# re: muduo 與 boost asio 吞吐量對比 2010-09-04 20:23 Cox

高人，期待后續(xù)大作  回復(fù)  更多評論   

# re: muduo 與 boost asio 吞吐量對比 2010-09-05 20:08 imjj

樓主的庫cmake 沒裝上，還沒測試
ACE 只實現(xiàn)了一個server，client 用的 asio 的
./client 127.0.0.1 $p 1 16384 $n 10
64位 win7 下 virtualbox 中的 ubuntu 10.04 32bit 桌面版，2 CPU 打開了 VT
ACE 是 5.8.1，reactor 模型， ACE_Dev_Poll_Reactor

5555 是 ACE 的服務(wù)器 6666 是 aiso 1.45 服務(wù)器

server:5555 sessions:1
941965312 total bytes written
941948928 total bytes read
server:6666 sessions:1
1029111808 total bytes written
1029095424 total bytes read

server:5555 sessions:10
2783215616 total bytes written
2783100928 total bytes read
server:6666 sessions:10
2227601408 total bytes written
2227503104 total bytes read

server:5555 sessions:100
1310179328 total bytes written
1308786688 total bytes read
server:6666 sessions:100
981581824 total bytes written
979976192 total bytes read

server:5555 sessions:1000
430227456 total bytes written
424837120 total bytes read
server:6666 sessions:1000
148946944 total bytes written
143015936 total bytes read

ACE 源碼如下
#include <ace/ACE.h>
#include <ace/Acceptor.h>
#include <ace/Task.h>
#include <ace/Svc_Handler.h>
#include <ace/SOCK_Stream.h>
#include <ace/SOCK_Acceptor.h>
#include <ace/INET_Addr.h>
#include <ace/Reactor.h>
#include <ace/Dev_Poll_Reactor.h>


int BLOCKSIZE = 16384;


class PingHandler : public ACE_Svc_Handler<ACE_SOCK_STREAM, ACE_NULL_SYNCH>
{
public:
PingHandler()
{
m_rbuff = NULL;
ACE_NEW(m_rbuff, ACE_Message_Block(BLOCKSIZE));
}
virtual ~PingHandler()
{
ACE_Message_Block::release(m_rbuff);
}

virtual int handle_input(ACE_HANDLE /*fd*/)
{
int ret = 0;
ssize_t bytes_required = m_rbuff->space();
ssize_t bytes_read = 0;
bytes_read = peer().recv(m_rbuff->wr_ptr(), m_rbuff->space());
if (bytes_read <= 0)
return -1;
m_rbuff->wr_ptr(bytes_read);
if (m_rbuff->space() == 0)
{
peer().send(m_rbuff->rd_ptr(), m_rbuff->length());
m_rbuff->reset();
}
return 0;
}

protected:
ACE_Message_Block * m_rbuff;
};


typedef ACE_Acceptor<PingHandler, ACE_SOCK_ACCEPTOR> PingAcceptor;

void usage()
{
ACE_DEBUG((LM_DEBUG,
ACE_TEXT("server <ip> <port> <threads> <blocksize>\n")));
}

int ACE_TMAIN(int argc, ACE_TCHAR * argv[])
{
if (argc != 5)
{
usage();
return 0;
}

ACE_Dev_Poll_Reactor dpr(1024);
ACE_Reactor r(&dpr);

const ACE_TCHAR * ip = argv[1];
u_short port = (u_short)ACE_OS::atoi(argv[2]);
int threads = ACE_OS::atoi(argv[3]);
BLOCKSIZE = ACE_OS::atoi(argv[4]);

PingAcceptor acceptor;
ACE_INET_Addr addr(port, ip);
if (acceptor.open(addr, &r) == -1)
{
ACE_DEBUG((LM_DEBUG, ACE_TEXT("faield:%p\n")));
return -1;
}

r.run_reactor_event_loop();
return 0;
}  回復(fù)  更多評論   

# re: muduo 與 boost asio 吞吐量對比 2010-09-06 11:59 普派

謝謝，收藏了  回復(fù)  更多評論   

# re: muduo 與 boost asio 吞吐量對比 2010-09-10 20:01 陳碩

@imjj
希望看到 client 端的 ACE 實現(xiàn)。  回復(fù)  更多評論   

# re: muduo 與 boost asio 吞吐量對比 2013-04-27 13:55 yayj

我修改了下asio的測試代碼，讓每個worker線程使用不同的io_service，但測試結(jié)果與只有一個io_service相比，沒有明顯的改善。所以可能不是io_service的數(shù)量問題。代碼鏈接https://gist.github.com/yayj/5472033, 有使用c++11的unique_ptr  回復(fù)  更多評論