I ran some tests.  I'm new at this, so bear with me.

I set up a new server configuration following the 2.0 HOWTO on openvpn.net.

The server machine is a Windows Vista Ultimate 64-bit on an Intel QX6700 (quad 
core extreme 2.66GHz) and 8GB RAM.  The client is an Ubuntu Hardy 64-bit 
virtual machine running under VMWare Workstation 6.5 (server is the vm host) 
using bridged networking.

On the client, I installed netserver from netperf-2.4.4 and openvpn-2.1rc7 
with network-manager-openvpn.  I configured the client with the certificates 
created when following the HOWTO.

I then performed three test phases, first with the server running 2.1rc13 
(32-bit), then with the server running the 64-bit test build, then again with 
the server running 2.1rc13 (32-bit). In each phase, I ran netperf three times 
from the server, specifying the client VPN address.  Here are the results:

***
*** rc13 32-bit server
***

PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.24
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.20
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.16

***
*** 64-bit test compile (server)
***

PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.48
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.47
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.39

***
*** Again, repeat 32-bit server:
***

PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.09
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.15
PS C:\Users\jaraco> netperf -H 10.8.0.6
TCP STREAM TEST to 10.8.0.6
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

   256   8192   8192    10.00       3.16

--- end tests ---

It appears as if the 64-bit build does have a 5-10% performance increase over 
the 32-bit build in this environment.

I hope these results are helpful.  Unfortunately, I don't have a testbed where 
I can configure two isolated, clean systems, which would probably result in 
more deterministic results.  Let me know if I can arrange the tests 
differently to highlight a particular aspect of the performance.

Regards,
Jason

> -----Original Message-----
> From: Alon Bar-Lev [mailto:alon.bar...@gmail.com]
> Sent: Wednesday, 05 November, 2008 15:53
> To: Jason R. Coombs
> Cc: openvpn devel
> Subject: Re: [Openvpn-devel] [WINDOWS] Request for test win64
>
> Yes.
> I think that there are people here who do this for fun... :)
> The question is if there is someone with amd64 machine that can
> compare previous tests to tests with this version.
>
> Thanks!
>
> On 11/5/08, Jason R. Coombs <jar...@jaraco.com> wrote:
> > How does one check performance?  Run throughput tests through the VPN
> >  connection in a controlled environment?
> >
> >
> >  > -----Original Message-----
> >  > From: Alon Bar-Lev [mailto:alon.bar...@gmail.com]
> >
> > > Sent: Wednesday, 05 November, 2008 15:17
> >  > To: Jason R. Coombs
> >  > Cc: openvpn devel
> >  > Subject: Re: [Openvpn-devel] [WINDOWS] Request for test win64
> >  >
> >
> > > Hello All,
> >  >
> >  > Can anyone check the performance of this [1] win64 openvpn build?
> >  >
> >  > It contains less assembly code, but C optimization may be better
> on
> >  > 64bit.
> >  >
> >  > So you may gain some performance... But it needs to be checked.
> >  >
> >  > Alon
> >  >
> >  > [1] http://alon.barlev.googlepages.com/openvpn-win64.tar.bz2
> >
> >

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to