[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: network transfer performance
From: |
Michael Cronenworth |
Subject: |
Re: network transfer performance |
Date: |
Fri, 22 Jun 2012 10:15:10 -0500 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:13.0) Gecko/20120605 Thunderbird/13.0 |
Michael Cronenworth wrote:
> Fedora uses libgcrypt. 128-bit vs 256-bit didn't provide any tangible
> difference in my testing. I'll give all those ciphers a shot and look
> into using libnettle. Thanks!
Using libnettle with ARCFOUR-128 and MD5 nets me about a 2 second
improvement. (8 seconds down from 10 seconds for 100MB)
Perf shows all the processing is happening in libnettle on both the
client and server side so none of my code should be holding up the transfer.
client (top two lines):
25.76% foo libnettle.so.4.3 [.] nettle_arcfour_crypt
18.61% foo libnettle.so.4.3 [.] _nettle_md5_compress
server (top two lines):
16.17% bar libnettle.so.4.3 [.] nettle_arcfour_crypt
13.19% bar libnettle.so.4.3 [.] _nettle_md5_compress
When I run "top" during the transfer the client and server process take
~30% of a CPU core each, so it still seems like something is
bottlenecking the processes, but I have exhausted my knowledge on
profiling. Anything else I can look at?