HOST IDENTITY PROTOCOL PROXY
Patrik Salmela, Jan Melén
Ericsson Research NomadicLab, Hirsalantie 11, 02420 Jorvas, Finland
Keywords: HIP, identifier-locator split, proxy.
Abstract: The Host Identity Protocol (HIP) is one of the more recent designs that challenge the current Internet
architecture. The main features of HIP are security and the identifier-locator split, which solves the problem
of overloading the IP address with two separate tasks. This paper studies the possibility of providing HIP
services to legacy hosts via a HIP proxy. Making a host HIP enabled requires that the IP-stack of the host is
updated to support HIP. From a network administrator's perspective this can be a large obstacle. However,
by providing HIP from a centralized point, a HIP proxy, the transition to begin using HIP can be made
smoother. This and other arguments for a HIP proxy will be presented in this paper along with an analysis
of a prototype HIP proxy and its performance.
1 INTRODUCTION
The current Internet is based on an over 20-year-
old architecture. That architecture has flaws - some
more serious than others. Many of these issues have
been addressed by tools and methods designed to
patch the flaws of the architecture. Examples of
these new designs are e.g. IPv6 that provides a new
larger addresses space in place of the one currently
used, and IPsec (Kent (1), 1998) that provides
security in the insecure network.
One of the more recent designs is the Host
Identity Protocol (Moskowitz (1), 2004). HIP is still
being researched and is not yet a complete product
and thus not being used in a large scale. Considering
that one of the main benefits of using HIP is secured
communication, one can assume that HIP might
appeal more to companies and organizations rather
than the average home computer user. When HIP
will begin to be utilized by a larger user group than
just the developers and some other interested parties,
as it is today, ease of use will be one factor that will
affect how well and wide HIP will spread. Enabling
HIP in a host requires that the host is updated with a
HIP enabled IP-stack. This might be a disadvantage
of HIP when a network administrator is considering
different methods of protecting the communication
to and from the network.
This paper studies the possibility of providing
HIP services to hosts without having to modify
them. Having legacy hosts communicating with HIP
enabled hosts, using HIP, is possible with a HIP
proxy. However, providing HIP to hosts via a proxy,
with the actual HIP implementation residing outside
of the host, puts some restrictions on the network
environment. A HIP proxy scenario is shown in
Figure 1.
The paper is structured as follows; first some
background information will be presented, with the
focus on some of the problems of the current
architecture. Different solutions for these problems
will be presented, including HIP. Next follows a
technical view of how HIP works and the reasoning
for a HIP proxy. After that, the functionality of a
HIP proxy is presented, followed by a look at the
design and performance of a prototype HIP proxy.
Then we look at how the prototype could be further
developed, after which the conclusions are
presented.
Fi
g
ure 1: HIP
p
rox
y
scenario
222
Salmela P. and Melén J. (2005).
HOST IDENTITY PROTOCOL PROXY.
In Proceedings of the Second International Conference on e-Business and Telecommunication Networks, pages 222-230
DOI: 10.5220/0001409002220230
Copyright
c
SciTePress
2 BACKGROUND
Changing the current Internet architecture is a quite
hot topic, and it has been that for some years
already. The topic has been discussed in various
papers, including the New Arch paper (Braden,
2000) and the Plutarch paper (Crowcroft, 2003).
There are many issues with the current architecture
that have helped to recognize the need for a change.
Maybe the most recognized issues include the lack
of support for security by the IP protocol, address
space depletion, the heavy load on routers and the
overloading of the IP address to serve as both
identifier and locator. Additionally, mobile hosts are
becoming more common which adds demand for an
always better mobility solution.
To some of the aforementioned problems there
are already working solutions; users who want
security can utilize one of the many available
security solutions e.g. IPsec, PGP, SSH or TLS. The
utilization of the IPv4 address space has been
improved with the help of Classless Inter-Domain
Routing (CIDR). Also mobility is possible in the
current Internet. Routers are heavily burdened
because the size of the IPv4 address does not allow
for much address aggregation. IPv6, with its four
times bigger address size compared to IPv4, will
improve the possibility for address aggregation.
However, there is still no widely deployed method
that provides an identifier-locator split.
2.1 Why do we need a change
So what is the big deal with using the IP address
both as an identifier and a locator? The problem can
be spotted by examining how the IP address behaves
when a host is changing its topological position in a
network, while remembering what qualities are
necessary for an identifier and a locator respectively.
Consider a host with the IP address IP
A
. The locator
of the host, i.e. the information used to route packets
to the host, is the IP address IP
A
. The same
information is used to identify the host. If the host
moves to another topological position the host has to
change its address to the new address IP
A
'. When a
host now wants to send packets to this host the new
IP address, IP
A
', is used to route the packets to the
host. This means that the locator has changed to
match the current location of the host, which is
exactly how a locator should function. However,
since the IP address serves as both an identifier and
a locator the host has now been assigned a new
identifier. This change is not welcome since having
an identifier that can change frequently makes the
identifier useless except for the short timeframe that
it stays constant. A true identifier should stay
constant, if not forever, at least for a very long time,
in the range of years.
Because the notion of an identifier is used in the
Internet, it should also fill the requirements set for
an identifier. Namely that it is constant and uniquely
identifies a host regardless of where in the network
the host is located. This makes the IP address an
unfit candidate for an identifier. What is needed is
another coexistent address space, actually an
``identifier space'', from which hosts are assigned an
identity. Another possibility could be something
along the lines of what was suggested in the GSE
proposal (Crawford, 1999); part of the IP address is
used for identifying the host while the rest is used as
a locator for the host. In this case the identifier part
has to stay constant when the host moves in the
network and updates the locator part to match the
current location of the host.
2.2 The HIP solution
The Host Identity Protocol is one of the new designs
that, amongst other things, target the identifier-
locator split. In addition, HIP also provides security,
mobility and multi-homing. All the features
provided by HIP are based on the solution for the
identifier-locator split.
HIP separates the identifier from the locator by
introducing a new name space for identifiers. The
entities in that set are called Host Identities (HI) and
are of variable length. A HI is the public key of an
asymmetric key-pair, which is used to provide
security in HIP. Because the HIs are of variable
length it is difficult to use them as such in HIP, so
instead a 128-bit hash over the HI, called a Host
Identity Tag (HIT), is used. When operating in an
IPv4 network a 32-bit hash over the HI, a Local
Scope Identifier (LSI), is used. Because of its length,
the LSI cannot be considered to be globally unique.
When a HIP enabled host sends a packet to another
HIP enabled host the packet is sent to a HIT, or an
LSI respectively, but the packet is transported using
the locator i.e. the IP address.
The use of HITs and LSIs is made possible by
introducing a new layer to the IP-stack. The HIP-
layer finds its place between the internetworking
layer and the transport layer, and is sometimes
referred to as layer 3,5. At the layers above the HIP-
HOST IDENTITY PROTOCOL PROXY
223
layer HITs, or LSIs, are used instead of IP addresses.
At the HIP-layer a translation takes place; from
HITs or LSIs to IPv6 or IPv4 addresses, or vice
versa. In the remaining layers the IP addresses are
used. Using HIP, the Host Identifier (HIT or LSI) of
a host is always constant as it should be, and the
locator can change when the peer moves to another
position.
2.3 Other similar solutions
HIP is one of the more complete solutions that
provide the identifier-locator split. However, there
are also some other proposals that target the same
problem. In this subsection three other solutions will
be presented: the Forwarding directive, Association,
and Rendezvous Architecture (FARA) (Clark,
2003), PeerNet (Eriksson, 2003) and the Internet
Indirection Infrastructure (I
3
) (Stoica, 2002).
FARA is a framework that can be used when
designing a new architecture. The FARA model is
divided into two layers; the upper layer contains the
communicating entities and the communication
endpoints, the lower layer handles the packet
forwarding. The communication link between two
entities is stateful and is called an Association. Each
Association is identified by a locally unique
Association ID (AId). When an entity moves its
AIds stay constant while the information used to
forward the packets to the entity changes. It is easy
to draw some parallels between this and how HIP
uses constant HIs while the IP address can change to
reflect the current position. In the FARA paper
(Clark, 2003) HIP is actually suggested as
something that could be used in a FARA
architecture.
PeerNet is based on peer-to-peer thinking. The
hosts are located as leafs in a binary tree, with the
path from the root presenting the address of the host.
When a new host attaches to the network it asks one
of the hosts in its vicinity for an address. The asked
host splits its address space into two and assigns one
of them to the new node and keeps the other for
itself. The hosts also have an identity that stays
constant regardless of node movements. PeerNet
uses distributed peer-to-peer routing with each host
storing some routing information, i.e. identity-to-
address mappings. PeerNet is not a ready solution, it
does have the identifier-locator split, but security
issues have not been addressed.
The I
3
design introduces some new elements to
the network, the I
3
servers. To be able to receive
packets hosts have to register their identity and
current locator into an I
3
server. This is called
inserting a trigger. The trigger has a limited lifetime
and thus it has to be updated periodically by the host
if it wishes to continue to receive packets via it. In I
3
packets are sent to identities and the sent packet
searches the I
3
servers for a trigger that matches the
destination identity. Once a match is found the
destination of the packet is changed for the IP
address found in the trigger. By updating the trigger
I
3
supports mobility, and by letting multiple hosts
register with the same identity a multicast property
is achieved. But just as PeerNet, I
3
is not a complete
solution. The biggest concern of I
3
is the lack of
security. To provide security for I
3
a combination of
HIP and I
3
, called Hi3, is being researched
(Nikander, 2004).
2.4 Problems with having a new
architecture
Even if these new designs might sound very good,
creating them is only part of the job, getting the
design deployed is also a big challenge. Deploying a
new architecture is not the same as deploying a new
standalone, e.g. security solution. Deploying the
design in a small test network which one has full
control over is easy, but when the target is a global
public network, the Internet, there is not really any
good way to get it done. The problem of deploying
e.g. HIP is similar to getting IPv6 deployed globally.
An ideal solution would be to get all of Internet
updated by the flick of a switch, moving from an all
IPv4 network to an all IPv6 network in a neglectable
time interval. However, this is not possible, not for
IPv6 nor HIP. An update of this proportion will
proceed incrementally, requiring some sort of
compatibility between the new and the old
architecture. Deploying HIP is not as difficult as the
IPv6 problem since HIP enabled hosts can still
communicate with legacy hosts using regular IP.
However, to truly benefit from all the features of
HIP, it would be desirable that as many hosts as
possible were HIP enabled.
3 HIP
To enable HIP in a host the IP-stack of the host has
to be updated to a HIP modified one. An asymmetric
key-pair has to be generated and the public key will
serve as the identity of the host, with hashes of the
ICETE 2005 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
224
key resulting in HITs and LSIs. To initiate a HIP
connection with another HIP enabled host the HIT
of the peer has to be obtained. This can be done
from a HIP modified DNS or other similar lookup
service.
The creation of a HIP connection between two
HIP enabled hosts is called the HIP base exchange
(Moskowitz (2), 2004) and it is depicted in Figure 2.
When the Initiator wants to establish a connection it
sends an I1 packet to the Responder. The packet
contains the HIT of the Initiator (HIT
I
), and if the
HIT of the Responder (HIT
R
) has been obtained it is
also included in the message. If the Initiator does not
know HIT
R
it is set to NULL in the I1 packet. This is
called opportunistic mode HIP. The I1 packet is
actually just an initiation message for the
connection.
Figure 2: The HIP base exchange
The Responder responds with an R1 packet
which contains the HITs used in the I1 packet, the
HI of the Responder and a challenge. If the Initiator
is attempting opportunistic mode HIP the Responder
has now added its HIT to the packet instead of the
received NULL HIT. The R1 packet also initiates
the Diffie-Hellman (Rescorla, 1999) exchange and
gives the preferences of the Responder in respect of
which IP Encapsulating Security Payload (ESP)
(Kent (2), 1998) mode to use. The supported
integrity and encryption algorithms are also
presented. The challenge in the packet is a puzzle
that the Initiator has to solve to prove that it is
serious about creating a connection. The Responder
can have in advance prepared R1 packets to ease its
load, while the puzzle requires the Initiator to do
heavy calculations. This makes connection initiation
expensive and is thus a form of Denial of Service
(DoS) protection.
When the Initiator has solved the puzzle it sends
an I2 packet to the Responder. The packet again
contains the two HITs and now also the solution to
the puzzle. Also the HI of the Initiator is included, it
is encrypted using the selected algorithms and
generated keys. Based on the information that the
Responder receives in the packet it can decrypt the
HI. The Responder also receives the Security
Parameter Index (SPI) to use when sending packets
to the Initiator.
The last packet of the HIP base exchange, the R2
packet sent to the Initiator, contains the SPI that the
Initiator should use along with the two HITs.
Similar to all but the first packet of the base
exchange, the R2 packet contains a digital signature,
and in addition a HMAC (Krawczyk, 1997)
calculated over the packet. Besides that, also other
consistency checks are done on each packet,
including checking that the received HITs are the
correct ones. The result of the HIP base exchange is
a pair of IPsec ESP security associations (SA). After
the base exchange all traffic between the Initiator
and the Responder is ESP protected.
The four packets used during the base exchange
(I1, R1, I2, R2) are HIP specific packets. Apart from
these packets there are also some other HIP specific
packets of which the Update packet is the most
important one. The Update packet is used for
signaling rekeying when the old SA needs to be
replaced, e.g. if the ESP sequence number is getting
too big. The Update packet is also used for handling
location updates by sending location update
messages.
The security provided by HIP is basically very
similar to IPsec without IKE. The HI of a host, and
the corresponding private key, are used for
authentication purposes and for negotiating security
parameters and SAs. The SAs are established
between two HITs, so when sending a packet the SA
is located based on the HITs found in the outgoing
packet. When receiving a packet the SA is located
based on the SPI, and the HITs for the connection
are found from the SA.
4 WHY A HIP PROXY
The difficulty of deploying a new architecture was
mentioned earlier; all hosts in a global network
cannot simultaneously be update to support a new
architecture, the migration to a new architecture will
take time. HIP does not need to spread to all hosts in
HOST IDENTITY PROTOCOL PROXY
225
the Internet, and it probably never will, but the wider
it spreads the more useful HIP is for its users. A HIP
proxy that makes it possible for a HIP host to
communicate with a legacy host, using HIP between
the HIP host and the HIP proxy, could help to
promote HIP. The more possibilities there are for
using HIP the more appeal it will have. The problem
with a HIP proxy is that if it is located in a public
network the security features of HIP are rendered
useless. The connection between the proxy and the
legacy host is not protected in any way. If one would
like, some other form of security could of course be
applied between the HIP proxy and the legacy host.
To be able to benefit from the security
functionality provided by HIP, when using a HIP
proxy, the proxy would have to be situated in a
secure network. One likely scenario might be a
private network, e.g. the internal network of a
company. By having a HIP proxy at the border
between the private network and the Internet, the
users of the private network could contact HIP
enabled hosts in the Internet using HIP. Because the
private network is considered to be secure the only
difference of this scenario, compared to two HIP
enabled hosts communicating with each other, is that
the legacy host cannot take advantage of all the
features provided by HIP, e.g. HIP mobility.
If the hosts of a private network do not need all
the features provided by HIP, a HIP proxy might
even be considered the preferred alternative
compared to enabling HIP in all the hosts. Enabling
HIP in all hosts might be considered to generate too
much work compared to having a HIP proxy
solution. With a static network configuration the
work estimates might actually be correct. However,
most networks are not static, and having a HIP
proxy in a dynamic network will generate excess
work in the form of keeping the proxy
configurations up-to-date. A HIP proxy is not the
preferred solution but it is well suited as a stepping-
stone when going from an all legacy network to an
all HIP network.
5 THE HIP PROXY PROTOTYPE
As a proof of concept a HIP proxy prototype has
been implemented. The implementation was done
for FreeBSD 5.2, and tested with the HIP
implementation developed at Ericsson Finland
(http://hip4inter.net). Besides implementing the HIP
proxy application also the kernel of FreeBSD had to
be modified; a new feature, divert sockets for IPv6,
had to be implemented. To perform its task the
proxy utilizes divert sockets and the firewalls (ipfw
and ip6fw) of FreeBSD. The network environment
where the proxy operates is between two small
LANs, one acting as a private network containing
the legacy hosts and the other acting as the Internet
containing the HIP enabled hosts. If the proxy was
to function in one network in which there are both
kinds of hosts the legacy hosts would have to be
configured to route all their packets via the HIP
proxy.
5.1 Functionality of a HIP proxy
When looking at the HIP proxy as a host in the
network its task is to serve as the endpoint for HIP
associations between itself and HIP enabled hosts.
HIP hosts connected via it believe that they are
communicating with the legacy host using HIP while
the legacy hosts believe that they are communicating
with the HIP host using plain IP. For the
communicating endpoints the HIP proxy is invisible.
The proxy itself can be seen as a host that performs
translation between the two communication formats;
plain IP and HIP.
When a legacy host wishes to communicate with
one of the HIP enabled hosts it queries DNS for the
IP address of the peer. The query travels through the
HIP proxy and on to a HIP modified DNS in the
Internet. The reply contains the IP address and the
HIT of the HIP host. When the reply passes the
proxy it caches the IP-HIT mapping for future use
when it possibly has to initiate a HIP base exchange
with the host. The legacy host receives the IP
address and can now use it to contact the HIP
enabled host. HIP enabled hosts can contact legacy
hosts via the proxy if the IP address of the proxy,
and the HITs assigned to the legacy hosts, are
registered into DNS. Thus a HIP enabled host will
receive an IP address and a HIT, as expected, when
querying the information about one of the legacy
hosts.
When a packet passes through the HIP proxy
host the packet must be diverted from its path and
sent to the HIP proxy application. If the packet is on
its way from a legacy host to a HIP enabled host the
proxy checks if there is an SA available for the
connection. If a matching SA is found the packet is
sent out using the SA. Otherwise the proxy has to
initiate the HIP base exchange to establish SAs for
the connection. Using the IP-HIT mapping it has
ICETE 2005 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
226
gotten from the DNS query, and the IP address of
the legacy host along with the HIT assigned to the
legacy host, the proxy can initiate the base
exchange. When the HIP association has been
established the packet sent by the legacy host can be
sent to the HIP enabled host and the communication
between the peers can begin. When a packet is
received over an SA, from a HIP enabled host, the
proxy decrypts the ESP packet and forwards it as a
plain IP packet with the IP addresses of the peers.
The packet is then sent to the legacy host whose IP
address was found based on the destination HIT. The
connection initiation is depicted in Figure 3.
When a HIP enabled host initiates a connection
to a legacy host it uses the information it has
received from DNS. The HIP host believes that it is
connecting to the legacy host, although the actual
HIP connection is established to the HIP proxy.
When the SAs have been established the HIP host
begins sending packets over them. The HIP proxy
converts the received packets to plain IP packets and
forwards them to the correct legacy host.
5.2 The prototype design
The prototype HIP proxy does not function
exactly as described in the previous section. We did
not have a HIP enabled DNS so the IP-HIT
mappings of both the legacy hosts and the HIP
enabled hosts were added to a configuration file for
the HIP proxy The proxy reads the configuration file
and stores the HIT-IP mappings into two linked lists,
one for legacy hosts and one for HIP enabled hosts.
Apart from the DNS issue the HIP proxy works as
described.
To get the received packets diverted to the proxy
application we use the IPv4 and IPv6 divert sockets
and the firewalls. Basically we tell the firewalls to
divert all packets received from the private network
except for broadcast packets and other packets that
we intuitively know that are not meant for the proxy.
This will result in that all connection initiations from
the legacy hosts, and the subsequent packets of the
connections, go through the proxy. To receive the
ESP packets sent from the HIP enabled hosts we tell
the proxy to divert all packets that have an address
prefix of 01
bin
for both source and destination
addresses. This is a characteristic of HITs; a HIT
always has the prefix 01
bin
(there is also a secondary
format for HITs with a 10
bin
prefix). Even if the
packets have IP addresses in the IP header while
they travel the Internet, IPsec processing, where the
IP addresses are replaced by HITs, happens before
the firewall rules are checked. Finally, to allow HIP
initiations from the HIP host, we tell the firewall to
allow all traffic that uses the HIP protocol, i.e. the
packets for the base exchange and the other HIP
specific packets such as the Update packet.
The structure of the application is divided into
two parts; in the first part the proxy is initialized, the
second part consists of a read/write loop where the
packets are processed. During the initialization part
the configuration file is read and the mappings found
in it are recorded. Before a HIP base exchange can
begin the Initiator has to have a HIP context for that
particular connection. A context consists of the
Initiator HIT along with the HIT and IP address of
the responder. The prototype creates the needed HIP
contexts after the configuration file is read. Before
the read/write loop begins the proxy also creates the
divert sockets so that it can receive packets.
In the read/write loop the proxy waits for packets
diverted to it. Once the proxy receives a packet it
examines the source and destination addresses of the
packet. Using the two linked lists with HIT-IP
mappings the proxy can conclude where the packet
Figure 3: Connection initiation via a HIP proxy
HOST IDENTITY PROTOCOL PROXY
227
is coming from and where it is going to, e.g. from
the legacy network to the HIP network. If either of
the addresses is not found in the linked lists the
proxy cannot process the packet correctly, in that
case the packet is forwarded unchanged by the
proxy. If mappings for both addresses are found, and
both addresses in the packet are found to be either
HITs or IP addresses (a mix of one IP and one HIT
is not accepted, it indicates an erroneous packet), the
proxy changes the IPs for HITs or vice versa. After
recalculating the checksums the packet is sent out
again. If the packets are going to the legacy host
they are forwarded via the output handling to the
private network. If the packet is going to one of the
HIP hosts it will have HITs as addresses in the IP
header. In this case the packet will be sent to IPsec
handling. If no SA is found for the specific
connection the HIP daemon is signaled to perform
the HIP base exchange after which the packet is sent
out utilizing the newly created SAs.
Before the read/write loop starts over again the
proxy checks if the configuration file should be re-
read. This makes it possible to add information
about new hosts without restarting the proxy. The
prototype uses a very basic method for finding out if
the file should be re-read; for each n packets the
configuration file is re-read. When testing this
feature, the value for n was set to 10. The value
should be adjusted based on how heavy traffic there
is through the proxy and the length of the list of
hosts entered into the file. With heavy traffic n
should be increased so that the re-read does not
happen very frequently. Also with a long list of
hosts n should be increased because with a longer
list the updating of the linked lists takes longer. A
more appropriate solution would be to check if the
file has been updated, and only when an update has
occurred should the file be re-read.
5.3 Performance
To measure how the HIP proxy prototype performs
some tests were conducted. The first test was done
to check how having the proxy in the path of the
packets affects the round trip times (RTT). First the
round trip times for ping6 were measured as an
average over 20 packets with the packets going
through the proxy but not being processed by it. To
get values to compare against the average round trip
times were also measured for the case when the
packets did not have to go via the proxy, the host
with the HIP proxy just forwarded the packets.
Finally we measure how the use of the HIP proxy,
and having it process packets affected the round trip
times. The results from these measurements are
presented in Table 1.
It can be concluded from the two first entries that
introducing the proxy does add delay; with the proxy
we get approximately 12% longer round-trip times.
This is something that can be expected since having
the packets go via the proxy adds processing on the
path. Having to pass a packet to an application in
user space, compared to only handling it in kernel
space, adds delay.
The last entry in Table 1 concentrates on how
applying IPsec ESP to the packets affect the delay.
Based on the result we can see that when the HIT-IP
mappings are found in the linked lists of the proxy
the round-trip time increases approximately by 22%
compared to having he proxy sending the packet to
output handling without any processing. When we
compare the delays of sending packets without using
the proxy and the case when the proxy is used and it
processes the packets we can see that the increase in
delay is approximately 36%. This increase in delay
includes both the added delay of having to send the
packet to user space, approximately 0,070ms, and
the delay that results from performing cryptographic
functions, approximately 0,150ms. The by the HIP
proxy added delay is mostly a result of doing the
cryptographic functions on the data. This is
something we cannot affect; if we want security it
will cost us time. The total delay added by the proxy
is not at an alarming level, and is as such acceptable.
Another interesting aspect of the performance of
the proxy is how the amount of entries in the linked
lists affects the delay. In the measurements
presented in Table 1 there were a total of three
entries, two in the HIP hosts list and one in the
legacy hosts list. In the next set of measurements we
had first 10 then 100 and finally 1000 entries per list.
The correct information was situated last in the
respective list so that the proxy would have to go
through all of the lists. For each packet both the
linked lists have to be examined. The results from
these measurements are presented in Table 2.
From the measured values we can see that if
we add enough entries to the lists it will show in the
round-trip times. But since the prototype proxy is
not meant for huge networks the delay added by
looking up mappings from long lists should not be
an issue. The values in Table 2 differ somewhat
from the corresponding values in Table 1. The
reason for the differing values is that the
ICETE 2005 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
228
measurements were performed at different times, so
the load on the network was different.
When the proxy reads the host information from
a configuration file, as is the case with this
prototype, the amount of hosts should be kept small
to keep the configuration file manageable. If some
automatic updating procedure is implemented it
allows for more hosts. Still, the delay caused by
having to look up host information from very long
lists will sooner or later add too much delay.
However, when the amount of hosts configured into
the proxy reaches that level it will probably be the
amount of traffic that the proxy has to handle that
will be the performance bottleneck, not the delay
from looking up the correct mappings.
6 FURTHER WORK
In the previous section we concluded that
approximately a third of the added delay that results
from using a HIP proxy compared to plain IP is a
result of the proxy application. This is one aspect of
the proxy that could be improved; by moving the
application from user space to kernel space the delay
induced by the proxy could probably be decreased.
Overall the proxy still performs well and as
expected. With a small set of hosts the delays are
kept at an acceptable level, keeping the RTTs in
roughly the same range as for legacy connections.
However, one must remember that a HIP proxy is
only a solution for a small set of nodes. When the
amount of nodes configured into the proxy gets too
big, either a second proxy should be introduced, and
the load balanced between the proxies, or then the
legacy hosts should be made HIP enabled. When a
HIP modified DNS is available it will increase the
limits of a HIP proxy by being able to dynamically
add HIT-IP mappings when they are needed. Also
old mappings that are considered obsolete can be
deleted since they can be re-fetched from DNS if
necessary. The amount of legacy hosts that the proxy
can serve will still be a limiting factor.
If the HIP proxy is situated in a public network
the security provided by HIP is in effect useless
since all the information also travels unencrypted in
the network, namely between the proxy and the
legacy host. This is quite alright as long as both
parties are aware of this. However, when using a
HIP proxy the HIP enabled host does not know that
it is communicating with a proxy but believes that it
is actually communicating with another HIP enabled
host. This puts the HIP enabled host at a
disadvantage, and it is a problem that needs to be
solved; the HIP enabled host must know when it is
communicating via a HIP proxy so that it knows that
the information it sends might not be secured all the
way to the actual endpoint.
A last issue that will be mentioned regarding the
HIP proxy is a problem that arises when the HIP
host, that is using the services of the HIP proxy, is
mobile. When a HIP host is mobile and moves to
another location, and thus gets a new locator, it
informs its communication parties of its new
location. With two HIP enabled hosts this works
well. However, when one of the endpoints is a HIP
proxy the location update message sent to the proxy
modifies established SAs as necessary, but the
information does not reach the proxy. If a
connection was established between a legacy host
and the HIP host before the location change, the
connection will continue to work even after the HIP
host has moved. If another legacy host now tries to
initiate a connection to the mobile host, using its
new locator, the connection will not be established
since the proxy has not gotten the new locator of the
mobile HIP host. This can be solved by updating the
proxy configuration file with the new information of
the mobile HIP host. This works well if there are no
connections established from legacy hosts to the old
locator of the HIP host. However, if there still are
connections to the old locator the result is that the
legacy host using the old locator of the mobile HIP
host will begin receiving packets from the HIP host's
new locator without knowing that it actually is the
same host. A solution for this problem could be that
the HIP proxy would keep a record of previous
locators of each HIP host, and state information for
each connection. Using this information all
connections could be maintained. All this of course
adds delay to the system. The solution presented
Table 2: The effects of serving many hosts
Hosts/list Avg. RTT
10 0,676ms
100 0,705ms
1000 0,770ms
Table 1: How the proxy affects round-trip times
Using
proxy
Using HIP Avg. RTT
No No 0,624ms
Yes No 0,698ms
Yes Yes 0,851ms
HOST IDENTITY PROTOCOL PROXY
229
here is probably not the optimal one and some more
research in this area is needed.
7 CONCLUSIONS
The HIP proxy prototype was constructed as a
proof-of-concept for a HIP proxy. The proxy
performs well and fills its tasks. However, as
mentioned in the previous section there are still
many areas in which the proxy may, and should, be
improved. The preferred solution for using HIP is of
course to have HIP enabled hosts. However, a HIP
proxy might be a good tool to help HIP get
spreading. The HIP proxy prototype described in
this paper is probably not something that should be
used as such for a HIP proxy. However, it might be a
good starting point for developing a new and
improved version that better fits the requirements of
a HIP proxy.
REFERENCES
Kent (1), Atkins, 1998. Security Architecture for the
Internet Protocol. RFC 2401
Moskowitz (1), Nikander, 2004. Host Identity Protocol
Architecture. Internet-draft, draft-moskowitz-hip-arch-
06 (work in progress).
Braden, Clark, Shenker, Wroclawski, 2000. Developing a
Next-Generation Internet Architecture.
Crowcroft, Hand, Mortier, Roscoe, Warfield, 2003.
Plutarch: An Argument for Network Pluralism.
Crawford, Mankin, Narten, Stewart, Zhang, 1999.
Separating Identifiers and Locators in Addresses: An
Analysis of the GSE Proposal for IPv6. Internet-draft.
Clark, Braden, Falk, Pingali, 2003. FARA: Reorganizing
the Addressing Architecture.
Eriksson, Faloutsos, Krishnamurthy, 2003. PeerNet:
Pushing Peer-to-Peer Down the Stack.
Stoica, Adkins, Ratnasamy, Shenker, Surana, Zhuang,
2002. Internet Indirection Infrastructure.
Nikander, Arkko, Ohlman, 2004. Host Identity Indirection
Infrastructure (Hi3).
Moskowitz (2), Nikander, Jokela, Henderson, 2004. Host
Identity Protocol. Internet-draft, draft-ietf-hip-base-00
(work in progress).
Rescorla, 1999. Diffie-Hellman Key Agreement Method.
RFC 2631.
Krawczyk, Bellare, Canetti, 1997. HMAC: Keyed-
Hashing for Message Authentication. RFC 2104.
Kent (2), Atkins, 1998. IP Encapsulating Security Payload
(ESP). RFC 2406.
HIP for BSD project, http://hip4inter.net [Referenced
11.02.2005].
ICETE 2005 - SECURITY AND RELIABILITY IN INFORMATION SYSTEMS AND NETWORKS
230