proper HIP association to process it. A possible solution to address this problem is to share HIP state information (e.g., HIP associations, sequence number of IPsec packets) amongst the related HIP proxies in a real-in-time fashion. However, during communication, some context information such as the sequence numbers of IPsec packets can change very fast. It is infeasible to synchronize the IPsec message counters for every transmitted or received IPsec packet, since such operations will occupy large amounts of bandwidth and seriously affect the performances of HIP proxies. [Nir 2009] indicates that this issue can be partially mitigated by synchronizing IPsec message counters only at regular intervals, for instance, every 10,000 packets. An issue similar with the one mentioned above is discussed in [TSC05], and an extended HIP base exchange is proposed. But the proposed solution only tries to help HIP-aware middle boxes obtain the SPIs used in a HIP base exchange and cannot be directly used to address the issue mentioned above. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 11] Internet-Draft Investigation in HIP Proxies May 2010 Note that all these issues can be simply addressed by adopting the preceding option because it can guarantee the packets transported between a ML host and a HIP host are intercepted by a same proxy. Therefore, in the following discussions, without mentioned otherwise we assume that a HIP proxy uses one of its IP addresses as the source IP address of a packet which it sends to a HIP host. 5. Issues with LBMs which also Support HIP Hosts to Initiate Communication Apart from the basic functions (i.e., supporting ML hosts to communicate with HIP hosts), in many typical scenarios, HIP proxies may also need to facilitate the communication initiated by HIP hosts. In this section, we attempt to analyze the issues that a HIP proxy has to face in the conditions where HIP hosts proactively initiate communication with legacy hosts. In order to support the communication initiated by HIP hosts, the HIP proxies of a private network should have the knowledge essential to represent the ML hosts to perform HIP BEXs. Such knowledge consists of the IP addresses of the legacy hosts, their pre-assigned HITs, the corresponding HI key pairs, and any other necessary information. In addition, such information of the ML hosts should be advertised in resolution systems (e.g., DNS and DHT) as HIP hosts. Otherwise, a HIP host has to obtain the HIT of the ML host in the opportunistic model which, however, should only be adopted in secure environments. 5.1. DNS Resource Records for ML Hosts In practice, the AAAA RR of a ML host can consist of either the IP address of the ML host or the address of its HIP proxy. In the preceding case, the packets destined to a legacy host are transported to the host directly, and thus HIP proxies must be located on the path of such packets to intercept them. In the succeeding case, the packets for a legacy host are firstly transported to the associated HIP proxy. Therefore the proxy can be deployed anywhere desired. In addition, the succeeding approach is more efficient than the preceding one in private networks where legacy hosts and HIP hosts are deployed in an intermixed way, since the HIP proxy will not have to process the packets transported between HIP hosts. However, the succeeding approach may cause problems in the process of packets sent by legacy hosts in the public network. Normally, a HIP proxy needs to serve a number of ML hosts. When using the succeeding approach, the packets destined to these ML hosts will have a same destination address (i.e., the IP Zhang Xu and Jiankang. Expires November 24, 2010 [Page 12] Internet-Draft Investigation in HIP Proxies May 2010 address of the proxy). Therefore, when receiving a packet from a legacy host located in the public network, the proxy may find it difficult to identify the ML host which the packet should be forwarded to. A simple approach which combines the advantages of the above two solutions but avoids their disadvantages is to extend the RDATA field in HIP RRs [RFC5205] with a new proxy field, which contains the IP address of a HIP proxy. In the extended HIP RR of a ML host, the proxy field consists of the IP address of its HIP proxy, while the proxy field in the RR of an ordinary HIP host is left empty. Therefore, a HIP host intending to communicate with the ML host can obtain the IP address of the proxy from DNS servers and set it as the destination address of the packets. The packets are then routed to the proxy. When a non-HIP host intends to communicate with the legacy host, it can obtain the IP address of the legacy host from the AAAA RR as usual and set it as the destination address of the packets; the packets are then transported to legacy host directly. Although it is also possible to use the RVS field in a HIP RR to transport the information of a HIP proxy, a special proxy field can bring additional benefits in security. For instance, it is normally assumed that the base-exchange protocol is able to establish a security channel for the hosts on the both sides of communication to securely exchange messages. However, this presumption may be no longer valid in the presence of HIP proxies, as the messages between legacy hosts and proxies can be transported in plain text. With the Proxy field, it is easy to distinguish the legacy hosts made up by HIP proxies from the ordinary HIP hosts. Therefore, a HIP host can assess the risks of exchanging sensitive information with its communicating peers in a more specific way. 5.2. An Asymmetric Path Issue In a load balancing scenario where multiple HIP proxies are deployed at the border of a private network, the packets transported between a legacy host and a HIP host may be routed via different HIP proxies. Therefore, when a packet is intercepted by a HIP proxy, the proxy may find that it lacks essential knowledge to appropriately process the packet. Hence, an asymmetric path issue occurs. In order to explain the asymmetric path issue in more detail, let us revisit the LBM illustrated in Figure 1. In addition, assume that the HIP proxies are DI1 proxies and their IP addresses are maintained in the DNS RRs of the ML hosts. When a HIP host (e.g., HH1) looks up a legacy host at a DNS server, the DNS server returns Zhang Xu and Jiankang. Expires November 24, 2010 [Page 13] Internet-Draft Investigation in HIP Proxies May 2010 the IP addresses of all the HIP proxies in an answer (see Figure 2). Upon receiving the answer, HH1 needs to select an IP address and sends an I1 packet to the associated HIP proxy. Assume the HIP proxy 1 is selected. Then after a base exchange, HIP proxy1 and HH1 establish a HIP association respectively. Upon receiving the first data packet from HH1, the HIP proxy uses the HIP association to de- capsulate the packet and forwards it to the legacy host. In the forwarded packets, the HIT of HH1 is adopted as the source IP address, and thus the HIT of HHI is adopted as the destination address in the reply packets sent by the legacy host. Assume that the HIT of HH1 is within the section managed by HIP proxy n. According the routes advertised by the proxy n, the packet is forwarded to the HIP proxy n which, however, does not have the corresponding HIP association to deal with the packet. Similarly with DI1 proxies, DI3 proxies and N-DI proxies also suffer from the asymmetric path issue in the load balancing scenarios, since they cannot guarantee the data packets which are transported between a legacy host and a HIP host stick to a single HIP proxy too. +----------------------+ +--------------------------+ | | | | | +---+-------+ | (3) | | (4) -|HIP proxy 1+-+<- | | / +---+-------+ | \ +--------+ (1)+------+| |+-----------+< - | . | -|HIP Host|--> | DNS || ||Legacy Host|- | . | | (HH1) |<-- |Server|| |+-----------+ \ +---+-------+ | +--------+(2) +------+| | (5) - >|HIP proxy n+-+ | | Private Network +---+-------+ | Public Network | | | | | +----------------------+ +--------------------------+ Figure 2. An example of the asymmetric path issue As we mentioned in section 3.3.1, the approach of sharing HIP associations and IPsec association amongst HIP proxies can be used to address this issue. However, this issue will introduce addition communication overhead on HIP proxies. Here, we discuss several other alternative solutions. The simplest solution is to allow a HIP proxy to discard the I1 packets it receives if they are not original from HIP hosts which the proxy takes in charge of. In addition, the proxy can inform the senders of the incidents using ICMP packets. Therefore, after waiting for a certain period or upon receiving a ICMP packet, a HIP host will try to select another HIP proxy from the list in the DNS Zhang Xu and Jiankang. Expires November 24, 2010 [Page 14] Internet-Draft Investigation in HIP Proxies May 2010 answer and send an I1 packet it. In the worst case, this process needs to be recursive until all the HIP proxies in the list have been contacted. Because a HIP host may have to send the multiple I1 packets in order to communicate with a ML host, this solution may yield a long delay. Note that in some DNS based load balancing approaches, a DNS server only returns one HIP proxy in an answer. On such occasions, HIP hosts have to communicate with DNS servers repeatedly, and the negative influence caused by the communication delay can be even exacerbated. A solution which is able to avoid the delay issue is to endow DNS servers with the capability of returning the IP address of an appropriate HIP proxy in an answer according to the HIT (if the proxy is a DI1 proxy or a N-DI proxy) or the IP address (if the proxy is a DI3 proxy) of a requester. That is, the HIP proxy described in a DNS answer should take in charge of the namespace section which the requester belongs to. In order to achieve this, DNS servers need to 1) maintain the information about the sections of the namespaces that HIP proxies take in charge of, 2) locate the appropriate HIP proxy according to the HIT or the IP address of a HIP requester. These requirements result in modifications to current DNS servers in the implementation of the DNS server applications and the conversation protocols between requesters and DNS servers. For instance, a HIP host may need to transport its HIT in DNS requests in order to help DNS servers locate an appropriate HIP proxy. It is also possible to register HIP proxies to a RVS server. Therefore, upon receiving an I1 packet, the RVS server can forward it to a proper proxy to process. The asymmetric path issue can be eliminated by adopting DI2 proxies. A DI2 proxy located at the border of a private network maintains a pool of IP addresses which are routable in the private network. After receiving a packet from a HIP host, the DI2 proxy processes the packet and forwards it to the destination legacy host. In addition, an IP address selected from the pool is adopted as the source address of the packet. Therefore, when the legacy host sends responding packets to the HIP host, the packets will be transported to the same HIP proxy. The asymmetric path issue is thus eliminated. 6. Issues with LBMs in supporting dynamic load balance and redundancy The load balancing solutions discussed above are simple and static. They cannot modify routes of packets according to the loads on different HIP proxies. In practice, there are requirements for LBMs to support dynamic load balance and redundancy. That is, when a Zhang Xu and Jiankang. Expires November 24, 2010 [Page 15] Internet-Draft Investigation in HIP Proxies May 2010 proxy (called a prim proxy) in a LBM is not able to work properly or the overheads imposed on it surpass a threshold, the proxy can delegate all of (or a part of) its job to other proxies (called backup proxies). In this section, we analyze the performance of different types of HIP proxies in supporting dynamic load balance and redundancy. In order to provide backup services, a backup proxy needs to advertise the same routes as those advertised by the prim proxy in both the private and the public networks. To avoid affecting the normal operations of the prim proxy, the routes advertised by the backup proxy have a much higher cost than that of the routes advertised by the prim proxy. When the abnormal conditions mentioned above occurs, the prim proxy can withdraw the routes it previously advertised so that the packets supposed to be processed by the prim proxy will be forwarded to the backup proxy. We refer to the routes advertised by a proxy for backup purposes as the backup routes of the proxy. In contrast, we refer to the routes advertised by a proxy to achieve its primary job as the prim routes of the proxy. In practice, the proxies in a LBM can provide backup services for one another. Therefore, a proxy in such a LBM may needs to advertise both prim and backup routes. The synchronization of state information between prim and backup proxies is also very important. Without proper HIP associations, a backup proxy cannot correctly take place of the prim proxy to process the packets. The state synchronization problem has been discussed above. If there is no state synchronization, a backup proxy may select to send signaling packets to HIP hosts to initiate new HIP BEXs. In the remainder of this section, we attempt to analyze the performance of different types of HIP proxies in supporting dynamic load balance and redundancy respectively. 6.1. Application of DI1 proxies in supporting dynamic load balance and redundancy As mentioned in section 3.1, a DI1 proxy needs to at least advertise two prim routes in the private network, one for a section of HITs, which is used to intercept data packets, and the other for a section of IP addresses, which is used to intercept DNS lookups. When the proxy cannot work properly, it can withdraw both routes to enable a backup proxy to take over its job. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 16] Internet-Draft Investigation in HIP Proxies May 2010 In some cases, a DI1 proxy may only want to delegate a part of its job to others so as to reduce the overloads it undertakes. To achieve this objective, the proxy can advertise multiple specific prim routes. When the overload on the proxy is high, it can only withdraw a subset of those advertised routes. For instance, a DI1 proxy can selectively only delegate a part of the responsibility in processing DNS lookups to a backup proxy by withdrawing one of its lookup intercepting routes. 6.2. Application of DI2 proxies in supporting dynamic load balance and redundancy A DI2 proxy needs to at least advertise two prim routes in the private network, a route for its IP address pool, used to intercept data packets, and the other for an IP address section, is used to intercept DNS lookups. When the proxy cannot work properly, it can withdraw both routes to enable a backup proxy to take over its job. In this case, the delegated backup proxy needs to maintain an IP address pool identical to the one maintained by the prim proxy. Moreover, apart from synchronizing HIP associations, the synchronization of mappings from IP addresses to HITs is also required. Otherwise, the backup proxy cannot translate the received packet correctly. If a DI2 proxy only intends to maintain existing communications between ML hosts and HIP hosts while not facilitating any more, it can withdraw the lookup intercepting route. As mentioned previously, DI2 proxies have the capability to stick the DNS lookups and the subsequent data packets to the same proxy. Therefore, the backup proxy can intercept DNS lookups as well as process the subsequent communications. 6.3. Application of DI3 proxies in supporting dynamic load balance and redundancy Unlike DI1 and DI2 proxies, the routes advertised by a DI3 proxy are used for intercepting both DNS lookups and data packets. Therefore, before a DI3 proxy withdraws a route, it needs to synchronize the states of the on-going communications affected by the routing adjustment to its backup proxies. 7. Security Consideration Security is an important benefit introduced by HIP. In the basic HIP architecture, security requirement on DNS communications is not compelled. But in practice, DNSSEC [RFC4305] is recommended in order Zhang Xu and Jiankang. Expires November 24, 2010 [Page 17] Internet-Draft Investigation in HIP Proxies May 2010 to prevent attackers from tampering or forging DNS lookups between resolvers and DNS server. This solution may affect the deployment of HIP proxies. For instance, DI1 and DI2 proxies need to modify the contents of NDS answers, and thus they should be only deployed on the path between legacy hosts and their resolvers. Therefore, a DI1 proxy (or a DI2 proxy) should not be deployed in the middle of DNSsec-enabled resolvers and authoritative DNS servers. When sharing HIP state information amongst HIP proxies, the integrity and confidentiality of the state information should be protected. The discussion about the similar issues can be found in [Nir 2009] and [Narayanan 07]. 8. Conclusions This document mainly analyzes and compares the performance of different kinds of HIP proxies in LBMs. Amongst the HIP proxies discussed in the document, DI2 proxies show its advantages in multiple scenarios. In addition, we argue that the state synchronization among HIP proxies is very important to achieve academic load balancing and redundancy. There is a topic which is important but not covered in this document is the compatibility among different HIP proxies. The different types of HIP proxies are designed based on different presumptions. The presumptions of different type of HIP proxies maybe conflict with each other. Then how to make a trade-off and enable different types of proxies work cooperatively is an important issue that the designers of HIP extensible solutions have to consider. 9. IANA Considerations No such considerations. 10. Acknowledgments 11. References Normative references [RFC5338] T. Henderson, P. Nikander and M. Komu, "Using the Host Identity Protocol with Legacy Applications," RFC 5338, Sep. 2008. [RFC5205] P. Nikander and J. Laganier, "Host Identity Protocol (HIP) Domain Name System (DNS) Extension," RFC 5205, April 2008. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 18] Internet-Draft Investigation in HIP Proxies May 2010 [RFC4035] R. Arends, R. Austein , M. Larson, D. Massey and S. Rose, "Protocol Modifications for the DNS Security Extensions," RFC 4035, March 2005. Informative references [Nir 2009] Y. Nir, "IPsec High Availability Problem Statement," Internet Draft, 2009. [Narayanan 07] V. Narayanan, "IPsec Gateway Failover and Redundancy - Problem Statement and Goals," Internet Draft, 2007. [PAT07] P. Salmela, J. Wall and P. Jokela, "Addressing Method and Method and Apparatus for Establishing Host Identity Protocol (Hip) Connections Between Legacy and Hip Nodes," US. 20070274312, 2007. [SAL05] P. Salmela, "Host Identity Protocol proxy in a 3G system," Master's thesis. Helsinki University of Technology 2005. [TSC05] H. Tschofenig, A. Gurtov, J. Ylitalo, A. Nagarajan and M. Shanmugam, "Traversing Middleboxes with the Host Identity Protocol," Proc. ACISP, 2005. Zhang Xu and Jiankang. Expires November 24, 2010 [Page 19] Internet-Draft Investigation in HIP Proxies May 2010 Authors' Addresses Dacheng Zhang Huawei Technologies Co.,Ltd KuiKe Building, No.9 Xinxi Rd., Hai-Dian District Beijing, 100085 P.R. China Email: zhangdacheng@huawei.com Xiaohu Xu Huawei Technologies Co.,Ltd KuiKe Building, No.9 Xinxi Rd., Hai-Dian District Beijing, 100085 P.R. China Email: xuxh@huawei.com Jiankang Yao CNNIC No.4 South 4th Street, Zhongguancun, Bejing, P.R. China Phone: +86 10 58813007 Email: yaojk@cnnic.cn Zhang Xu and Jiankang. Expires November 24, 2010 [Page 20]