[Openvpn-users] Retake on openvpn speed over gigabit

[Ski M. <ski...@ya...>]

On 5/2/11 11:20 AM, ope...@li... wrote:
> Message: 6
> Date: Mon, 02 May 2011 10:20:19 -0500
> From: Les Mikesell<les...@gm...>
> Subject: Re: [Openvpn-users] Retake on openvpn speed over gigabit
>     ethernet (SECOND attempt)
> To:ope...@li...
> Message-ID:<4DB...@gm...>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> On 5/2/2011 8:59 AM, Jan Just Keijser wrote:
>> >
>> >  as promised, my write-up on OpenVPN speeds on gigabit LANs:
>> >      https://community.openvpn.net/openvpn/wiki/Gigabit_Networks_Linux
>> >
>> >  in short
>> >  - use openssl 1.0 + the aes-ni patch, even on older CPUs
>> >  - use a FAST CPU : clockspeed matters
>> >  - I managed to get 585 Mbps using aes-256-cbc on a non-aesni capable cpu
>> >  - I managed to get 885 Mbps using aes-256-cbc on a aesni capable cpu
>> >  - openvpn speed is limited by encryption/decryption; the 
>>kernel-space/user-space division comes into play a bit later (@1.3 Gbps)
>> >
> Are those large MTU sizes shown in the tests actually practical or are
> they just shown to demonstrate how they interact with the encryption
> settings with the test data originating on the VPN machine?  That is,
> can you always count on the underlying OS stack to fragment down to size
> the output media can handle or to figure out what to do with packets
> with the DF bit set?  And, is this likely to have any real effect when
> routing for a LAN where the initial packet size will be limited by the
> first-hop ethernet?
> 
> -- Les Mikesell les...@gm...

Is it possible to make OpenVPN automatically calculate the best MUT for each 
network connection that is made?  Since computers are much better at caculating 
things like best MTU size then I as a human could be constantly calculating and 
testing different MTU sizes.  Possibly having OpenVPN sending a couple different 
packets of different MTU sizes to determine the best size.