| Internet-Draft | Benchmarking Stateful Gateways | October 2021 |
| Lencse & Shima | Expires 13 April 2022 | [Page] |
RFC 2544 has defined a benchmarking methodology for network interconnect devices. RFC 5180 addressed IPv6 specificities and it also provided a technology update, but excluded IPv6 transition technologies. RFC 8219 addressed IPv6 transition technologies, including stateful NAT64. However, none of them discussed how to apply RFC 4814 pseudorandom port numbers to any stateful NAT (NAT44, NAT64, NAT66) technologies. We discuss why using pseudorandom port numbers with stateful NAT gateways is a hard problem and recommend a solution.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 13 April 2022.¶
Copyright (c) 2021 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 (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
[RFC2544] has defined a comprehensive benchmarking methodology for network interconnect devices, which is still in use. It was mainly IP version independent, but it used IPv4 in its examples. [RFC5180] addressed IPv6 specificities and also added technology updates, but declared IPv6 transition technologies out of its scope. [RFC8219] addressed the IPv6 transition technologies, including stateful NAT64. It has reused several benchmarking procedures from [RFC2544] (e.g. throughput, frame loss rate), it has redefined the latency measurement, and added further ones, e.g. the PDV (packet delay variation) measurement.¶
However, none of them discussed, how to apply [RFC4814] pseudorandom port numbers, when benchmarking stateful NATxy (NAT44, NAT64, NAT66) gateways. We are not aware of any other RFCs that address this question.¶
First, we discuss why using pseudorandom port numbers with stateful NATxy gateways is a hard problem.¶
Then we recommend a solution.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
In its appendix, [RFC2544] has defined a frame format for test frames including specific source and destination port numbers. [RFC4814] recommends to use pseudorandom and uniformly distributed values for both source and destination port numbers. However, stateful NATxy (NAT44, NAT64, NAT66) solutions use the port numbers to identify connections. The usage of pseudorandom port numbers causes different problems depending on the direction.¶
As for the private to public direction, pseudorandom source and destination port numbers could be used, however, this approach would be a denial of service attack against the stateful NATxy gateway, because it would exhaust its connection tracking table. To that end, let us see some calculations using the recommendations of RFC 4814:¶
We note that section 12 of [RFC2544] also requires testing with 256 destination networks, which further increases the number of connection tracking table entries.¶
Our methodology works with any IP version. We use IPv4 in the Test Setup shown in Figure 1 to facilitate its easy understanding based on the well-known stateful NAT44 (also called NAPT: Network Address and Port Translation) solution.¶
+--------------------------------------+
10.0.0.2 |Initiator Responder| 198.19.0.2
+-------------| Tester |<------------+
| private IPv4| [state table]| public IPv4 |
| +--------------------------------------+ |
| |
| +--------------------------------------+ |
| 10.0.0.1 | DUT: | 198.19.0.1 |
+------------>| Sateful NATxy gateway |-------------+
private IPv4| [connection tracking table] | public IPv4
+--------------------------------------+
As for transport layer protocol, [RFC2544] recommended testing with UDP, and it was kept also in [RFC8219]. For the general recommendation, we also keep UDP, thus the port numbers in the following text are to be understood as UDP port numbers. We discuss the limitation of this approach in Section 6.¶
We define the most important elements of our proposed benchmarking system as follows.¶
The Initiator SHOULD use restricted ranges for source and destination port numbers to avoid the denial of service attack like event against the connection tracking table of the DUT described in Section 2. The size of the source port number range SHOULD be larger (e.g. in the order of a few times ten thousand), whereas the size of the destination port number range SHOULD be smaller (may vary from a few to several hundreds or thousands as needed). The rationale is that source and destination port numbers that can be observed in the Internet traffic are not symmetrical. Whereas source port numbers may be random, there are a few very popular destination port numbers (e.g. 443, 80, etc., see [IIR2020]) and others hardly occur. And we have found that their role is also asymmetric in the Linux kernel routing hash function [LEN2020].¶
The product of the sizes of the two ranges can be used as a parameter. The performance of the stateful NATxy gateway MAY be examined as a function of this parameter.¶
The preliminary phase serves two purposes:¶
Whereas the above two things are always necessary before the real test phase, the preliminary phase can be used without the real test phase. It is done so, when the maximum connection establishment rate is measured (as described in Section 4.4).¶
A preliminary test phase MUST be performed before all tests performed in the real test phase. In this phase, the following things happen:¶
When the preliminary test phase is performed in preparation to the real test phase, the applied frame rate and the duration of the preliminary phase SHOULD be carefully selected so that:¶
Our experience with iptables shows that the handling of a frame requires significantly more amount of work from the NAT44 gateway, when the frame creates a new connection, than when the frame belongs to an existing connection. Further more, we have also experienced that the depletion of the connection tracking table of iptables lasted significantly longer than its filling time at maximum connection establishment rate. Therefore, it is necessary to be able to control the connection tracking table entries of the DUT in order to achieve clear conditions for the measurements. We can simply achieve the following two extreme situations:¶
From this point we use the following two assumptions:¶
The first extreme situation can be achieved by¶
The second extreme situation can be achieved by¶
[RFC4814] REQUIRES pseudorandom port numbers, which we believe is a good approximation of the distribution of the source port numbers a NATxy gateway on the Internet may face with.¶
We note that pseudorandom all different source port number destination port number combinations may be computing efficiently generated by preparing a random permutation of the previously enumerated all possible source port number destination port number combinations using Dustenfeld's random shuffle algorithm [DUST1964]. This method also satisfies the criterion for the second case that all possible source port number destination port number combinations must be enumerated during the preliminary phase.¶
Important warning: in normal (non-NAT) router testing, the port number selection algorithm, whether it is pseudo-random or enumerated in increasing (or decreasing) order does not affect final results. However, our experience with iptables shows that if the connection tracking table is filled using port number enumeration in increasing order, then the maximum connection establishment rate of iptables degrades significantly compared to its performance using pseudorandom port numbers [LEN2021].¶
The enumeration of the source port number destination port number combinations in increasing or decreasing order (or in any other specific order) MAY be used as an additional measurement.¶
The maximum connection establishment rate is an important characteristic of the stateful NATxy gateway and its determination is necessary for the safe execution of the preliminary test phase (without frame loss) before the real test phase.¶
The measurement procedure of the maximum connection establishment rate is very similar to the throughput measurement procedure defined in [RFC2544].¶
Procedure: The Initiator sends a specific number of test frames using all different source port number destination port number combinations at a specific rate through the DUT. The Responder counts the frames that are successfully translated by the DUT. If the count of offered frames is equal to the count of received frames, the rate of the offered stream is raised and the test is rerun. If fewer frames are received than were transmitted, the rate of the offered stream is reduced and the test is rerun.¶
The maximum connection establishment rate is the fastest rate at which the count of test frames successfully translated by the DUT is equal to the number of test frames sent to it by the Initiator.¶
Notes:¶
As for the traffic direction, there are three possible cases during the real test phase:¶
If the Initiator sends test frames, then it uses pseudorandom source port numbers and destination port numbers from the restricted port number ranges. The responder receives the test frames, updates its state table and processes the test frames as required by the given measurement procedure (e.g. only counts them for throughput test, handles timestamps for latency or PDV tests, etc.).¶
If the Responder sends test frames, then it uses the four tuples from its state table. The reading order of the state table may follow different policies (discussed in Section 4.6). The Initiator receives the test frames, and processes them as required by the given measurement procedure.¶
As for the actual measurement procedures, we RECOMMEND to use the updated ones from Section 7 of [RFC8219].¶
As for writing policy of the state table of the Responder, we RECOMMEND round robin, because it ensures that its entries are automatically kept fresh and thus there is no need to handle timeout.¶
The Responder can read its state table in various orders. We RECOMMEND one of the following ones:¶
Pseudorandom reading order of the state table MAY NOT be used with unidirectional traffic from the Responder to the Initiator, because if a four tuple is not used until timeout time, then its connection is deleted from the connection tracking table of the DUT and a later use of the given four tuple will cause frame loss. There is no such problem, when bidirectional traffic is used, because then the state table of the Responder is periodically refreshed.¶
We do not see any problem in the round robin reading order, because the state table is filled using pseudorandom port numbers.¶
Stateful testing involves some issues not present in stateless testing.¶
Even though we do black box testing, one MUST consider timeout and carefully manage timeout budget. For example, if the frame rate is high enough, then every single entry of the state table of the Responder is refreshed within timeout time and it prevents frame sending with a stale four tuple. If the entries of the state table are not refreshed (due to testing with single directional traffic from the Responder to the Initiator) then using all four tuples within timeout time can keep all connection tracking table entries of the DUT alive.¶
Special care should be taken for the lower frame rate in the preliminary phase.¶
If the binary search (or the decreasing of the applied frame rates during the frame loss rate test) results in a frame rate that is too low to prevent the deletion of the connection tracking table entries of the DUT due to timeout, then it results in the failure of the consecutive tests (the binary search of the throughput test counts down to zero).¶
Several network performance tester vendors include a parameter called "Loss Tolerance" (or similar) for the throughput test and several benchmarking professionals actually use nonzero values [TOL2001]. If frames are lost during stateful testing (especially if it happens during a test with unidirectional traffic from the Responder to the Initiator) the refreshing of the corresponding connection tracking table element of the DUT is not ensured and it may result in the loss of further frames (not due to the low performance of the DUT, but due to using a stale four tuple).¶
The "stateful" branch of siitperf [SIITPERF] is an implementation of this concept.¶
Our experience is partially documented in a paper currently under review [LEN2021].¶
Stateful NATxy solutions handle TCP and UDP differently, e.g. iptables uses 30s timeout for UDP and 60s timeout for TCP. Thus benchmarking results produced using UDP do not necessarily characterize the performance of a NATxy gateway well enough, when they are used for forwarding Internet traffic. As for the given example, timeout values of the DUT may be adjusted, but it requires extra consideration.¶
Other differences in handling UDP or TCP are also possible. Thus we recommend that further investigations are to be performed in this field.¶
As a mitigation of this problem, we recommend that testing with protocols usig TCP (like HTTP and HTTPS) can be performed as described in [I-D.ietf-bmwg-ngfw-performance]. This approach also solves the potential problem of protocol helpers may be present in the stateful DUT.¶
The authors would like to thank Edwin Cordeiro, Lukasz Bromirski and Sandor Repas for their comments.¶
This document does not make any request to IANA.¶
We have no further security considerations beyond that of [RFC8219]. Perhaps they should be cited here so that they be applied not only for the benchmarking of IPv6 transition technologies, but also for the benchmarking of stateful NATxy gateways.¶
Initial version.¶
Updates based on the comments received on the BMWG mailing list and minor corrections.¶
Section 4.3 was completely re-written. As a consequence, the occurrences of the now undefined "mostly different" source port number destination port number combinations were deleted from Section 4.4, too.¶