RFC 8985 | RACK | February 2021 |
Cheng, et al. | Standards Track | [Page] |
This document presents the RACK-TLP loss detection algorithm for TCP. RACK-TLP uses per-segment transmit timestamps and selective acknowledgments (SACKs) and has two parts. Recent Acknowledgment (RACK) starts fast recovery quickly using time-based inferences derived from acknowledgment (ACK) feedback, and Tail Loss Probe (TLP) leverages RACK and sends a probe packet to trigger ACK feedback to avoid retransmission timeout (RTO) events. Compared to the widely used duplicate acknowledgment (DupAck) threshold approach, RACK-TLP detects losses more efficiently when there are application-limited flights of data, lost retransmissions, or data packet reordering events. It is intended to be an alternative to the DupAck threshold approach.¶
This is an Internet Standards Track document.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc8985.¶
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.¶
This document presents RACK-TLP, a TCP loss detection algorithm that improves upon the widely implemented duplicate acknowledgment (DupAck) counting approach described in [RFC5681] and [RFC6675]; it is RECOMMENDED as an alternative to that earlier approach. RACK-TLP has two parts. Recent Acknowledgment (RACK) detects losses quickly using time-based inferences derived from ACK feedback. Tail Loss Probe (TLP) triggers ACK feedback by quickly sending a probe segment to avoid retransmission timeout (RTO) events.¶
In traditional TCP loss recovery algorithms [RFC5681] [RFC6675], a sender starts fast recovery when the number of DupAcks received reaches a threshold (DupThresh) that defaults to 3 (this approach is referred to as "DupAck counting" in the rest of the document). The sender also halves the congestion window during the recovery. The rationale behind the partial window reduction is that congestion does not seem severe since ACK clocking is still maintained. The time elapsed in fast recovery can be just one round trip, e.g., if the sender uses SACK-based recovery [RFC6675] and the number of lost segments is small.¶
If fast recovery is not triggered or is triggered but fails to repair all the losses, then the sender resorts to RTO recovery. The RTO timer interval is conservatively the smoothed RTT (SRTT) plus four times the RTT variation, and is lower bounded to 1 second [RFC6298]. Upon RTO timer expiration, the sender retransmits the first unacknowledged segment and resets the congestion window to the loss window value (by default, 1 full-sized segment [RFC5681]). The rationale behind the congestion window reset is that an entire flight of data and the ACK clock were lost, so this deserves a cautious response. The sender then retransmits the rest of the data following the slow start algorithm [RFC5681]. The time elapsed in RTO recovery is one RTO interval plus the number of round trips needed to repair all the losses.¶
Fast recovery is the preferred form of loss recovery because it can potentially recover all losses in the timescale of a single round trip, with only a fractional congestion window reduction. RTO recovery and congestion window reset should ideally be the last resort and should ideally be used only when the entire flight is lost. However, in addition to losing an entire flight of data, the following situations can unnecessarily resort to RTO recovery with traditional TCP loss recovery algorithms [RFC5681] [RFC6675]:¶
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.¶
RACK-TLP allows senders to recover losses more effectively in all three scenarios described in the previous section. There are two design principles behind RACK-TLP. The first principle is to detect losses via ACK events as much as possible, to repair losses at round-trip timescales. The second principle is to gently probe the network to solicit additional ACK feedback, to avoid RTO expiration and subsequent congestion window reset. At a high level, the two principles are implemented in RACK and TLP, respectively.¶
The rationale behind RACK is that if a segment is delivered out of order, then the segments sent chronologically before that were either lost or reordered. This concept is not fundamentally different from those described in [RFC5681], [RFC6675], or [FACK]. RACK's key innovation is using per-segment transmission timestamps and widely deployed SACK [RFC2018] options to conduct time-based inferences instead of inferring losses by counting ACKs or SACKed sequences. Time-based inferences are more robust than DupAck counting approaches because they do not depend on flight size and thus are effective for application-limited traffic.¶
Conceptually, RACK keeps a virtual timer for every data segment sent (including retransmissions). Each timer expires dynamically based on the latest RTT measurements plus an additional delay budget to accommodate potential packet reordering (called the "reordering window"). When a segment's timer expires, RACK marks the corresponding segment as lost for retransmission.¶
In reality, as an algorithm, RACK does not arm a timer for every segment sent because it's not necessary. Instead, the sender records the most recent transmission time of every data segment sent, including retransmissions. For each ACK received, the sender calculates the latest RTT measurement (if eligible) and adjusts the expiration time of every segment sent but not yet delivered. If a segment has expired, RACK marks it as lost.¶
Since the time-based logic of RACK applies equally to retransmissions and original transmissions, it can detect lost retransmissions as well. If a segment has been retransmitted but its most recent (re)transmission timestamp has expired, then, after a reordering window, it's marked as lost.¶
RACK infers losses from ACK feedback; however, in some cases, ACKs are sparse, particularly when the inflight is small or when the losses are high. In some challenging cases, the last few segments in a flight are lost. With the operations described in [RFC5681] or [RFC6675], the sender's RTO would expire and reset the congestion window when, in reality, most of the flight has been delivered.¶
Consider an example where a sender with a large congestion window transmits 100 new data segments after an application write and only the last three segments are lost. Without RACK-TLP, the RTO expires, the sender retransmits the first unacknowledged segment, and the congestion window slow starts from 1. After all the retransmits are acknowledged, the congestion window is increased to 4. The total delivery time for this application transfer is three RTTs plus one RTO, a steep cost given that only a tiny fraction of the flight was lost. If instead the losses had occurred three segments sooner in the flight, then fast recovery would have recovered all losses within one round trip and would have avoided resetting the congestion window.¶
Fast recovery would be preferable in such scenarios; TLP is designed to trigger the feedback RACK needed to enable that. After the last (100th) segment was originally sent, TLP sends the next available (new) segment or retransmits the last (highest-sequenced) segment in two round trips to probe the network, hence the name "Tail Loss Probe". The successful delivery of the probe would solicit an ACK. RACK uses this ACK to detect that the 98th and 99th segments were lost, trigger fast recovery, and retransmit both successfully. The total recovery time is four RTTs, and the congestion window is only partially reduced instead of being fully reset. If the probe was also lost, then the sender would invoke RTO recovery, resetting the congestion window.¶
Upon receiving an ACK indicating a SACKed segment, a sender cannot tell immediately whether that was a result of reordering or loss. It can only distinguish between the two in hindsight if the missing sequence ranges are filled in later without retransmission. Thus, a loss detection algorithm needs to budget some wait time -- a reordering window -- to try to disambiguate packet reordering from packet loss.¶
The reordering window in the DupAck counting approach is implicitly defined as the elapsed time to receive DupThresh SACKed segments or duplicate acknowledgments. This approach is effective if the network reordering degree (in sequence distance) is smaller than DupThresh and at least DupThresh segments after the loss is acknowledged. For cases where the reordering degree is larger than the default DupThresh of 3 packets, one alternative is to dynamically adapt DupThresh based on the FlightSize (e.g., the sender adjusts the DupThresh to half of the FlightSize). However, this does not work well with the following two types of reordering:¶
In the above two cases, the degree of reordering in packet distance is highly variable. This makes the DupAck counting approach ineffective, including dynamic adaptation variants as in [RFC4653]. Instead, the degree of reordering in time difference in such cases is usually within a single round-trip time. This is because the packets either traverse disjoint paths with similar propagation delays or are repaired quickly by the local access technology. Hence, using a time threshold instead of a packet threshold strikes a middle ground, allowing a bounded degree of reordering resilience while still allowing fast recovery. This is the rationale behind the RACK-TLP reordering resilience design.¶
Specifically, RACK-TLP introduces a new dynamic reordering window parameter in time units, and the sender considers a data segment S lost if both of these conditions are met:¶
Note that condition (1) implies at least one round trip of time has elapsed since S has been sent.¶
The RACK reordering window adapts to the measured duration of reordering events within reasonable and specific bounds to disincentivize excessive reordering. More specifically, the sender sets the reordering window as follows:¶
Rules 2 and 3 are required to adapt to reordering caused by dynamics such as the prolonged link-layer loss recovery episodes described earlier. Each increase in the reordering window requires a new round trip where the sender receives a DSACK; thus, depending on the extent of reordering, it may take multiple round trips to fully adapt.¶
For short flows, the low initial reordering window helps recover losses quickly, at the risk of spurious retransmissions. The rationale is that spurious retransmissions for short flows are not expected to produce excessive additional network traffic. For long flows, the design tolerates reordering within a round trip. This handles reordering in small timescales (reordering within the round-trip time of the shortest path).¶
However, the fact that the initial reordering window is low and the reordering window's adaptive growth is bounded means that there will continue to be a cost to reordering that disincentivizes excessive reordering.¶
The following example in Figure 1 illustrates the RACK-TLP algorithm in action:¶
Figure 1 illustrates a sender sending four segments (P0, P1, P2, P3) and losing the last three segments. After two round trips, TLP sends a loss probe, retransmitting the last segment, P3, to solicit SACK feedback and restore the ACK clock (Event 3). The delivery of P3 enables RACK to infer (Event 5b) that P1 and P2 were likely lost because they were sent before P3. The sender then retransmits P1 and P2. Unfortunately, the retransmission of P1 is lost again. However, the delivery of the retransmission of P2 allows RACK to infer that the retransmission of P1 was likely lost (Event 7a); hence, P1 should be retransmitted (Event 7b). Note that [RFC5681] mandates a principle that loss in two successive windows of data or the loss of a retransmission must be taken as two indications of congestion and therefore results in two separate congestion control reactions.¶
In addition to enhancing fast recovery, RACK improves the accuracy of RTO recovery by reducing spurious retransmissions.¶
Without RACK, upon RTO timer expiration, the sender marks all the unacknowledged segments as lost. This approach can lead to spurious retransmissions. For example, consider a simple case where one segment was sent with an RTO of 1 second and then the application writes more data, causing a second and third segment to be sent right before the RTO of the first segment expires. Suppose none of the segments were lost. Without RACK, if there is a spurious RTO, then the sender marks all three segments as lost and retransmits the first segment. If the ACK for the original copy of the first segment arrives right after the spurious RTO retransmission, then the sender continues slow start and spuriously retransmits the second and third segments since it (erroneously) presumed they are lost.¶
With RACK, upon RTO timer expiration, the only segment automatically marked as lost is the first segment (since it was sent an RTO ago); for all the other segments, RACK only marks the segment as lost if at least one round trip has elapsed since the segment was transmitted. Consider the previous example scenario, but this time with RACK. With RACK, when the RTO expires, the sender only marks the first segment as lost and retransmits that segment. The other two very recently sent segments are not marked as lost because they were sent less than one round trip ago and there were no ACKs providing evidence that they were lost. Upon receiving the ACK for the RTO retransmission, the RACK sender would not yet retransmit the second or third segment, but rather would re-arm the RTO timer and wait for a new RTO interval to elapse before marking the second or third segment as lost.¶
To summarize, RACK-TLP aims to adapt to small time-varying degrees of reordering, quickly recover most losses within one to two round trips, and avoid costly RTO recoveries. In the presence of reordering, the adaptation algorithm can impose sometimes needless delays when it waits to disambiguate loss from reordering, but the penalty for waiting is bounded to one round trip, and such delays are confined to flows long enough to have observed reordering.¶
The reader is expected to be familiar with the definitions given in the TCP congestion control [RFC5681], selective acknowledgment [RFC2018], and loss recovery [RFC6675] RFCs. RACK-TLP has the following requirements:¶
The reader is expected to be familiar with the variables SND.UNA, SND.NXT, SEG.ACK, and SEG.SEQ in [RFC793]; Sender Maximum Segment Size (SMSS) and FlightSize in [RFC5681]; DupThresh in [RFC6675]; and RTO and SRTT in [RFC6298]. A RACK-TLP implementation uses several new terms and needs to store new per-segment and per-connection state, described below.¶
These terms are used to explain the variables and algorithms below:¶
These variables indicate the status of the most recent transmission of a data segment:¶
These timers augment the existing timers maintained by a sender, including the RTO timer [RFC6298]. A RACK-TLP sender arms one of these three timers -- RACK reordering timer, TLP PTO timer, or RTO timer -- when it has unacknowledged segments in flight. The implementation can simplify managing all three timers by multiplexing a single timer among them with an additional variable to indicate the event to invoke upon the next timer expiration.¶
Upon transmitting a new segment or retransmitting an old segment, record the time in Segment.xmit_ts and set Segment.lost to FALSE. Upon retransmitting a segment, set Segment.retransmitted to TRUE.¶
RACK_transmit_new_data(Segment): Segment.xmit_ts = Now() Segment.lost = FALSE RACK_retransmit_data(Segment): Segment.retransmitted = TRUE Segment.xmit_ts = Now() Segment.lost = FALSE¶
Step 1: Update RACK.min_RTT.¶
Use the RTT measurements obtained via [RFC6298] or [RFC7323] to update the estimated minimum RTT in RACK.min_RTT. The sender SHOULD track a windowed min-filtered estimate of recent RTT measurements that can adapt when migrating to significantly longer paths rather than tracking a simple global minimum of all RTT measurements.¶
Step 2: Update the state for the most recently sent segment that has been delivered.¶
In this step, RACK updates the state that tracks the most recently sent segment that has been delivered: RACK.segment. RACK maintains its latest transmission timestamp in RACK.xmit_ts and its highest sequence number in RACK.end_seq. These two variables are used in later steps to estimate if some segments not yet delivered were likely lost. Given the information provided in an ACK, each segment cumulatively ACKed or SACKed is marked as delivered in the scoreboard. Because an ACK can also acknowledge retransmitted data segments and because retransmissions can be spurious, the sender needs to take care to avoid spurious inferences. For example, if the sender were to use timing information from a spurious retransmission, the RACK.rtt could be vastly underestimated.¶
To avoid spurious inferences, ignore a segment as invalid if any of its sequence range has been retransmitted before and if either of two conditions is true:¶
The second check is a heuristic when the TCP Timestamp option is not available or when the round-trip time is less than the TCP Timestamp clock granularity.¶
Among all the segments newly ACKed or SACKed by this ACK that pass the checks above, update the RACK.rtt to be the RTT sample calculated using this ACK. Furthermore, record the most recent Segment.xmit_ts in RACK.xmit_ts if it is ahead of RACK.xmit_ts. If Segment.xmit_ts equals RACK.xmit_ts (e.g., due to clock granularity limits), then compare Segment.end_seq and RACK.end_seq to break the tie when deciding whether to update the RACK.segment's associated state.¶
Step 2 may be summarized in pseudocode as:¶
RACK_sent_after(t1, seq1, t2, seq2): If t1 > t2: Return true Else if t1 == t2 AND seq1 > seq2: Return true Else: Return false RACK_update(): For each Segment newly acknowledged, cumulatively or selectively, in ascending order of Segment.xmit_ts: rtt = Now() - Segment.xmit_ts If Segment.retransmitted is TRUE: If ACK.ts_option.echo_reply < Segment.xmit_ts: Continue If rtt < RACK.min_rtt: Continue RACK.rtt = rtt If RACK_sent_after(Segment.xmit_ts, Segment.end_seq RACK.xmit_ts, RACK.end_seq): RACK.xmit_ts = Segment.xmit_ts RACK.end_seq = Segment.end_seq¶
Step 3: Detect data segment reordering.¶
To detect reordering, the sender looks for original data segments being delivered out of order. To detect such cases, the sender tracks the highest sequence selectively or cumulatively acknowledged in the RACK.fack variable. ".fack" stands for the most "Forward ACK" (this term is adopted from [FACK]). If a never-retransmitted segment that's below RACK.fack is (selectively or cumulatively) acknowledged, it has been delivered out of order. The sender sets RACK.reordering_seen to TRUE if such a segment is identified.¶
RACK_detect_reordering(): For each Segment newly acknowledged, cumulatively or selectively, in ascending order of Segment.end_seq: If Segment.end_seq > RACK.fack: RACK.fack = Segment.end_seq Else if Segment.end_seq < RACK.fack AND Segment.retransmitted is FALSE: RACK.reordering_seen = TRUE¶
Step 4: Update the RACK reordering window.¶
The RACK reordering window, RACK.reo_wnd, serves as an adaptive allowance for settling time before marking a segment as lost. This step documents a detailed algorithm that follows the principles outlined in the "Reordering Window Adaptation" section.¶
If no reordering has been observed based on the previous step, then one way the sender can enter fast recovery is when the number of SACKed segments matches or exceeds DupThresh (similar to [RFC6675]). Furthermore, when no reordering has been observed, the RACK.reo_wnd is set to 0 both upon entering and during fast recovery or RTO recovery.¶
Otherwise, if some reordering has been observed, then RACK does not trigger fast recovery based on DupThresh.¶
Whether or not reordering has been observed, RACK uses the reordering window to assess whether any segments can be marked as lost. As a consequence, the sender also enters fast recovery when there are any number of SACKed segments, as long as the reorder window has passed for some non-SACKed segments.¶
When the reordering window is not set to 0, it starts with a conservative RACK.reo_wnd of RACK.min_RTT/4. This value was chosen because Linux TCP used the same factor in its implementation to delay Early Retransmit [RFC5827] to reduce spurious loss detections in the presence of reordering, and experience showed this worked reasonably well [DMCG11].¶
However, the reordering detection in the previous step, Step 3, has a self-reinforcing drawback when the reordering window is too small to cope with the actual reordering. When that happens, RACK could spuriously mark reordered segments as lost, causing them to be retransmitted. In turn, the retransmissions can prevent the necessary conditions for Step 3 to detect reordering since this mechanism requires ACKs or SACKs only for segments that have never been retransmitted. In some cases, such scenarios can persist, causing RACK to continue to spuriously mark segments as lost without realizing the reordering window is too small.¶
To avoid the issue above, RACK dynamically adapts to higher degrees of reordering using DSACK options from the receiver. Receiving an ACK with a DSACK option indicates a possible spurious retransmission, suggesting that RACK.reo_wnd may be too small. The RACK.reo_wnd increases linearly for every round trip in which the sender receives some DSACK option so that after N round trips in which a DSACK is received, the RACK.reo_wnd becomes (N+1) * min_RTT / 4, with an upper-bound of SRTT.¶
If the reordering is temporary, then a large adapted reordering window would unnecessarily delay loss recovery later. Therefore, RACK persists using the inflated RACK.reo_wnd for up to 16 loss recoveries, after which it resets RACK.reo_wnd to its starting value, min_RTT / 4. The downside of resetting the reordering window is the risk of triggering spurious fast recovery episodes if the reordering remains high. The rationale for this approach is to bound such spurious recoveries to approximately once every 16 recoveries (less than 7%).¶
To track the linear scaling factor for the adaptive reordering window, RACK uses the variable RACK.reo_wnd_mult, which is initialized to 1 and adapts with the observed reordering.¶
The following pseudocode implements the above algorithm for updating the RACK reordering window:¶
RACK_update_reo_wnd(): /* DSACK-based reordering window adaptation */ If RACK.dsack_round is not None AND SND.UNA >= RACK.dsack_round: RACK.dsack_round = None /* Grow the reordering window per round that sees DSACK. Reset the window after 16 DSACK-free recoveries */ If RACK.dsack_round is None AND any DSACK option is present on latest received ACK: RACK.dsack_round = SND.NXT RACK.reo_wnd_mult += 1 RACK.reo_wnd_persist = 16 Else if exiting Fast or RTO recovery: RACK.reo_wnd_persist -= 1 If RACK.reo_wnd_persist <= 0: RACK.reo_wnd_mult = 1 If RACK.reordering_seen is FALSE: If in Fast or RTO recovery: Return 0 Else if RACK.segs_sacked >= DupThresh: Return 0 Return min(RACK.reo_wnd_mult * RACK.min_RTT / 4, SRTT)¶
Step 5: Detect losses.¶
For each segment that has not been SACKed, RACK considers that segment lost if another segment that was sent later has been delivered and the reordering window has passed. RACK considers the reordering window to have passed if the RACK.segment was sent a sufficient time after the segment in question, if a sufficient time has elapsed since the RACK.segment was S/ACKed, or some combination of the two. More precisely, RACK marks a segment as lost if:¶
RACK.xmit_ts >= Segment.xmit_ts AND RACK.xmit_ts - Segment.xmit_ts + (now - RACK.ack_ts) >= RACK.reo_wnd¶
Solving this second condition for "now", the moment at which a segment is marked as lost, yields:¶
now >= Segment.xmit_ts + RACK.reo_wnd + (RACK.ack_ts - RACK.xmit_ts)¶
Then (RACK.ack_ts - RACK.xmit_ts) is the round-trip time of the most recently (re)transmitted segment that's been delivered. When segments are delivered in order, the most recently (re)transmitted segment that's been delivered is also the most recently delivered; hence, RACK.rtt == RACK.ack_ts - RACK.xmit_ts. But if segments were reordered, then the segment delivered most recently was sent before the most recently (re)transmitted segment. Hence, RACK.rtt > (RACK.ack_ts - RACK.xmit_ts).¶
Since RACK.RTT >= (RACK.ack_ts - RACK.xmit_ts), the previous equation reduces to saying that the sender can declare a segment lost when:¶
now >= Segment.xmit_ts + RACK.reo_wnd + RACK.rtt¶
In turn, that is equivalent to stating that a RACK sender should declare a segment lost when:¶
Segment.xmit_ts + RACK.rtt + RACK.reo_wnd - now <= 0¶
Note that if the value on the left-hand side is positive, it represents the remaining wait time before the segment is deemed lost. But this risks a timeout (RTO) if no more ACKs come back (e.g., due to losses or application-limited transmissions) to trigger the marking. For timely loss detection, it is RECOMMENDED that the sender install a reordering timer. This timer expires at the earliest moment when RACK would conclude that all the unacknowledged segments within the reordering window were lost.¶
The following pseudocode implements the algorithm above. When an ACK is received or the RACK reordering timer expires, call RACK_detect_loss_and_arm_timer(). The algorithm breaks timestamp ties by using the TCP sequence space since high-speed networks often have multiple segments with identical timestamps.¶
RACK_detect_loss(): timeout = 0 RACK.reo_wnd = RACK_update_reo_wnd() For each segment, Segment, not acknowledged yet: If RACK_sent_after(RACK.xmit_ts, RACK.end_seq, Segment.xmit_ts, Segment.end_seq): remaining = Segment.xmit_ts + RACK.rtt + RACK.reo_wnd - Now() If remaining <= 0: Segment.lost = TRUE Segment.xmit_ts = INFINITE_TS Else: timeout = max(remaining, timeout) Return timeout RACK_detect_loss_and_arm_timer(): timeout = RACK_detect_loss() If timeout != 0 Arm the RACK timer to call RACK_detect_loss_and_arm_timer() after timeout¶
As an optimization, an implementation can choose to check only segments that have been sent before RACK.xmit_ts. This can be more efficient than scanning the entire SACK scoreboard, especially when there are many segments in flight. The implementation can use a separate doubly linked list ordered by Segment.xmit_ts, insert a segment at the tail of the list when it is (re)transmitted, and remove a segment from the list when it is delivered or marked as lost. In Linux TCP, this optimization improved CPU usage by orders of magnitude during some fast recovery episodes on high-speed WAN networks.¶
Upon RTO timer expiration, RACK marks the first outstanding segment as lost (since it was sent an RTO ago); for all the other segments, RACK only marks the segment as lost if the time elapsed since the segment was transmitted is at least the sum of the recent RTT and the reordering window.¶
RACK_mark_losses_on_RTO(): For each segment, Segment, not acknowledged yet: If SEG.SEQ == SND.UNA OR Segment.xmit_ts + RACK.rtt + RACK.reo_wnd - Now() <= 0: Segment.lost = TRUE¶
Reset TLP.is_retrans and TLP.end_seq when initiating a connection, fast recovery, or RTO recovery.¶
TLP_init(): TLP.end_seq = None TLP.is_retrans = false¶
The sender schedules a loss probe timeout (PTO) to transmit a segment during the normal transmission process. The sender SHOULD start or restart a loss probe PTO timer after transmitting new data (that was not itself a loss probe) or upon receiving an ACK that cumulatively acknowledges new data unless it is already in fast recovery, RTO recovery, or segments have been SACKed (i.e., RACK.segs_sacked is not zero). These conditions are excluded because they are addressed by similar mechanisms, like Limited Transmit [RFC3042], the RACK reordering timer, and Forward RTO-Recovery (F-RTO) [RFC5682].¶
The sender calculates the PTO interval by taking into account a number of factors.¶
First, the default PTO interval is 2*SRTT. By that time, it is prudent to declare that an ACK is overdue since under normal circumstances, i.e., no losses, an ACK typically arrives in one SRTT. Choosing the PTO to be exactly an SRTT would risk causing spurious probes given that network and end-host delay variance can cause an ACK to be delayed beyond the SRTT. Hence, the PTO is conservatively chosen to be the next integral multiple of SRTT.¶
Second, when there is no SRTT estimate available, the PTO SHOULD be 1 second. This conservative value corresponds to the RTO value when no SRTT is available, per [RFC6298].¶
Third, when the FlightSize is one segment, the sender MAY inflate the PTO by TLP.max_ack_delay to accommodate a potentially delayed acknowledgment and reduce the risk of spurious retransmissions. The actual value of TLP.max_ack_delay is implementation specific.¶
Finally, if the time at which an RTO would fire (here denoted as "TCP_RTO_expiration()") is sooner than the computed time for the PTO, then the sender schedules a TLP to be sent at that RTO time.¶
Summarizing these considerations in pseudocode form, a sender SHOULD use the following logic to select the duration of a PTO:¶
TLP_calc_PTO(): If SRTT is available: PTO = 2 * SRTT If FlightSize is one segment: PTO += TLP.max_ack_delay Else: PTO = 1 sec If Now() + PTO > TCP_RTO_expiration(): PTO = TCP_RTO_expiration() - Now()¶
When the PTO timer expires, the sender MUST check whether both of the following conditions are met before sending a loss probe:¶
If either one of these two conditions is not met, then the sender MUST skip sending a loss probe and MUST proceed to re-arm the RTO timer, as specified at the end of this section.¶
If both conditions are met, then the sender SHOULD transmit a previously unsent data segment, if one exists and the receive window allows, and increment the FlightSize accordingly. Note that the FlightSize could be one packet greater than the congestion window temporarily until the next ACK arrives.¶
If such an unsent segment is not available, then the sender SHOULD retransmit the highest-sequence segment sent so far and set TLP.is_retrans to true. This segment is chosen to deal with the retransmission ambiguity problem in TCP. Suppose a sender sends N segments and then retransmits the last segment (segment N) as a loss probe, after which the sender receives a SACK for segment N. As long as the sender waits for the RACK reordering window to expire, it doesn't matter if that SACK was for the original transmission of segment N or the TLP retransmission; in either case, the arrival of the SACK for segment N provides evidence that the N-1 segments preceding segment N were likely lost.¶
In a case where there is only one original outstanding segment of data (N=1), the same logic (trivially) applies: an ACK for a single outstanding segment tells the sender that the N-1=0 segments preceding that segment were lost. Furthermore, whether there are N>1 or N=1 outstanding segments, there is a question about whether the original last segment or its TLP retransmission were lost; the sender estimates whether there was such a loss using TLP recovery detection (see below).¶
The sender MUST follow the RACK transmission procedures in the "Upon Transmitting a Data Segment" section upon sending either a retransmission or a new data loss probe. This is critical for detecting losses using the ACK for the loss probe.¶
After attempting to send a loss probe, regardless of whether a loss probe was sent, the sender MUST re-arm the RTO timer, not the PTO timer, if the FlightSize is not zero. This ensures RTO recovery remains the last resort if TLP fails. The following pseudocode summarizes the operations.¶
TLP_send_probe(): If TLP.end_seq is None and Sender has taken a new RTT sample since last probe or the start of connection: TLP.is_retrans = false Segment = send buffer segment starting at SND.NXT If Segment exists and fits the peer receive window limit: /* Transmit the lowest-sequence unsent Segment */ Transmit Segment RACK_transmit_data(Segment) TLP.end_seq = SND.NXT Increase FlightSize by Segment length Else: /* Retransmit the highest-sequence Segment sent */ Segment = send buffer segment ending at SND.NXT Transmit Segment RACK_retransmit_data(Segment) TLP.end_seq = SND.NXT TLP.is_retrans = true If FlightSize is not zero: Rearm RTO timer to fire at timeout = now + RTO¶
When there is packet loss in a flight ending with a loss probe, the feedback solicited by a loss probe will reveal one of two scenarios, depending on the pattern of losses.¶
If the loss probe and the ACK that acknowledges the probe are delivered successfully, RACK-TLP uses this ACK -- just as it would with any other ACK -- to detect if any segments sent prior to the probe were dropped. RACK would typically infer that any unacknowledged data segments sent before the loss probe were lost, since they were sent sufficiently far in the past (where at least one PTO has elapsed, plus one round trip for the loss probe to be ACKed). More specifically, RACK_detect_loss() (Step 5) would mark those earlier segments as lost. Then the sender would trigger a fast recovery to recover those losses.¶
If the TLP retransmission repairs all the lost in-flight sequence ranges (i.e., only the last segment in the flight was lost), the ACK for the loss probe appears to be a regular cumulative ACK, which would not normally trigger the congestion control response to this packet loss event. The following TLP recovery detection mechanism examines ACKs to detect this special case to make congestion control respond properly [RFC5681].¶
After a TLP retransmission, the sender checks for this special case of a single loss that is recovered by the loss probe itself. To accomplish this, the sender checks for a duplicate ACK or DSACK indicating that both the original segment and TLP retransmission arrived at the receiver, which means there was no loss. If the TLP sender does not receive such an indication, then it MUST assume that the original data segment, the TLP retransmission, or a corresponding ACK was lost for congestion control purposes.¶
If the TLP retransmission is spurious, a receiver that uses DSACK would return an ACK that covers TLP.end_seq with a DSACK option (Case 1). If the receiver does not support DSACK, it would return a DupAck without any SACK option (Case 2). If the sender receives an ACK matching either case, then the sender estimates that the receiver received both the original data segment and the TLP probe retransmission. The sender considers the TLP episode to be done and records that fact by setting TLP.end_seq to None.¶
Upon receiving an ACK that covers some sequence number after TLP.end_seq, the sender should have received any ACKs for the original segment and TLP probe retransmission segment. At that time, if the TLP.end_seq is still set and thus indicates that the TLP probe retransmission remains unacknowledged, then the sender should presume that at least one of its data segments was lost. The sender then SHOULD invoke a congestion control response equivalent to a fast recovery.¶
More precisely, on each ACK, the sender executes the following:¶
TLP_process_ack(ACK): If TLP.end_seq is not None AND ACK's ack. number >= TLP.end_seq: If not TLP.is_retrans: TLP.end_seq = None /* TLP of new data delivered */ Else if ACK has a DSACK option matching TLP.end_seq: TLP.end_seq = None /* Case 1, above */ Else If ACK's ack. number > TLP.end_seq: TLP.end_seq = None /* Repaired the single loss */ (Invoke congestion control to react to the loss event the probe has repaired) Else If ACK is a DupAck without any SACK option: TLP.end_seq = None /* Case 2, above */¶
The RACK reordering timer, the TLP PTO timer, the RTO, and Zero Window Probe (ZWP) timer [RFC793] are mutually exclusive and are used in different scenarios. When arming a RACK reordering timer or TLP PTO timer, the sender SHOULD cancel any other pending timers. An implementation is expected to have one timer with an additional state variable indicating the type of the timer.¶
The biggest advantage of RACK-TLP is that every data segment, whether it is an original data transmission or a retransmission, can be used to detect losses of the segments sent chronologically prior to it. This enables RACK-TLP to use fast recovery in cases with application-limited flights of data, lost retransmissions, or data segment reordering events. Consider the following examples:¶
The examples above show that RACK-TLP is particularly useful when the sender is limited by the application, which can happen with interactive or request/response traffic. Similarly, RACK still works when the sender is limited by the receive window, which can happen with applications that use the receive window to throttle the sender.¶
RACK-TLP works more efficiently with TCP Segmentation Offload (TSO) compared to DupAck counting. RACK always marks the entire TSO aggregate as lost because the segments in the same TSO aggregate have the same transmission timestamp. By contrast, the algorithms based on sequence counting (e.g., [RFC6675], [RFC5681]) may mark only a subset of segments in the TSO aggregate as lost, forcing the stack to perform expensive fragmentation of the TSO aggregate or to selectively tag individual segments as lost in the scoreboard.¶
The main drawback of RACK-TLP is the additional state required compared to DupAck counting. RACK requires the sender to record the transmission time of each segment sent at a clock granularity that is finer than 1/4 of the minimum RTT of the connection. TCP implementations that already record this for RTT estimation do not require any new per-packet state. But implementations that are not yet recording segment transmission times will need to add per-packet internal state (expected to be either 4 or 8 octets per segment or TSO aggregate) to track transmission times. In contrast, the loss detection approach described in [RFC6675] does not require any per-packet state beyond the SACK scoreboard; this is particularly useful on ultra-low RTT networks where the RTT may be less than the sender TCP clock granularity (e.g., inside data centers). Another disadvantage is that the reordering timer may expire prematurely (like any other retransmission timer) and cause higher spurious retransmissions, especially if DSACK is not supported.¶
The primary motivation of RACK-TLP is to provide a general alternative to some of the standard loss recovery algorithms [RFC5681] [RFC6675] [RFC5827] [RFC4653]. In particular, the SACK loss recovery algorithm for TCP [RFC6675] is not designed to handle lost retransmissions, so its NextSeg() does not work for lost retransmissions, and it does not specify the corresponding required additional congestion response. Therefore, the algorithm [RFC6675] MUST NOT be used with RACK-TLP; instead, a modified recovery algorithm that carefully addresses such a case is needed.¶
The Early Retransmit mechanism [RFC5827] and NCR for TCP [RFC4653] dynamically adjust the duplicate ACK threshold based on the current or previous flight sizes. RACK-TLP takes a different approach by using a time-based reordering window. RACK-TLP can be seen as an extended Early Retransmit [RFC5827] without a FlightSize limit but with an additional reordering window. [FACK] considers an original segment to be lost when its sequence range is sufficiently far below the highest SACKed sequence. In some sense, RACK-TLP can be seen as a generalized form of FACK that operates in time space instead of sequence space, enabling it to better handle reordering, application-limited traffic, and lost retransmissions.¶
RACK-TLP is compatible with the standard RTO [RFC6298], RTO Restart [RFC7765], F-RTO [RFC5682], and Eifel algorithms [RFC3522]. This is because RACK-TLP only detects loss by using ACK events. It neither changes the RTO timer calculation nor detects spurious RTOs. RACK-TLP slightly changes the behavior of [RFC6298] by preceding the RTO with a TLP and reducing potential spurious retransmissions after RTO.¶
RACK-TLP intentionally decouples loss detection from congestion control. RACK-TLP only detects losses; it does not modify the congestion control algorithm [RFC5681] [RFC6937]. A segment marked as lost by RACK-TLP MUST NOT be retransmitted until congestion control deems this appropriate. As mentioned in the paragraph following Figure 1 (Section 3.4, Paragraph 3), [RFC5681] mandates a principle that loss in two successive windows of data or the loss of a retransmission must be taken as two indications of congestion and therefore trigger two separate reactions. The Proportional Rate Reduction (PRR) algorithm [RFC6937] is RECOMMENDED for the specific congestion control actions taken upon the losses detected by RACK-TLP. In the absence of PRR [RFC6937], when RACK-TLP detects a lost retransmission, the congestion control MUST trigger an additional congestion response per the aforementioned principle in [RFC5681]. If multiple original transmissions or retransmissions were lost in a window, the congestion control specified in [RFC5681] only reacts once per window. The congestion control implementer is advised to carefully consider this subtle situation introduced by RACK-TLP.¶
The only exception -- the only way in which RACK-TLP modulates the congestion control algorithm -- is that one outstanding loss probe can be sent even if the congestion window is fully used. However, this temporary overcommit is accounted for and credited in the in-flight data tracked for congestion control, so that congestion control will erase the overcommit upon the next ACK.¶
If packet losses happen after reordering has been observed, RACK-TLP may take longer to detect losses than the pure DupAck counting approach. In this case, TCP may continue to increase the congestion window upon receiving ACKs during this time, making the sender more aggressive.¶
The following simple example compares how RACK-TLP and non-RACK-TLP loss detection interact with congestion control: suppose a sender has a congestion window (cwnd) of 20 segments on a SACK-enabled connection. It sends 10 data segments, and all of them are lost.¶
Without RACK-TLP, the sender would time out, reset cwnd to 1, and retransmit the first segment. It would take four round trips (1 + 2 + 4 + 3 = 10) to retransmit all the 10 lost segments using slow start. The recovery latency would be RTO + 4*RTT, with an ending cwnd of 4 segments due to congestion window validation.¶
With RACK-TLP, a sender would send the TLP after 2*RTT and get a DupAck, enabling RACK to detect the losses and trigger fast recovery. If the sender implements Proportional Rate Reduction [RFC6937], it would slow start to retransmit the remaining 9 lost segments since the number of segments in flight (0) is lower than the slow start threshold (10). The slow start would again take four round trips (1 + 2 + 4 + 3 = 10) to retransmit all the lost segments. The recovery latency would be 2*RTT + 4*RTT, with an ending cwnd set to the slow-start threshold of 10 segments.¶
The difference in recovery latency (RTO + 4*RTT vs 6*RTT) can be significant if the RTT is much smaller than the minimum RTO (1 second in [RFC6298]) or if the RTT is large. The former case can happen in local area networks, data center networks, or content distribution networks with deep deployments. The latter case can happen in developing regions with highly congested and/or high-latency networks.¶
Delayed or stretched ACKs complicate the detection of repairs done by TLP since, with such ACKs, the sender takes a longer time to receive fewer ACKs than would normally be expected. To mitigate this complication, before sending a TLP loss probe retransmission, the sender should attempt to wait long enough that the receiver has sent any delayed ACKs that it is withholding. The sender algorithm described above features such a delay in the form of TLP.max_ack_delay. Furthermore, if the receiver supports DSACK, then, in the case of a delayed ACK, the sender's TLP recovery detection mechanism (see above) can use the DSACK information to infer that the original and TLP retransmission both arrived at the receiver.¶
If there is ACK loss or a delayed ACK without a DSACK, then this algorithm is conservative because the sender will reduce the congestion window when, in fact, there was no packet loss. In practice, this is acceptable and potentially even desirable: if there is reverse path congestion, then reducing the congestion window can be prudent.¶
RACK-TLP can be implemented in other transport protocols (e.g., [QUIC-LR]). The [SPROUT] loss detection algorithm was also independently designed to use a 10 ms reordering window to improve its loss detection similar to RACK.¶
RACK-TLP algorithm behavior is based on information conveyed in SACK options, so it has security considerations similar to those described in the Security Considerations section of [RFC6675].¶
Additionally, RACK-TLP has a lower risk profile than the loss recovery algorithm [RFC6675] because it is not vulnerable to ACK-splitting attacks [SCWA99]: for an MSS-sized segment sent, the receiver or the attacker might send MSS ACKs that selectively or cumulatively acknowledge one additional byte per ACK. This would not fool RACK. In such a scenario, RACK.xmit_ts would not advance because all the sequence ranges within the segment were transmitted at the same time and thus carry the same transmission timestamp. In other words, SACKing only one byte of a segment or SACKing the segment in entirety have the same effect with RACK.¶
This document has no IANA actions.¶
The authors thank Matt Mathis for his insights in FACK and Michael Welzl for his per-packet timer idea that inspired this work. Eric Dumazet, Randy Stewart, Van Jacobson, Ian Swett, Rick Jones, Jana Iyengar, Hiren Panchasara, Praveen Balasubramanian, Yoshifumi Nishida, Bob Briscoe, Felix Weinrank, Michael Tüxen, Martin Duke, Ilpo Jarvinen, Theresa Enghardt, Mirja Kühlewind, Gorry Fairhurst, Markku Kojo, and Yi Huang contributed to this document or the implementations in Linux, FreeBSD, Windows, and QUIC.¶