You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

424 lines
17 KiB

%** Design.tex: How was the problem attacked, what was the design
% the architecture
In this chapter we describe the architecture of our solution and our
design choices.
% ----------------------------------------------------------------------
\section{\label{design:configuration}IPv6 and IPv4 configuration}
The following sections refer to host and network configurations. In
this section we describe the IPv6 and IPv4 configurations as a basis
for the discussion.
All IPv6 addresses are from the documentation block
\textit{2001:DB8::/32}~\cite{rfc3849}. In particular the following sub
networks and IPv6 addresses are used:
\begin{tabular}{| c | c |}
\textbf{Address} & \textbf{Description} \\
2001:db8:42::/64 & IPv6 host network \\
2001:db8:23::/96 & IPv6 mapping to the IPv4 Internet \\
2001:db8:42::42 & IPv6 host address \\
2001:db8:42::77 & IPv6 router address \\
2001:db8:42::a00:2a & In-network IPv6 address mapped to (p4)\\
2001:db8:23::a00:2a & IPv6 address mapped to (tayga) \\
2001:db8:23::2a & IPv6 address mapped to (jool)\\
\caption{IPv6 address and network overview}
We use private IPv4 addresses as specified by RFC1918~\cite{rfc1918}
from the range as follows:
\begin{tabular}{| c | c |}
\textbf{Address} & \textbf{Description} \\
\hline & IPv4 host network \\
\hline & IPv4 network mapping to IPv6\\
\hline & IPv4 router address\\
\hline & In-network IPv4 address mapped to 2001:db8:42::42 (p4)\\
\hline & IPv4 address mapped to 2001:db8:42::42 (tayga)\\
\hline & IPv4 address mapped to 2001:db8:42::42 (jool)\\
\caption{IPv4 address and network overview}
% ok
% ----------------------------------------------------------------------
\section{\label{design:tests}NAT64 Verification}
We use socat~\cite{rieger:_multip} to verify basic operation of the
NAT64 gateway and iperf~\cite{dugan:_tcp_udp_sctp} to test stability
of the implementation and measure bandwidth.
In particular we use
the commands listed in table \ref{tab:nat64verification}. The socat
commands allow interactive testing on TCP and UDP connections, while
the iperf commands fully utilise the available bandwidth with test
The socat and iperf commands are used to verify all three NAT64
implementations (p4, tayga, jool).
\begin{tabular}{| c | c | c |}
\textbf{Command} & \textbf{Example} & \textbf{Description} \\
\texttt{socat - TCP6:HOST:PORT} & socat -
TCP6:[2001:db8:42::a00:2a]:2345 & Connect via IPv6/TCP\\
& & to IPv4 host\\
\texttt{socat - UDP6:HOST:PORT} & socat -
UDP6:[2001:db8:42::a00:2a]:2345 & Connect via IPv6/UDP \\ & & to IPv4 host\\
\texttt{socat - TCP:HOST:PORT} & socat -
TCP: & Connect via IPv4/TCP \\ & & to IPv6 host \\
\texttt{socat - UDP:HOST:PORT} & socat -
UDP: & Connect via IPv4/UDP \\ & & to IPv6 host \\
\texttt{socat - UDP6-LISTEN:PORT} & socat -
UDP6-LISTEN:2345 & Listen on IPv6/UDP \\
\texttt{socat - TCP6-LISTEN:PORT} & socat -
TCP6-LISTEN:2345 & Listen on IPv6/TCP \\
\texttt{socat - UDP-LISTEN:PORT} & socat -
UDP-LISTEN:2345 & Listen on IPv4/UDP \\
\texttt{socat - TCP-LISTEN:PORT} & socat -
TCP-LISTEN:2345 & Listen on IPv4/TCP \\
\texttt{iperf3 -PROTO -p PORT} & iperf3 -4 -p 2345 & IPv4 iperf server\\
\texttt{-B IP -s} & -B -s &\\
& iperf3 -6 -p 2345 & IPv6 iperf server\\
& -B 2001:db8:42::42 -s & \\
\texttt{iperf3 -PROTO -p PORT } & iperf3 -6 -p 2345& Connect to iperf server\\
\texttt{-O IGNORETIME -t RUNTIME} & -O 10 -t 190 &
Run for 190 seconds, \\
& & skip first 10 seconds\\
\texttt{-P PARALLEL -c IP} & -P20 -c 2001:db8:23::2a &
with 20 sessions\\
& & connecting to\\
& & 2001:db8:23::2a\\
\texttt{iperf3 -PROTO -p PORT} & & Same as above,\\
\texttt{-O IGNORETIME -t RUNTIME} & & but connect via UDP\\
\texttt{-P PARALLEL -c IP} & & \\
\texttt{-u -b0} & & \\
\caption{NAT64 verification commands}
% ----------------------------------------------------------------------
\section{\label{design:nat64}NAT64 with P4}
\caption{P4 Switch Architecture}
In section \ref{background:transition} we discussed different
translation mechanisms for IPv6 and IPv4. In this thesis we focus on
the translation mechansims stateless and stateful NAT64. While higher
layer protocol dependent translations are more flexible, this topic
has already been addressed in
\cite{nico18:_implem_layer_ipv4_ipv6_rever_proxy} and the focus in
this thesis is on the practicability of high speed NAT64 with P4.
The high level design can be seen in figure \ref{fig:switchdesign}: a
P4 capable switch is running our code to provide NAT64
functionality. A P4 switch cannot manage its tables on it own and
needs support for this from a controller. The controller also has the
role to handle unknown packets and can modify the runtime
configuration of the switch. This is especially useful in the case of
stateful NAT64.
If only static table entries
are required, they can usually be added at the start of a P4 switch
and the controller can also be omitted. However stateful
NAT64 requires the use of a controller to create session entries in the
switch tables.
The P4 switch can use any protocol to communicate with the controller, as
the connection to the controller is implemented as a separate ethernet
\caption{Standard NAT64 translation}
Software NAT64 solutions typically require routing to be applied to
transport the packet to the NAT64 translator as shown in figure
Our design differs here: while routing could be used like described
above, NAT64 with P4 does not require any routing to be setup. Figure
\ref{fig:v6v4mixed} shows the network design that we realise using
P4. This design has multiple advantages: first it reduces the number
of devices to pass and thus directly reduces the RTT, secondly it
allows translation of IP addresses within the same logic network
\caption{In-network NAT64 translation}
% ----------------------------------------------------------------------
\section{\label{design:statelessnat64}Stateless NAT64}
As seen in section \ref{background:transition:stateless}, stateless
NAT64 can be implemented using various factors. Our design for the
stateless depends on the capabilities of the environment and is
summarised in table \ref{tab:statelessnat64factors}.
\begin{tabular}{| c | c |}
\textbf{Implementation} & \textbf{NAT64 match}\\
P4/BMV2 & LPM (both directions)\\
& and individual entries (both directions)\\
P4/NetPFGA & Individual entries\\
Tayga & LPM (IPv6 to IPv4) and individual entries (IPv4 to IPv6)\\
Jool & LPM (both directions)\\
\caption{NAT64 match factors}
When using LPM for translating from IPv6 to IPv4, a /96 IPv6 network
is configured for covering the whole IPv4 Internet and the individual
IPv4 address is appended to the prefix (compare section
\ref{design:configuration}). We also use LPM to match on an IPv4 sub
network that translates to an IPv6 sub network. Individual
entries are configured differently depending on the implementation:
Limitations in the P4/NetFPGA environment require to use table
entries. Jool supports individual entries as a special case of LPM,
with a network mask matching only one IP address. Tayga
support LPM for translation from IPv6 to IPv4, but requires invidiual
entries for translating from IPv4 to IPv6. Our P4/BMV2 offers the
highest degree of flexibility, as it provides support for invidual
entries based on table entries and LPM table entries.
% ----------------------------------------------------------------------
\section{\label{design:statefulnat64}Stateful NAT64}
Similar to stateless NAT64, the design of stateful NAT64 depends on
the features of the invidual implementation. As pointed out in section
\ref{background:transition:statefulnat64}, stateful NAT64 is very
similar to stateless NAT64, with the main difference being an
additional stateful table that helps to create 1:n mappings.
We use different approaches within the implementations
to solve this problem:
\item For P4/BMV2 and P4/NetPFGA a python controller handles packets
that don't have a table entry, sets the table entry in the P4 switch
and inserts the original packet afterwards back into the switch.
\item With tayga we rely on the Linux kernel NAT44 capabilities
\item Jool implements its own stateful mechanism based on a port
All methods though operate in a very similar fashion: A ``controller''
inspects the IPv6 packet and depending on the source address,
destination address, protocol (TCP, UDP,
ICMP, ICMP6, etc.) and the protocol ID (source / destination TCP/UDP
port, ICMP identifier) it selects an outgoing IPv4 address, and source
port or ICMP identifier.
In case of Jool and Tayga this decision is based on a session table
inside the Linux kernel, in case of P4 this decision is based on a
session table inside the python controller. While the Jool and Tayga
both support cleaning up old session entries,
our P4 based solution does not support this feature at the moment.
% ----------------------------------------------------------------------
/* checksumming for icmp6_na_ns_option */
update_checksum_with_payload(meta.chk_icmp6_na_ns == 1,
hdr.ipv6.src_addr, /* 128 */
hdr.ipv6.dst_addr, /* 128 */
meta.cast_length, /* 32 */
24w0, /* 24 0's */
PROTO_ICMP6, /* 8 */
hdr.icmp6.type, /* 8 */
hdr.icmp6.code, /* 8 */
\caption{P4/BMV2 checksumming}
The software emulated switch that is implemented using
Open vSwitch~\cite{openvswitch} and the
behavioral model~\cite{_implem_your_switc_target_with_bmv2}
offers the fastest and easiest way of P4 development. All NAT64
features are tested first on P4/BMV2 and in a second step ported to
P4/NetFPGA and modified, where necessary.
The development follows closely the general design shown in section
As outlined in section \ref{background:checksums}, checksums inside
higher level protocols need to be adjusted after translation.
Within the software emulation checksums can be
computed with two different methods:
\item Recalculating the checksum by inspecting headers and payload
\item Calculating the difference between the translated headers
The BMV2 model is sophisticated and provides direct support
for calculating the checksum over the payload. This allows the BMV2
model to operate as a full featured host, including advanced features
like responding to ICMP6 Neighbor discovery requests~\cite{rfc4861}
that include payload checksums. Sample code that calculates the
required checksum for answering NDP queries is shown in figure
\ref{fig:bmv2checksum}. The code shows how the field
\texttt{hdr.icmp6.checksum} is updated with the \texttt{csum16} method
depending on the IPv6 and ICMP6 headers as well as the payload. The
second option of using the differences is described in section
% ok
% ----------------------------------------------------------------------
\section{\label{design:netpfga}NetFPGA - FIXME: relate things}
While the P4-NetFPGA project ~\cite{netfpga:_p4_netpf_public_github}
allows compiling P4 to the NetPFGA, the design slightly varies.
In particular, the NetFPGA P4 compiler does not support reading
the payload. For this reason it also does not support
creating the checksum based on the payload.
To support checksum modifications in NAT64 on the NetFPGA, the
checksum was calculated on the netpfga using differences between
the IPv6 and IPv4 headers. Figure \ref{fig:checksumbydiff} shows an
excerpt of the code used for calculating checksums in the netpfga.
action v4sum() {
bit<16> tmp = 0;
tmp = tmp + (bit<16>) hdr.ipv4.src_addr[15:0]; // 16 bit
tmp = tmp + (bit<16>) hdr.ipv4.src_addr[31:16]; // 16 bit
tmp = tmp + (bit<16>) hdr.ipv4.dst_addr[15:0]; // 16 bit
tmp = tmp + (bit<16>) hdr.ipv4.dst_addr[31:16]; // 16 bit
tmp = tmp + (bit<16>) hdr.ipv4.totalLen -20; // 16 bit
tmp = tmp + (bit<16>) hdr.ipv4.protocol; // 8 bit
meta.v4sum = ~tmp;
/* analogue code for v6sum skipped */
action delta_tcp_from_v6_to_v4()
bit<17> tmp = (bit<17>) hdr.tcp.checksum + (bit<17>) meta.v4sum;
if (tmp[16:16] == 1) {
tmp = tmp + 1;
tmp[16:16] = 0;
tmp = tmp + (bit<17>) (0xffff - meta.v6sum);
if (tmp[16:16] == 1) {
tmp = tmp + 1;
tmp[16:16] = 0;
hdr.tcp.checksum = (bit<16>) tmp;
\caption{Calculating checksum based on header differences}
The checksums for IPv4, TCP, UDP and ICMP6 are all based on the
``Internet Checksum''~\cite{rfc791},~\cite{rfc1071}.
Its calculation can be summarised as follows:
The checksum field is the 16-bit one's complement of the one's
complement sum of all 16-bit words in the header. For purposes of
computing the checksum, the value of the checksum field
is zero.\footnote{Quote from Wikipedia~\cite{wikipedia:_ipv4}.}.
As the calculation mainly depends on on (1-complement) sums, the
checksums after translating the protocol can be corrected by
subtracting the differences of the relevant fields. It is notable that
not the full headers are used, but the pseudo headers (compare figures
\ref{fig:ipv6pseudoheader} and \ref{fig:ipv4pseudoheader}).
To compensate the carry bit, our code uses 17 bit integers for
correcting the carry.
% FIXME: add note to python script / checksum diffing
% ----------------------------------------------------------------------
The benchmarks were performed on two hosts, a load generator and a
nat64 translator. Both hosts were equipped with a dual port
Intel X520 10 Gbit/s network card. Both hosts were connected using DAC
without any equipment in between. TCP offloading was enabled in the
X520 cards. Figure \ref{fig:softwarenat64design}
shows the network setup.
\caption{NAT64 in software benchmark}
When testing the NetPFGA/P4 performance, the X520 cards in the NAT64
translator were diconnected and instead the NetPFGA ports were
connected, as show in figure \ref{fig:netpfgadesign}. The load
generator is equipped with a quad core CPU (Intel(R) Core(TM) i7-6700
CPU @ 3.40GHz), enabled with hyperthreading and 16 GB RAM. The NAT64
translator is also equipped with a quard core CPU (Intel(R) Core(TM)
i7-4770 CPU @ 3.40GHz) and 16 GB RAM.
The first 10 seconds of the benchmark were excluded to avoid the TCP
warm up phase.\footnote{iperf -O 10 parameter}
\caption{NAT64 with NetFPGA benchmark}
% ok