DMM T. Murakami Internet-Draft Arrcus, Inc. Intended status: Informational S. Matsushima Expires: 11 January 2024 L. Fujita SoftBank 10 July 2023 Impact analysis from IPv6 GTP-U checksum calculation draft-murakami-dmm-udp-checksum-impact-gtpu-01 Abstract This document describes about the impact on the performance when calculating the checksum for IPv6 GTP-U packet upon encapsulating the packet into IPv6 GTP-U. Status of This Memo 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 11 January 2024. Copyright Notice Copyright (c) 2023 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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Murakami, et al. Expires 11 January 2024 [Page 1] Internet-Draft GTP-U Checksum calculation July 2023 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1. Requirements Notation . . . . . . . . . . . . . . . . . . 2 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3 3. Impact analysis from UDP checksum calculation . . . . . . . . 3 3.1. Vector Packet Processing . . . . . . . . . . . . . . . . 3 3.2. Software base checksum calculation . . . . . . . . . . . 3 3.3. Checksum calculation offload . . . . . . . . . . . . . . 5 4. Security Considerations . . . . . . . . . . . . . . . . . . . 6 5. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 6 6. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 6 7. References . . . . . . . . . . . . . . . . . . . . . . . . . 6 7.1. Normative References . . . . . . . . . . . . . . . . . . 6 7.2. Informative References . . . . . . . . . . . . . . . . . 7 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 7 1. Introduction [RFC6935] allows to use zero checksum for IPv6 UDP when using IPv6 UDP for encapsulating a packet. Eliminating the checksum calculation contributes huge performance improvement in terms of forwarding packets. 3GPP also allows UDP checksum zero for GTP-U over IPv6 UDP encapsulation since Release-16 onward[TS.29281]. However, UDP checksum seem still remained in GTP-U over IPv6/UDP encapsulation implementations. This can be causing non-negligible performance impact on nodes, especially in NFV environment, which nodes are encapsulating the packet into IPv6 GTP-U. This document describes an analysis of network performance impact caused by IPv6 UDP checksum calculation. To do the analysis, we measured latency variation on three environments, (1) UDP checksum zero, (2) UDP checksum calculated by software, (3) offloading UDP checksum calculation. These latencies were measured on a VPP (Vector Packet Processing) instance when a packet encapsulated with an IPv6, a UDP and a GTP-U headers. 1.1. Requirements Notation 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 BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. Murakami, et al. Expires 11 January 2024 [Page 2] Internet-Draft GTP-U Checksum calculation July 2023 2. Terminology GTP-U: GPRS Tunneling Protocol for User Plane VPP: Vector Packet Processing NIC: Network Interface Card 3. Impact analysis from UDP checksum calculation 3.1. Vector Packet Processing VPP is doing the batching process on the received packets. VPP stores some received packets and process these packets at once. Hence, it causes some small degration on latency to send out the packets to the network. INPUT: +--+ +--+ +--+ +--+ |P1| |P2| |P3| ... |Pn| +--+ +--+ +--+ +--+ Batching Process | (n packets) | |<------------------->| OUTPUT: | +--+ +--+ +--+ +--+ | |P1| |P2| |P3| ... |Pn| | +--+ +--+ +--+ +--+ | | |<------------ batching time ------------> Based on VPP, the impact on the latency is measured when encapsulating the packets into IPv6 GTP-U with software base checksum calculation and with checksum calculation offload in order to figure out the impact on the network performance. In order to simplify the impact analysis, only 1 CPU core is assigned to the packet processing in VPP and the packets are arrived at one interface and sent out to another interface. The traffic generator outside of VPP is sending out the packets and receives IPv6 GTP-U encapsulated packets. Based on this, the impact on the latency is measured in each case. 3.2. Software base checksum calculation This section describes the performance impact analysis when using software base checksum calculation. Murakami, et al. Expires 11 January 2024 [Page 3] Internet-Draft GTP-U Checksum calculation July 2023 No checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 15,336 | 14,535 | 124,397 | +----------------------+----------------------+--------------------+ Software checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 15,477 | 14,592 | 123,337 | +----------------------+----------------------+--------------------+ INPUT: 100 pps, 1492 byte packet In this case, there is no impact on the network performance. Since the incoming packet rate is enough small, the software checksum calculation can be done within the batching time to process the packets. No checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 120,005 | 62,650 | 1,300,217 | +----------------------+----------------------+--------------------+ Software checksum calculation: 3.676% packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 8,167,461 | 52,467 | 9,341,807 | +----------------------+----------------------+--------------------+ INPUT: 275k pps, 1492 byte packet In this case, there is huge impact on the network performance. If the total CPU time required for calculating UDP checksum is exceeding the batching time to process the packets, it causes huge impact on the latency. In addition, since the checksum calculation steals CPU times and the software can not acquire enough CPU times to process the packets, it causes the huge packet loss. Murakami, et al. Expires 11 January 2024 [Page 4] Internet-Draft GTP-U Checksum calculation July 2023 3.3. Checksum calculation offload Some of NIC can support UDP checksum calculation offload. When enabling this function on NIC, UDP checksum is calculated by NIC. In this case, CPU time is not consumed for calculating UDP checksum. This section describes the performance impact analysis when enabling UDP checksum offload on NIC. No checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 15,336 | 14,535 | 124,397 | +----------------------+----------------------+--------------------+ UDP checksum offload: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 15,349 | 14,537 | 63,742 | +----------------------+----------------------+--------------------+ INPUT: 100 pps, 1492 byte packet In this case, there is no impact on the network performance. Since the incoming packet rate is enough small, all received packet can be processed within the batching time. Murakami, et al. Expires 11 January 2024 [Page 5] Internet-Draft GTP-U Checksum calculation July 2023 No checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 120,005 | 62,650 | 1,380,217 | +----------------------+----------------------+--------------------+ Software checksum calculation: No packet loss +----------------------+----------------------+--------------------+ | Store-Forward | Store-Forward | Store-Forward | | Avg Latency (ns) | Min Latency (ns) | Max Letency (ns) | +----------------------+----------------------+--------------------+ | 134,126 | 74,090 | 2,443,992 | +----------------------+----------------------+--------------------+ INPUT: 275k pps, 1492 byte packet In this case, there is small impact on the network performance. The time required to calculate UDP checksum by NIC can not be done within the batching time and hence it causes small degrade on the latency. However, even though there are huge packet loss when using software base checksum calculation with same condition, there is no packet loss when using UDP checksum offload. 4. Security Considerations No secturity consideration. 5. IANA Considerations No IANA consideration. 6. Contributors In addition to the authors listed on the front page, the following individuals have also made significant contributions to the draft: (Artwork only available as : No external link available, see draft- murakami-dmm-udp-checksum-impact-gtpu-01.html for artwork.) 7. References 7.1. Normative References Murakami, et al. Expires 11 January 2024 [Page 6] Internet-Draft GTP-U Checksum calculation July 2023 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC6935] Eubanks, M., Chimento, P., and M. Westerlund, "IPv6 and UDP Checksums for Tunneled Packets", RFC 6935, DOI 10.17487/RFC6935, April 2013, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . 7.2. Informative References [TS.23501] 3GPP, "System architecture for the 5G System (5GS)", 3GPP TS 23.501 17.2.0, 24 September 2021, . [TS.29281] 3GPP, "General Packet Radio System (GPRS) Tunnelling Protocol User Plane (GTPv1-U)", 3GPP TS 29.281 16.1.0, September 2020. Authors' Addresses Tetsuya Murakami Arrcus, Inc. 2077 Gateway Place, Suite 400 San Jose, CA 95110 United States of America Email: tetsuya@arrcus.com Satoru Matsushima SoftBank Japan Email: satoru.matsushima@g.softbank.co.jp Leo Fujita SoftBank Japan Email: leo.fujita@g.softbank.co.jp Murakami, et al. Expires 11 January 2024 [Page 7]