[pptp-server] Performance difference between TCPIP and SMB tr affic

[Cowles, Steve]

> -----Original Message-----
> From: Jamin Collins [mailto:JaminC at adapt-tele.com]
> Sent: Tuesday, April 03, 2001 9:22 AM
> To: 'robert'; Tife Chan; pptp-server at lists.schulte.org
> Subject: RE: [pptp-server] Performance difference between
> TCPIP and SMB tr affic
>
>
> I may be reading too much into this, but I'm getting the
> impression that both of the tests are being perfromed through
> the PPTP connection.  Thus, the encryption/decryption is happening
> for both the FTP and the SMB transfers.  Thus his concern over why
> FTP is 10 times faster than SMB.
>
> Jamin W. Collins

I agree, there is something else wrong if he is getting a 10x differnece
between FTP and SMB packets. Personally, I have also noticed a differnece in
speed when using FTP verus SMB, but not on the order of 10x. I have never
measured the differnece, but I would say I see a 15-20% decrease when using
SMB. 

FWIW: SMB packets are also an encapsulated (netbios) which is then
encapsulated again (and possibly encrypted) into a GRE packet before being
sent across the tunnel. In contrast, FTP uses raw TCP/IP as its transport,
but its packets are still encapsulated/encrpyted into a GRE packet before
being sent across the tunnel.

If I was to guess at where the problem lies, it would be in one of the
following areas:
1) How Netbios is configured on Windows or Samba (like the "socket options"
in smb.conf)
2) The TCP Window size (TCPWIN)
3) The MTU/MRU parameters of the PPTP tunnel. 

Without using a packet sniffer its hard to tell, but its possible that the
SMB packets (after being encapsulated twice) are being fragmented at the
PPTP server/router. Thus causing additional overhead because the GRE packets
are now lager than the allowed MRU/MTU (bytes) and have to be broken up into
two packets to be properly transmitted across the tunnel. 

That my two bits. If I get some time, I will try to create a packet capture
on the difference between FTP/SMB transfers across a PPTP tunnel. If I
remember right, packets that have been fragmented are flagged as such and
easliy spotted.

Steve Cowles 

>
> > -----Original Message-----
> > From: robert [mailto:berzerke at swbell.net]
> > Sent: Friday, March 30, 2001 9:07 AM
> > To: Tife Chan; pptp-server at lists.schulte.org
> > Subject: Re: [pptp-server] Performance difference between
> > TCPIP and SMB traffic
> >
> >
> > Understand that *some* slowness is normal.  Think what has to
> > happen to those little packets.  (This order may be wrong..)
> > First they are encrypted, then encapsulated, then the routing
> > is changed, then they are sent over the wire, were the process
> > is reversed.  These changes take time, although not much for
> > each packet.  None of these steps occur with ftp, so there is
> > less overhead.  Now how much slowness is normal I don't know.
> > I'll have to do some tests myself and post the results here
> > later.
> >
> >
> > > On Thursday 29 March 2001 21:43, Tife Chan wrote:
> > > Hi all,
> > >
> > > I found that passing SMB traffic through the pptp link is
> > > much slower than TCPIP. I have network A and network B and
> > > they are connected together with two linux servers using pptp.
> > > When ftp a file from network A to network B, the speed is fine.
> > > But when I try to copy the same file from network A to network
> > > B through Windows Explorer, the speed is much much slower.
> > >
> > > Any idea?
> > >
> > > Thanks.
> > >
> > > Regards,
> > > Tife Chan