| Internet-Draft | TCP ACK Rate Request Option | October 2020 |
| Gomez & Crowcroft | Expires 4 May 2021 | [Page] |
TCP Delayed Acknowledgments (ACKs) is a widely deployed mechanism that allows reducing protocol overhead in many scenarios. However, Delayed ACKs may also contribute to suboptimal performance. When a relatively large congestion window (cwnd) can be used, less frequent ACKs may be desirable. On the other hand, in relatively small cwnd scenarios, eliciting an immediate ACK may avoid unnecessary delays that may be incurred by the Delayed ACKs mechanism. This document specifies the TCP ACK Rate Request (TARR) option. This option allows a sender to indicate the ACK rate to be used by a receiver, and it also allows to request immediate ACKs from a receiver.¶
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 4 April 2021.¶
Copyright (c) 2020 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.¶
Delayed Acknowledgments (ACKs) were specified for TCP with the aim to reduce protocol overhead [RFC1122]. With Delayed ACKs, a TCP delays sending an ACK by up to 500 ms (often 200 ms, with lower values in recent implementations such as ~50 ms also reported), and typically sends an ACK for at least every second segment received in a stream of full-sized segments. This allows combining several segments into a single one (e.g. the application layer response to an application layer data message, and the corresponding ACK), and also saves up to one of every two ACKs, under many traffic patterns (e.g. bulk transfers). The "SHOULD" requirement level for implementing Delayed ACKs in RFC 1122, along with its expected benefits, has led to a widespread deployment of this mechanism.¶
However, there exist scenarios where Delayed ACKs contribute to suboptimal performance. We next roughly classify such scenarios into two main categories, in terms of the congestion window (cwnd) size and the Maximum Segment Size (MSS) that would be used therein: i) "large" cwnd scenarios (i.e. cwnd >> MSS), and ii) "small" cwnd scenarios (e.g. cwnd up to ~MSS).¶
In "large" cwnd scenarios, increasing the number of data segments after which a receiver transmits an ACK beyond the typical one (i.e. 2 when Delayed ACKs are used) may provide significant benefits. One example is mitigating performance limitations due to asymmetric path capacity (e.g. when the reverse path is significantly limited in comparison to the forward path) [RFC3449]. Another advantage is reducing the computational cost both at the sender and the receiver, and reducing network packet load, due to the lower number of ACKs involved.¶
In many "small" cwnd scenarios, a sender may want to request the receiver to acknowledge a data segment immediately (i.e. without the additional delay incurred by the Delayed ACKs mechanism). In high bit rate environments (e.g. data centers), a flow's fare share of the available Bandwidth Delay Product (BDP) may be in the order of one MSS, or even less. For an accordingly set cwnd value (e.g. cwnd up to MSS), Delayed ACKs would incur a delay that is several orders of magnitude greater than the RTT, severely degrading performance. Note that the Nagle algorithm may produce the same effect for some traffic patterns in the same type of environments [RFC8490]. In addition, when transactional data exchanges are performed over TCP, or when the cwnd size has been reduced, eliciting an immediate ACK from the receiver may avoid idle times and allow timely continuation of data transmission and/or cwnd growth, contributing to maintaining low latency.¶
Further "small" cwnd scenarios can be found in Internet of Things (IoT) environments. Many IoT devices exhibit significant memory constraints, such as only enough RAM for a send buffer size of 1 MSS. In that case, if the data segment does not elicit an application-layer response, the Delayed ACKs mechanism unnecessarily contributes a delay equal to the Delayed ACK timer to ACK transmission. The sender cannot transmit a new data segment until the ACK corresponding to the previous data segment is received and processed.¶
With the aim to provide a tool for performance improvement in both "large" and "small" cwnd scenarios, this document specifies the TCP ACK Rate request (TARR) option. This option allows a sender to indicate the ACK rate to be used by a receiver, and it also allows to request immediate ACKs from a receiver.¶
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 [RFC2119].¶
A TCP endpoint announces that it supports the TARR option by including the TARR option format in packets that have the SYN bit set. In such packets, the values carried by the TARR option format other than Kind, Length and ExID MUST be ignored by the receiving TCP.¶
TBD1: perhaps two formats (with one codepoint each), a shorter one just to advertise that TARR is supported, and a larger one which is the current TARR option format defined in section 4?¶
The next two subsections define the sender and receiver behaviors for devices that support the TARR option, respectively.¶
A TCP sender MUST NOT include the TARR option in TCP data segments to be sent if the TCP receiver does not support the TARR option.¶
A TCP sender MAY request a TARR-option-capable receiver to modify the ACK rate of the latter to one ACK every R full-sized data segments received from the sender. This request is performed by the sender by including the TARR option in the TCP header of a data segment. The TARR option carries the R value requested by the sender (see section 4). For the described purpose, the value of R MUST NOT be zero. The TARR option also carries the N field, which MUST be ignored when R is not set to zero.¶
When a TCP sender needs a data segment to be acknowledged immediately by a TARR-option-capable receiving TCP, the sender includes the TARR option in the TCP header of the data segment, with a value of R equal to zero. When R is set to zero, the N field of the same option indicates the number of subsequent data segments for which the sender also requests immediate ACKs.¶
A TCP sender MAY indicate that it has a reordering tolerance of R packets by setting the Ignore Order field of the TARR option to True (see Section 4).¶
A receiving TCP conforming to this specification MUST process a TARR option present in a received data segment.¶
When the TARR option of a received segment carries an R value different from zero, a TARR-option-capable receiving TCP MUST modify its ACK rate to one ACK every R full-sized received data segments from the sender, as long as packet reordering does not occur. When R is different from zero, the receiving TCP MUST ignore the N field of the TARR option.¶
A TARR-option-capable TCP that receives a TARR option with the Ignore Order field set to True (see Section 4), MUST NOT send an ACK after each reordered data segment. Instead, it MUST continue to send one ACK every R received data segments. Otherwise (i.e., Ignore Order = False), such a receiver will need to send an ACK after each reordered data segment received.¶
If a TARR-option-capable TCP receives a segment carrying the TARR option with R=0, the receiving TCP MUST send an ACK immediately, and it MUST also send an ACK immediately after each one of the N next consecutive segments to be received. N refers to the corresponding field in the TARR option of the received segment (see Section 4).¶
The TARR option has the format and content shown in Fig. 1.¶
0 1 2 3
01234567 89012345 67890123 45678901
+--------+--------+--------+--------+
| Kind | Length | ExID |
+--------+--------+--------+--------+
| R | IgnOrd | N |
+--------+--------+--------+
Kind: The Kind field value is TBD.¶
Length: The Length field value is 7 bytes.¶
ExID: The experiment ID field size is 2 bytes, and its value is 0x00AC.¶
R: The size of this field is 1 byte. If all bits of this field are set to 0, the field indicates a request by the sender for the receiver to trigger one or more ACKs immediately. Otherwise, the field carries the binary encoding of the number of full-sized segments received after which the receiver is requested by the sender to send an ACK.¶
Ignore Order: The size of this field is 1 byte. This field MUST have the value 0x01 ("True") or 0x00 ("False"). When this field is set to True, the receiver MUST NOT send an ACK after each reordered data segment. Instead, it MUST continue to send one ACK every R received data segments.¶
TBD2: perhaps 7 bits for R and 1 bit for Ignore Order?¶
N: The size of this field is 1 byte. When R=0, the N field indicates the number of subsequent consecutive data segments to be sent for which immediate ACKs are requested by the sender.¶
This document specifies a new TCP option (TCP ACK Rate Request) that uses the shared experimental options format [RFC6994], with ExID in network-standard byte order.¶
The authors plan to request the allocation of ExID value 0x00AC for the TCP option specified in this document.¶
Bob Briscoe, Jonathan Morton, Richard Scheffenegger, Neal Cardwell, Michael Tuexen, Yuchung Cheng, Matt Mathis, Jana Iyengar, Gorry Fairhurst, and Stuart Cheshire provided useful comments and input for this document. Jana Iyengar suggested including a field to allow a sender communicate its tolerance to reordering. Gorry Fairhurst suggested adding a mechanism to request a number of consecutive immediate ACKs.¶
Carles Gomez has been funded in part by the Spanish Government through project PID2019-106808RA-I00, and by Secretaria d'Universitats i Recerca del Departament d'Empresa i Coneixement de la Generalitat de Catalunya 2017 through grant SGR 376.¶