From 2ac9808a0953a9c12dfe5d81319bea565210a19c Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Beno=C3=AEt=20CORTIER?= Date: Fri, 16 Aug 2024 02:26:41 +0900 Subject: [PATCH] perf(jetsocat): increase the maximum JMUX message size This has almost no effect on the throughput when there is a significant delay, but the throughput is improved when the delay is very small to non-inexistant. The main benefit is a reduced CPU-usage. 1. Benchmark results before this patch a. With 50ms delay on loopback 1 connection: [ 1] 0.0000-600.4197 sec 16.1 GBytes 230 Mbits/sec 2 connections: [ 1] 0.0000-605.0387 sec 8.19 GBytes 116 Mbits/sec [ 2] 0.0000-605.1395 sec 8.19 GBytes 116 Mbits/sec [SUM] 0.0000-605.1395 sec 16.4 GBytes 233 Mbits/sec 10 connections: [ 3] 0.0000-625.7966 sec 1.69 GBytes 23.2 Mbits/sec [ 8] 0.0000-625.9956 sec 1.69 GBytes 23.2 Mbits/sec [ 1] 0.0000-626.0966 sec 1.69 GBytes 23.2 Mbits/sec [ 5] 0.0000-626.0964 sec 1.69 GBytes 23.2 Mbits/sec [ 2] 0.0000-626.1983 sec 1.69 GBytes 23.2 Mbits/sec [ 7] 0.0000-626.1964 sec 1.69 GBytes 23.2 Mbits/sec [ 6] 0.0000-626.1964 sec 1.69 GBytes 23.2 Mbits/sec [ 9] 0.0000-626.1981 sec 1.69 GBytes 23.2 Mbits/sec [ 10] 0.0000-626.2973 sec 1.69 GBytes 23.2 Mbits/sec [ 4] 0.0000-626.3984 sec 1.69 GBytes 23.2 Mbits/sec [SUM] 0.0000-626.3986 sec 16.9 GBytes 232 Mbits/sec b. Without delay 1 connection: [ 1] 0.0000-600.0518 sec 1.33 TBytes 19.4 Gbits/sec 2 connections: [ 2] 0.0000-600.0706 sec 681 GBytes 9.75 Gbits/sec [ 1] 0.0000-600.0705 sec 681 GBytes 9.75 Gbits/sec [SUM] 0.0000-600.0705 sec 1.33 TBytes 19.5 Gbits/sec 10 connections: [ 3] 0.0000-600.3608 sec 112 GBytes 1.60 Gbits/sec [ 5] 0.0000-600.3606 sec 112 GBytes 1.60 Gbits/sec [ 6] 0.0000-600.3605 sec 112 GBytes 1.60 Gbits/sec [ 8] 0.0000-600.3598 sec 112 GBytes 1.60 Gbits/sec [ 7] 0.0000-600.3594 sec 112 GBytes 1.60 Gbits/sec [ 1] 0.0000-600.3606 sec 112 GBytes 1.60 Gbits/sec [ 9] 0.0000-600.3597 sec 112 GBytes 1.60 Gbits/sec [ 10] 0.0000-600.3606 sec 112 GBytes 1.60 Gbits/sec [ 2] 0.0000-600.3602 sec 112 GBytes 1.60 Gbits/sec [ 4] 0.0000-600.3719 sec 112 GBytes 1.60 Gbits/sec [SUM] 0.0000-600.3721 sec 1.09 TBytes 16.0 Gbits/sec 2. Benchmark results after this patch a. With 50ms delay on loopback 1 connection: [ 1] 0.0000-600.4552 sec 16.1 GBytes 231 Mbits/sec 2 connections: [ 1] 0.0000-605.1600 sec 8.16 GBytes 116 Mbits/sec [ 2] 0.0000-605.1599 sec 8.16 GBytes 116 Mbits/sec [SUM] 0.0000-605.1599 sec 16.3 GBytes 232 Mbits/sec 10 connections: [ 8] 0.0000-625.8346 sec 1.69 GBytes 23.2 Mbits/sec [ 9] 0.0000-626.1828 sec 1.69 GBytes 23.2 Mbits/sec [ 2] 0.0000-626.1820 sec 1.69 GBytes 23.2 Mbits/sec [ 5] 0.0000-626.1817 sec 1.69 GBytes 23.2 Mbits/sec [ 6] 0.0000-626.1815 sec 1.69 GBytes 23.2 Mbits/sec [ 4] 0.0000-626.1827 sec 1.69 GBytes 23.2 Mbits/sec [ 3] 0.0000-626.1814 sec 1.69 GBytes 23.2 Mbits/sec [ 7] 0.0000-626.1821 sec 1.69 GBytes 23.2 Mbits/sec [ 1] 0.0000-626.2831 sec 1.69 GBytes 23.1 Mbits/sec [ 10] 0.0000-626.2819 sec 1.69 GBytes 23.1 Mbits/sec [SUM] 0.0000-626.2832 sec 16.9 GBytes 232 Mbits/sec b. Without delay 1 connection: [ 1] 0.0000-600.0402 sec 1.68 TBytes 24.6 Gbits/sec 2 connections: [ 1] 0.0000-600.0628 sec 752 GBytes 10.8 Gbits/sec [ 2] 0.0000-601.0794 sec 751 GBytes 10.7 Gbits/sec [SUM] 0.0000-601.0794 sec 1.47 TBytes 21.5 Gbits/sec 10 connections: [ 6] 0.0000-600.3015 sec 127 GBytes 1.82 Gbits/sec [ 3] 0.0000-600.3014 sec 127 GBytes 1.82 Gbits/sec [ 7] 0.0000-600.3012 sec 127 GBytes 1.82 Gbits/sec [ 5] 0.0000-600.2992 sec 127 GBytes 1.82 Gbits/sec [ 9] 0.0000-600.3014 sec 127 GBytes 1.82 Gbits/sec [ 1] 0.0000-600.3006 sec 127 GBytes 1.82 Gbits/sec [ 2] 0.0000-600.3601 sec 127 GBytes 1.82 Gbits/sec [ 10] 0.0000-600.3592 sec 127 GBytes 1.82 Gbits/sec [ 8] 0.0000-600.3604 sec 127 GBytes 1.82 Gbits/sec [ 4] 0.0000-600.3586 sec 127 GBytes 1.82 Gbits/sec [SUM] 0.0000-600.3605 sec 1.24 TBytes 18.2 Gbits/sec --- crates/jmux-proxy/src/lib.rs | 9 +++- docs/JMUX-proxy-performance.md | 86 ++++++++++++++++++++++++++++++++-- 2 files changed, 88 insertions(+), 7 deletions(-) diff --git a/crates/jmux-proxy/src/lib.rs b/crates/jmux-proxy/src/lib.rs index 3606ba095..bbfb6ac5f 100644 --- a/crates/jmux-proxy/src/lib.rs +++ b/crates/jmux-proxy/src/lib.rs @@ -30,7 +30,8 @@ use tokio::task::JoinHandle; use tokio_util::codec::FramedRead; use tracing::{Instrument as _, Span}; -const MAXIMUM_PACKET_SIZE_IN_BYTES: u16 = 4 * 1024; // 4 kiB +const DATA_PACKET_OVERHEAD: u16 = 8; +const MAXIMUM_PACKET_SIZE_IN_BYTES: u16 = 8 * 1024 + DATA_PACKET_OVERHEAD; // 8 kiB + packet overhead const WINDOW_ADJUSTMENT_THRESHOLD: u32 = 4 * 1024; // 4 kiB pub type ApiResponseSender = oneshot::Sender; @@ -777,7 +778,11 @@ impl DataReaderTask { } = self; let codec = tokio_util::codec::BytesCodec::new(); - let mut bytes_stream = FramedRead::new(reader, codec); + let mut bytes_stream = FramedRead::with_capacity( + reader, + codec, + usize::from(MAXIMUM_PACKET_SIZE_IN_BYTES - DATA_PACKET_OVERHEAD), + ); let maximum_packet_size = usize::from(maximum_packet_size); trace!("Started forwarding"); diff --git a/docs/JMUX-proxy-performance.md b/docs/JMUX-proxy-performance.md index 693d6b8b7..bc6602482 100644 --- a/docs/JMUX-proxy-performance.md +++ b/docs/JMUX-proxy-performance.md @@ -70,10 +70,9 @@ iperf -c "$ADDR" -p $PORT -P 10 -t 600 Let’s assume the script is in a file named `run_iperf.sh`. -Running `iperf` for long enough is important to ensure that the buffering happening at the socket level is not influencing the numbers too much. -When running less a minute, we end up measuring the rate at which `iperf` enqueue bytes into the socket’s buffer. -Filling the buffer can be done very quickly and can have a significant impact on the measured average speed. -10 minutes is long enough to obtain convergent results. +It's important to note that `iperf` should be run for an extended period to account for the initial filling of TCP socket buffers, +which can artificially inflate the average throughput if tested for less than a minute. +Running `iperf` for 10 minutes is enough to ensure the results accurately reflect the effective average throughput. ## Applied optimizations @@ -274,7 +273,7 @@ The flow control algorithm, particularly the window size, is a critical paramete Since such delays are common in almost all practical setups, it’s safe to say that this is the most important metric to optimize. Other optimizations, while beneficial, primarily serve to reduce CPU usage and increase throughput on very high-speed networks. -A speed of 30 Mbits/s is already considered high, but networks with throughput exceeding 1 Gbits/s also exist. +A speed of 30 Mbits/s is already considered high, but networks with throughput exceeding 1 Gbits/s also exist (e.g.: ultra-high speed local area networks). Enhancing performance for these networks is valuable, particularly in reducing CPU usage as the volume of data processed increases. Measurements indicate that our JMUX proxy should perform well, even on high-speed networks. @@ -286,3 +285,80 @@ In real-world wide-area networks, packet loss will inevitably occur. Nevertheless, these results provide valuable data, confirming that our optimizations are effective with a high degree of confidence. While further optimization could be pursued to address more specific scenarios, the current implementation is likely sufficient for most practical purposes. + +## 2025.2.0 update + +Related patches: + +- +- + +### Results + +```shell +./run_iperf.sh 5000 +``` + +#### With 50ms delay on loopback + +1 connection: + +``` +[ 1] 0.0000-600.4552 sec 16.1 GBytes 231 Mbits/sec +``` + +2 connections: + +``` +[ 1] 0.0000-605.1600 sec 8.16 GBytes 116 Mbits/sec +[ 2] 0.0000-605.1599 sec 8.16 GBytes 116 Mbits/sec +[SUM] 0.0000-605.1599 sec 16.3 GBytes 232 Mbits/sec +``` + +10 connections: + +``` +[ 8] 0.0000-625.8346 sec 1.69 GBytes 23.2 Mbits/sec +[ 9] 0.0000-626.1828 sec 1.69 GBytes 23.2 Mbits/sec +[ 2] 0.0000-626.1820 sec 1.69 GBytes 23.2 Mbits/sec +[ 5] 0.0000-626.1817 sec 1.69 GBytes 23.2 Mbits/sec +[ 6] 0.0000-626.1815 sec 1.69 GBytes 23.2 Mbits/sec +[ 4] 0.0000-626.1827 sec 1.69 GBytes 23.2 Mbits/sec +[ 3] 0.0000-626.1814 sec 1.69 GBytes 23.2 Mbits/sec +[ 7] 0.0000-626.1821 sec 1.69 GBytes 23.2 Mbits/sec +[ 1] 0.0000-626.2831 sec 1.69 GBytes 23.1 Mbits/sec +[ 10] 0.0000-626.2819 sec 1.69 GBytes 23.1 Mbits/sec +[SUM] 0.0000-626.2832 sec 16.9 GBytes 232 Mbits/sec +``` + +#### Without delay + +1 connection: + +``` +[ 1] 0.0000-600.0402 sec 1.68 TBytes 24.6 Gbits/sec +``` + +2 connections: + +``` +[ 1] 0.0000-600.0628 sec 752 GBytes 10.8 Gbits/sec +[ 2] 0.0000-601.0794 sec 751 GBytes 10.7 Gbits/sec +[SUM] 0.0000-601.0794 sec 1.47 TBytes 21.5 Gbits/sec +``` + +10 connections: + +``` +[ 6] 0.0000-600.3015 sec 127 GBytes 1.82 Gbits/sec +[ 3] 0.0000-600.3014 sec 127 GBytes 1.82 Gbits/sec +[ 7] 0.0000-600.3012 sec 127 GBytes 1.82 Gbits/sec +[ 5] 0.0000-600.2992 sec 127 GBytes 1.82 Gbits/sec +[ 9] 0.0000-600.3014 sec 127 GBytes 1.82 Gbits/sec +[ 1] 0.0000-600.3006 sec 127 GBytes 1.82 Gbits/sec +[ 2] 0.0000-600.3601 sec 127 GBytes 1.82 Gbits/sec +[ 10] 0.0000-600.3592 sec 127 GBytes 1.82 Gbits/sec +[ 8] 0.0000-600.3604 sec 127 GBytes 1.82 Gbits/sec +[ 4] 0.0000-600.3586 sec 127 GBytes 1.82 Gbits/sec +[SUM] 0.0000-600.3605 sec 1.24 TBytes 18.2 Gbits/sec +```