I'm running an exit relay configured to give up to 6,4MB/s on an Athlon64 3000+ Freebsd box, and I'm seeing cpu usage between 30-60% is this considered normal ? Any means of getting it down? (except getting a faster cpu or hardware crypto)
Yes, but should be across one CPU if you have stuck to a basic config. Tor is a CPU hog.
Sent from my iPhone
On Apr 5, 2011, at 5:56 AM, Gijs Peskens gpeskens85@gmail.com wrote:
I'm running an exit relay configured to give up to 6,4MB/s on an Athlon64 3000+ Freebsd box, and I'm seeing cpu usage between 30-60% is this considered normal ? Any means of getting it down? (except getting a faster cpu or hardware crypto) _______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
Configured for 150KB/300KB bandwidth I see 1.5% CPU utilization on my machine. This is on a 1.8GHz Pentium-M (circa 2004) CPU.
This is running on a RHEL 5.5 system. The version of OpenSSL is so old in this distribution that I doubled my encryption throughput by statically linking Tor with the current OpenSSL release.
I build the current versions of Tor + libevent + OpenSSL as a monolithic application, for profiled execution (GCC v4.4.4). Run the build for several hours, then rebuild using the generated profile data. The CPU bottleneck is clearly the crypto stuff, which is written in x86 ASM and so won't benefit from the profiled build. Still, the peripheral functions are improved slightly by the profiling.
The major improvement, though, was from moving to the current OpenSSL.
On 04/05/2011 08:56 AM, Gijs Peskens wrote:
I'm running an exit relay configured to give up to 6,4MB/s on an Athlon64 3000+ Freebsd box, and I'm seeing cpu usage between 30-60% is this considered normal ? Any means of getting it down? (except getting a faster cpu or hardware crypto)
tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
tor-relays@lists.torproject.org