Geek Answers Handbook

Linux socket performance

Following my last question:

Performance issue using javas socket streams with sockets

I am looking at socket performance on Linux. In the above example, I get a round-trip time of ~ 65 μs. If I make two commits in the file system, it will decrease to ~ 45 μs. The extra time using localhost sockets should be due to the fact that I'm on the network stack.

Is there some kind of OS configuration that can make the localhost socket go as fast as the named pipe?

uname -a
Linux fiatpap1d 2.4.21-63.ELhugemem #1 SMP Wed Oct 28 23:12:58 EDT 2009 i686 athlon i386 GNU/Linux

Thanks in advance!

+5
source share
3 answers

~ 65 . , ~ 45 . , localhost, , .

, .

FIFO - . bool. . , .

. TCP. (recv, send ). , - , - . Linux : / . , , , , .

, , FIFO, 20us - , Linux.

P.S. 65us rtt = ~ 35us . 1s/35us = ~ 30K . , , .

+2

Java, UNIX. , Java:

UNIX Java?

+1

Your previous questions contain two false assumptions:

  • ICMP_ECHO (aka ping) gives meaningful time information. This does not mean that, among other things, the ICMP level can be (and should be) with a low priority of service.
  • The fact that data marching through intermediate Java interfaces is not a bottleneck. Because it.

Your testing methods are very suspicious. What are you trying to accomplish?

+1
source

More articles:


All Articles