在我的 C 程式中，我發送了許多小的 (< KB) UDP 訊息。當我測量此類 UDP 發送的性能時，我發現它們中的大多數在 300 毫秒內完成，但偶爾（每 ~ 150 次）需要 3 秒。請注意，這些測量是通過 wi-fi 網路進行的，這可能解釋了偶爾的延遲。但是，如果我想支持 wi-fi 場景，我懷疑增加發送緩沖區將是我的第一步。我最初的研究指出我通過net.core.rmem_maxand增加了 UDP 緩沖區大小net.core.rmem_default，我這樣做了：
$sudo sysctl net.core.rmem_max
net.core.rmem_max = 26214400
$sudo sysctl net.core.rmem_default
net.core.rmem_default = 26214400
但是，這似乎對我的程式沒有任何影響。接下來，我嘗試在 code: 中設定緩沖區大小
setsockopt(sock_fd, SOL_SOCKET, SO_SNDBUF, &size, size_len)，其中大小為26214400. 這解決了問題 - 不再有延遲的資料包。但是，通過 檢查緩沖區大小后getsockopt(socket_fd, SOL_SOCKET, SO_SNDBUF, &size, &size_len)，我看到緩沖區設定為5228800.
因此，我的問題是：增加 UDP 發送緩沖區的正確方法是什么？
我的環境：
cat /etc/os-release
PRETTY_NAME="Ubuntu 22.04.1 LTS"
NAME="Ubuntu"
VERSION_ID="22.04"
VERSION="22.04.1 LTS (Jammy Jellyfish)"
此問題僅在我使用 wi-fi 時出現，這是我要支持的用例。換句話說，在針對以太網介面進行測驗時，我看不到偶爾的緩慢發送。
這是我與UDP相關的源代碼：
#include <netinet/in.h>
#include <arpa/inet.h>
#include <stdexcept>
#include <cstring>
#include <cstdlib>
#include <cerrno>
#include <boost/log/trivial.hpp>
#include "udp_sender.h"
void soc_get_so_snd_buf(int socket_fd)
{
unsigned int size = 0;
socklen_t size_len = sizeof(size);
if(getsockopt(socket_fd, SOL_SOCKET, SO_SNDBUF, &size, &size_len) == -1 ) {
std::string err_msg = "Failed to get the socket send buffer for socket: " std::to_string(socket_fd);
BOOST_LOG_TRIVIAL(error) << err_msg;
throw std::runtime_error(err_msg);
}
BOOST_LOG_TRIVIAL(debug) << "Send buffer size (SO_SNDBUF) is: " << size;
}
void soc_set_so_snd_buf(int sock_fd, size_t size)
{
socklen_t size_len = sizeof(size);
if (setsockopt(sock_fd, SOL_SOCKET, SO_SNDBUF, &size, size_len) == -1) {
std::string err_msg = "Failed to set the socket send buffer to " std::to_string(size);
BOOST_LOG_TRIVIAL(error) << err_msg;
throw std::runtime_error(err_msg);
}
}
UDPSender::UDPSender(int id) : my_id(id) {
char my_ip[100];
sprintf(my_ip, "10.1.1.%d", id);
send_socket_fd = socket(AF_INET, SOCK_DGRAM, 0);
if(send_socket_fd < 0)
throw std::runtime_error("Failed to create a send socket.");
int enable = 1;
if (setsockopt(send_socket_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(int)) < 0) {
throw std::runtime_error("Failed to assign REUSADDR option.");
}
soc_get_so_snd_buf(send_socket_fd);
soc_set_so_snd_buf(send_socket_fd, 26214400); // setting this to 26214400 results in 5228800 which seems to solve my issue for now.
soc_get_so_snd_buf(send_socket_fd);
struct sockaddr_in srcAddr{};
memset(&srcAddr, 0, sizeof(srcAddr));
srcAddr.sin_family = AF_INET;
srcAddr.sin_port = htons(0); // Any port will do.
inet_pton(AF_INET, my_ip, &srcAddr.sin_addr);
if(bind(send_socket_fd, (struct sockaddr*)&srcAddr, sizeof(srcAddr)) < 0) {
BOOST_LOG_TRIVIAL(error) << "Failed to bind sending socket " << strerror(errno);
throw std::runtime_error("Failed to bind a send socket.");
}
}
UDPSender::~UDPSender() {
close(send_socket_fd);
send_socket_fd = -1;
}
ssize_t UDPSender::send(
int dest_id,
int dest_port,
char* msg,
size_t msg_size) const {
char dest_ip[100];
struct sockaddr_in dest_addr{};
sprintf(dest_ip, "10.1.1.%d", dest_id);
memset(&dest_addr, 0, sizeof(dest_addr));
dest_addr.sin_family = AF_INET;
dest_addr.sin_port = htons(dest_port);
inet_pton(AF_INET, dest_ip, &dest_addr.sin_addr);
ssize_t bytes_sent = sendto(
send_socket_fd,
msg,
msg_size,
0,
(struct sockaddr*)&dest_addr, sizeof(dest_addr));
if (bytes_sent < 0) {
std::string error_msg = "Failed to send from: "
std::to_string(my_id)
" to: "
std::to_string(dest_id) " errno: " std::to_string(errno) " ";
error_msg = strerror(errno);
if (errno == EPERM) {
error_msg = "; error was due iptables bock rule - expected for some targets.";
BOOST_LOG_TRIVIAL(debug) << error_msg;
return 0;
} else {
BOOST_LOG_TRIVIAL(error) << error_msg;
throw std::runtime_error(error_msg);
}
}
if (bytes_sent != msg_size) {
BOOST_LOG_TRIVIAL(error) << "Failed to send the expected number of bytes, wanted :"
<< msg_size << " sent " << bytes_sent;
throw std::runtime_error("Failed to send the expected number of bytes.");
}
return bytes_sent;
}
測驗：
#include "udp_sender.h"
#include "message_factory.h"
#include "heartbeats.h"
#include <thread>
#include <chrono>
#include "gtest/gtest.h"
#include <boost/log/trivial.hpp>
using namespace std::chrono;
TEST(UDPSenderTest, Broadcast) {
int sender_id = 0;
UDPSender sender = UDPSender(sender_id);
char serialized[65000];
Heartbeat msg = Heartbeat(sender_id, Peers{Node{2, 3}}, false);
long serialized_size = MessageFactory::serialize_heartbeat(msg, serialized);
auto start_cycle = high_resolution_clock::now();
for(int i=0; i<256; i ) {
auto start = high_resolution_clock::now();
sender.send(i, 7777, serialized, serialized_size);
auto stop = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(stop - start);
if (duration.count() > 300) {
BOOST_LOG_TRIVIAL(error) << "Sent broadcast to " << i << " in " << duration.count() << " milliseconds";
}
}
auto stop_cycle = high_resolution_clock::now();
auto duration = duration_cast<milliseconds>(stop_cycle - start_cycle);
BOOST_LOG_TRIVIAL(debug) << "Completed broadcast cycle in " << duration.count() << " milliseconds";
}
當我沒有設定套接字選項而是依賴 Linux 配置時輸出：
[ RUN ] UDPSenderTest.Broadcast
37: [2022-10-06 09:56:08.183035] [0x00007f5f15f8b740] [debug] Send buffer size (SO_SNDBUF) is: 212992
37: [2022-10-06 09:56:11.256173] [0x00007f5f15f8b740] [debug] Sent broadcast to 247 in milliseconds 3069
37: [2022-10-06 09:56:11.256590] [0x00007f5f15f8b740] [debug] Completed broadcast cycle in 3073 milliseconds
uj5u.com熱心網友回復：
您設定了錯誤的內核引數。
net.core.rmem_default和net.core.rmem_max引數分別設定 UDP接收緩沖區的默認大小和最大大小。UDP發送緩沖區的相應引數是net.core.wmem_default和net.core.wmem_max。
一般來說，為您的程式設定 UDP 發送緩沖區的正確方法是使用SO_SNDBUFsocket 選項，您設定的值將受net.core.wmem_max內核引數的限制。然后，如果您需要一個大于最大值的值，您可以增加 的值net.core.wmem_max。
您通常不想修改，net.core.wmem_default否則net.core.rmem_default會影響系統上的其他行程。
轉載請註明出處，本文鏈接：https://www.uj5u.com/qiye/512327.html
標籤：C C插座网络编程UDP
下一篇：未在NestJS微服務中接收資料