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. ------------------ Before this patch: ------------------ => With 50ms delay on loopback ==> 1 connection [ 1] 0.0000-600.4660 sec 16.0 GBytes 229 Mbits/sec ==> 2 connections [ 2] 0.0000-605.0918 sec 8.18 GBytes 116 Mbits/sec [ 1] 0.0000-605.1930 sec 8.18 GBytes 116 Mbits/sec [SUM] 0.0000-605.1930 sec 16.4 GBytes 232 Mbits/sec ==> 10 connections [ 9] 0.0000-626.1216 sec 1.69 GBytes 23.1 Mbits/sec [ 10] 0.0000-626.2644 sec 1.69 GBytes 23.1 Mbits/sec [ 1] 0.0000-626.2643 sec 1.69 GBytes 23.1 Mbits/sec [ 3] 0.0000-626.2633 sec 1.69 GBytes 23.1 Mbits/sec [ 8] 0.0000-626.2619 sec 1.69 GBytes 23.2 Mbits/sec [ 6] 0.0000-626.3628 sec 1.69 GBytes 23.1 Mbits/sec [ 7] 0.0000-626.3644 sec 1.69 GBytes 23.1 Mbits/sec [ 5] 0.0000-626.3640 sec 1.69 GBytes 23.2 Mbits/sec [ 4] 0.0000-626.3627 sec 1.69 GBytes 23.2 Mbits/sec [ 2] 0.0000-626.4901 sec 1.69 GBytes 23.1 Mbits/sec [SUM] 0.0000-626.4902 sec 16.9 GBytes 231 Mbits/sec => Without delay ==> 1 connection [ 1] 0.0000-600.0847 sec 1.28 TBytes 18.8 Gbits/sec ==> 2 connections [ 1] 0.0000-600.0795 sec 656 GBytes 9.39 Gbits/sec [ 2] 0.0000-600.0958 sec 656 GBytes 9.38 Gbits/sec [SUM] 0.0000-600.0958 sec 1.28 TBytes 18.8 Gbits/sec ==> 10 connections [ 8] 0.0000-600.3803 sec 108 GBytes 1.54 Gbits/sec [ 4] 0.0000-600.3804 sec 108 GBytes 1.54 Gbits/sec [ 2] 0.0000-600.3789 sec 108 GBytes 1.54 Gbits/sec [ 6] 0.0000-600.3795 sec 108 GBytes 1.54 Gbits/sec [ 3] 0.0000-600.3793 sec 108 GBytes 1.54 Gbits/sec [ 9] 0.0000-600.3789 sec 108 GBytes 1.54 Gbits/sec [ 10] 0.0000-600.3802 sec 108 GBytes 1.54 Gbits/sec [ 5] 0.0000-600.3791 sec 108 GBytes 1.54 Gbits/sec [ 1] 0.0000-600.3794 sec 108 GBytes 1.54 Gbits/sec [ 7] 0.0000-600.3803 sec 108 GBytes 1.54 Gbits/sec [SUM] 0.0000-600.3803 sec 1.05 TBytes 15.4 Gbits/sec ----------------- After this patch: ----------------- => With 50ms delay on loopback ==> 1 connection [ 1] 0.0000-600.5881 sec 16.0 GBytes 229 Mbits/sec ==> 2 connections [ 2] 0.0000-605.1890 sec 8.06 GBytes 114 Mbits/sec [ 1] 0.0000-605.2897 sec 8.06 GBytes 114 Mbits/sec [SUM] 0.0000-605.2898 sec 16.1 GBytes 229 Mbits/sec ==> 10 connections [ 3] 0.0000-626.0706 sec 1.68 GBytes 23.0 Mbits/sec [ 6] 0.0000-626.3093 sec 1.68 GBytes 23.0 Mbits/sec [ 9] 0.0000-626.3096 sec 1.68 GBytes 23.0 Mbits/sec [ 10] 0.0000-626.3078 sec 1.68 GBytes 23.0 Mbits/sec [ 8] 0.0000-626.3076 sec 1.68 GBytes 23.0 Mbits/sec [ 1] 0.0000-626.3096 sec 1.67 GBytes 22.9 Mbits/sec [ 2] 0.0000-626.4217 sec 1.68 GBytes 23.0 Mbits/sec [ 4] 0.0000-626.4200 sec 1.68 GBytes 23.0 Mbits/sec [ 5] 0.0000-626.4215 sec 1.67 GBytes 23.0 Mbits/sec [ 7] 0.0000-626.5331 sec 1.68 GBytes 23.0 Mbits/sec [SUM] 0.0000-626.5333 sec 16.8 GBytes 230 Mbits/sec => Without delay ==> 1 connection [ 1] 0.0000-600.0912 sec 1.67 TBytes 24.5 Gbits/sec ==> 2 connections [ 1] 0.0000-600.0679 sec 747 GBytes 10.7 Gbits/sec [ 2] 0.0000-601.0850 sec 747 GBytes 10.7 Gbits/sec [SUM] 0.0000-601.0851 sec 1.46 TBytes 21.3 Gbits/sec ==> 10 connections [ 9] 0.0000-600.2833 sec 127 GBytes 1.82 Gbits/sec [ 4] 0.0000-600.2828 sec 127 GBytes 1.82 Gbits/sec [ 10] 0.0000-600.3096 sec 127 GBytes 1.82 Gbits/sec [ 7] 0.0000-600.3094 sec 127 GBytes 1.82 Gbits/sec [ 5] 0.0000-600.3092 sec 127 GBytes 1.82 Gbits/sec [ 3] 0.0000-600.3100 sec 127 GBytes 1.82 Gbits/sec [ 8] 0.0000-600.3096 sec 127 GBytes 1.82 Gbits/sec [ 1] 0.0000-600.3093 sec 127 GBytes 1.82 Gbits/sec [ 6] 0.0000-600.3095 sec 127 GBytes 1.82 Gbits/sec [ 2] 0.0000-600.3091 sec 127 GBytes 1.82 Gbits/sec [SUM] 0.0000-600.3098 sec 1.24 TBytes 18.2 Gbits/sec
Devolutions · Aug 17, 2024 · bc758f2 · bc758f2
1 parent ac60d0e
commit bc758f2
Showing 1 changed file with 7 additions and 12 deletions.
diff --git a/crates/jmux-proxy/src/lib.rs b/crates/jmux-proxy/src/lib.rs
@@ -30,17 +30,8 @@ use tokio::task::JoinHandle;
 use tokio_util::codec::FramedRead;
 use tracing::{Instrument as _, Span};
 
-// PERF/FIXME: changing this parameter to 16 * 1024 greatly improves the throughput,
-//  but we need to wait until 2025 before making this change.
-//
-//  iperf result for 4 * 1024:
-//  > 0.0000-10.0490 sec 23.0 GBytes 19.7 Gbits/sec
-//
-//  iperf result for 16 * 1024:
-//  > 0.0000-10.0393 sec 30.6 GBytes 26.2 Gbits/sec
-//
-//  This is an improvement of ~32.9%.
-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<JmuxApiResponse>;
@@ -790,7 +781,11 @@ 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");