You are not logged in.
I want to know that why my non-blocking socket's write will consume such a long time. sometimes 400ms worst.
and this is my strace result:
19:17:48.886276 write(1306, "\n\365\31:h#\t\7\237\20\3\21\337\36\276\317\6\3023\6\204\223,-\306\30\322\"\f\1\30\0\n"..., 121) = 121 <0.403088>
It use 0.403088 seconds to finish the write, And I am sure of that the fd 1306 is a non-blocking socket.
Another records shows:
19:15:27.949736 write(995, "\n\365'\236UN\24\10\26Y\16\30\373\26\306\240\36\267z\10\20\342\f\17\301&\253\240\2\325\\\314\354"..., 160) = 160 <0.368412>
19:15:28.410003 write(1466, "\n\365't\360`\20\10\20\342&\253\240\36\267z\10\26Y\26\306\240\16\30\373\2\325\\\f\17\301\314\354"..., 194) = 194 <0.199244>
19:15:28.627854 write(1821, "\n\365'\260*s*\f\17\301\16\30\373\10\26Y\10\20\342\26\306\240\2\325\\&\253\240\36\267z\314\354"..., 160) = 160 <0.397758>
19:15:29.029411 write(755, "\n\365'f\356\1\1&\253\240\36\267z\10\26Y\26\306\240\16\30\373\f\17\301\2\325\\\10\20\342\314\354"..., 160) = 160 <0.399817>
19:15:29.440261 write(614, "\n\365'b\345H\6\36\267z&\253\240\10\26Y\2\325\\\26\306\240\10\20\342\f\17\301\16\30\373\314\354"..., 305) = 305 <0.396985>
19:15:29.846446 write(892, "\n\365'\354\2256t\16\30\373\f\17\301\2\325\\\10\20\342\10\26Y&\253\240\26\306\240\36\267z\314\354"..., 160) = 160 <0.399170>
19:15:30.250257 write(515, "\n\365'\376\343\214i\10\20\342\26\306\240\f\17\301&\253\240\10\26Y\2\325\\\36\267z\16\30\373\314\354"..., 231) = 231 <0.398990>
19:15:30.652666 write(262, "\n\365'\274so\6&\253\240\10\26Y\2\325\\\26\306\240\36\267z\16\30\373\f\17\301\10\20\342\314\354"..., 168) = 168 <0.396569>
19:17:46.885289 write(908, "t\213\16\222\0\0\0\0\240@\33L\0\0\3219\0\0\333J\0\0\2218\0\0\2\0|\203\16\222\0"..., 189) = 189 <0.396330>
19:17:47.351419 write(1529, "\n\365\31\352\2\10\252\36\276\317\30\322\"\6\204\223\6\3023\f\1\30\7\237\20\3\21\337,-\306\0\n"..., 128) = 128 <0.293829>
19:17:47.651080 write(683, "\n\365\31~ZG=\30\322\"\6\204\223\7\237\20\36\276\317\6\3023\3\21\337,-\306\f\1\30\0\n"..., 121) = 121 <0.398178>
19:17:48.062419 write(175, "\n\365\31\224H\260\313\f\1\30\6\3023\3\21\337\36\276\317\6\204\223\7\237\20,-\306\30\322\"\0\n"..., 121) = 121 <0.398811>
19:17:48.467481 write(569, "\n\365\31v5]H\3\21\337\30\322\"\f\1\30\6\3023\6\204\223\7\237\20,-\306\36\276\317\0\n"..., 121) = 121 <0.192335>
19:17:48.663023 write(173, "\n\365\31\376\nJ\0\36\276\317,-\306\7\237\20\30\322\"\3\21\337\6\204\223\6\3023\f\1\30\0\n"..., 121) = 121 <0.206703>
19:17:48.886276 write(1306, "\n\365\31:h#\t\7\237\20\3\21\337\36\276\317\6\3023\6\204\223,-\306\30\322\"\f\1\30\0\n"..., 121) = 121 <0.403088>
BTW, This application is for the service of 3-4K client.
And I do want to ask that:
1.Why non-blocking write will make our process lose the CPU ( I think it is blocked ).
2.What our process waiting for? Why not just return EAGAIN?
3.Can we avoid this problem?
Our server's info:
Linux hz172-96 2.6.32-bpo.5-amd64 #1 SMP Mon May 2 11:40:03 UTC 2011 x86_64 GNU/Linux
Looking forward to your answer, Thank you!
Offline
Does the device (kernel module) you are writing to support non-blocking I/O?
Offline
Maybe you're getting EWOULDBLOCK when the write would block instead of just EAGAIN. Check the return value of the write instead of assuming the write just worked.
Offline
thank you for your answer.
and i have resolve this ploblem in using the linux kernel 2.6.30, using kernel 2.6.32 before.
but, i don't kown Is there some bug in linux kernel 2.6.32 or for other reason.
Offline