Hi folks,
I'm currently thinking about a new scalable network architecture for Tor. In order to get some idea about the performance of this architecture I need some data about the CREATE/CREATED handshake. This handshake involves PK encryption which is of course time consuming, and this probably determines the maximum client-capacity of a node in my new network architecture. I'm interested in two values:
1) The average circuit-rate of a single client (circuits/sec) 2) The average CPU-time it takes to perform a CREATE/CREATED handshake.
Any data on these questions is much appreciated.
On Wed, 12 Oct 2016 13:37:24 +0200 Rob van der Hoeven robvanderhoeven@ziggo.nl wrote:
- The average CPU-time it takes to perform a CREATE/CREATED
handshake.
This "depends" entirely on your CPU and which handshake is used, though I don't particularly consider TAP handshake performance relevant because it is slow and superseded by ntor.
`src/test/bench` will give concrete numbers (~140 usec on a modern Intel processor).
Regards,
On 13 Oct 2016, at 00:57, Yawning Angel yawning@schwanenlied.me wrote:
On Wed, 12 Oct 2016 13:37:24 +0200 Rob van der Hoeven robvanderhoeven@ziggo.nl wrote:
- The average CPU-time it takes to perform a CREATE/CREATED
handshake.
This "depends" entirely on your CPU and which handshake is used, though I don't particularly consider TAP handshake performance relevant because it is slow and superseded by ntor.
In 0.2.9 and later, TAP is only used by the receiving party in the hidden service protocol. (That is, client intro and service rendezvous.)
Also, relays that only have TAP onion keys are ignored by authorities, other relays, and clients.
T
`src/test/bench` will give concrete numbers (~140 usec on a modern Intel processor).
Regards,
-- Yawning Angel _______________________________________________ tor-dev mailing list tor-dev@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-dev
T
-- Tim Wilson-Brown (teor)
teor2345 at gmail dot com PGP C855 6CED 5D90 A0C5 29F6 4D43 450C BA7F 968F 094B ricochet:ekmygaiu4rzgsk6n xmpp: teor at torproject dot org ------------------------------------------------------------------------------
On Wed, 2016-10-12 at 13:57 +0000, Yawning Angel wrote:
On Wed, 12 Oct 2016 13:37:24 +0200 Rob van der Hoeven robvanderhoeven@ziggo.nl wrote:
- The average CPU-time it takes to perform a CREATE/CREATED
handshake.
This "depends" entirely on your CPU and which handshake is used, though I don't particularly consider TAP handshake performance relevant because it is slow and superseded by ntor.
`src/test/bench` will give concrete numbers (~140 usec on a modern Intel processor).
This is what I was looking for. Running the benchmark on two very different systems was revealing: on my Pentium G620 the ntor server-side time was ~300 uSec, an Allwinner A20 system completed the server-side code in ~10600 uSec.
Thanks! Rob. https://hoevenstein.nl
On Thu, 13 Oct 2016 15:05:05 +0200 Rob van der Hoeven robvanderhoeven@ziggo.nl wrote:
This is what I was looking for. Running the benchmark on two very different systems was revealing: on my Pentium G620 the ntor server-side time was ~300 uSec, an Allwinner A20 system completed the server-side code in ~10600 uSec.
One of the things on my TODO list is to use NEON for the X25519 scalar mult on ARM targets that are capable of such since it's a decent performance increase, at least on 32 bit ARM.
One day I will also get an Aarch64 target and figure out optimization there, since it's the way of the future.
Regards,
-
Rob van der Hoeven -
teor -
Yawning Angel