Firewall Wizards mailing list archives

RE: Re: Free S/wan over satellite


From: David Klein <dklein () netscreen com>
Date: Thu, 30 May 2002 08:57:22 -0700

From: Joseph S D Yao [mailto:jsdy () center osis gov]
...
Not the window size problem I had thought, but this might explain it.

But, why can't they play the same games with IPsec ESP 
packets?  Aren't
packets just packets?


All the TCP window size negotiation and TCP seq/ack numbers are encrypted
within the ESP header.  The point of encrypting and secure-hashing a packet
is so someone can't see the packet nor change the packet in transit.  This
would include your satellite provider.

IPSec ESP doesn't maintain a sliding window nor "replicate" the TCP seq/ack
numbers.  [Well, there are SPI numbers and there is a sequence number in the
ESP header but these are used for tunnel/SA identification and anti-replay.]
In an IP network, "maintaining the state of a connection" has always fallen
on the end-systems to do (via TCP).  Routers (dealing only with IP headers)
and VPN/IPSec gateways (dealing with IP and ESP headers) generally don't
care about this client-to-server "connection state".

Firewalls, NAT boxes, and Satellite gear tend to want to mess with IP and
TCP headers for security, enforcing state, hiding things, and in the case of
Satellites, tweaking/spoofing TCP headers to transparently overcome the
[standard TCP window size + long roundtrip delay] combo that negatively
effects throughput.  Just like a firewall can't see the TCP header nor
screen sessions/connections within an IPSec tunnel going through it, neither
can the Satellite terminals see nor muck with the TCP headers once they have
been tunneled (i.e., encapsulated and encrypted into an ESP header).

Dave Klein
Netscreen SE


-----Original Message-----
From: Joseph S D Yao [mailto:jsdy () center osis gov]
Sent: Thursday, May 30, 2002 10:23 AM
To: David Klein
Cc: Ben Swanner; firewall-wizards () nfr com
Subject: Re: [fw-wiz] Re: Free S/wan over satellite


On Thu, May 30, 2002 at 07:38:54AM -0700, David Klein wrote:
On Fri, May 24, 2002 at 12:26:11PM -0500, Ben Swanner wrote:
Set up on Linux over vsat connection and speed dropped by a 
factor of ten.
Any ideas?

IP-over-Satellite service providers do a couple of things 
to overcome the
throughput issue related to the normal TCP window size 
coupled with long
latency delays related to satellites.  Specifically, they 
1) dip into the TCP header and increase the window size; and/or
2) locally spoof TCP acks at the satellite terminal where the TCP
transmitter is.

Item (1) will cause the TCP transmitter to send more 
packets into the
network before stopping and waiting for acknowledgements.

Item (2) will cause the TCP transmitter to think the 
intended TCP receiver
has already received the packet it sent and so the TCP 
transmitter will
continue sliding its TCP window along and transmitting more packets.

Once you put TCP into IPSec ESP (or any encrypted packet), 
they can't play
these games.  So now your real TCP window size comes into 
play.  And with
the massive round-trip time associated with satellite 
links, this means your
TCP transmitter blasts a few packets out (a normal TCP 
window size worth)
and then effectively waits a while for the acks to come 
back from the real
TCP receiver before sending more traffic.  

Dave Klein
Netscreen SE

Not the window size problem I had thought, but this might explain it.

But, why can't they play the same games with IPsec ESP 
packets?  Aren't
packets just packets?

-- 
Joe Yao                               jsdy () center osis gov - 
Joseph S. D. Yao
OSIS Center Systems Support                                   EMT-B
--------------------------------------------------------------
---------
   This message is not an official statement of OSIS Center policies.

_______________________________________________
firewall-wizards mailing list
firewall-wizards () nfr com
http://list.nfr.com/mailman/listinfo/firewall-wizards


Current thread: