While the XCP code release has proven stable and interesting in the
lab, there are certainly some untested code paths and undetected bugs
still lurking around.  If you download it, please be advised that it
is beta quality.  If you find a bug, please tell us how to replicate it
so we can fix it.  Even better would be a patch and a description of
the problem.

The current version has addtional fields in the xcphdr like queue_length,
te_timer value, positive_feedback and negative_feedback. These parameters
have been added for debugging purposes only and would be removed as we
progress with our tests. 

Differences between the previous XCP release and current.
                                                                             
1. The previous release dealt in terms of congestion
window i.e. the sender put the cwnd in the packet
header and the router calculates the throughput using
cwnd/rtt.
In the current release the cwnd/rtt calculation is moved
to the end host to avoid the division in the router. The
header contains throughput (Kbytes/s)
                                                                             
2. The current release configures the "capacity" variable
used by XCP router per interface. It has been integrated
with ifconfig.
In the previous release this was hard coded.

3. The current release is compliant to xcp-spec-XX 
(See webpage)
                                                                             

Additional Information:

Inducing delays: 
	To test XCP in an environment with delays we use
DUMMYNET option in the Kernel config. A sample list of commands
is provided below. Here we are delaying all packets to host
10.10.0.2 by 100ms.
$) ipfw add 2000 pipe 1 ip from any to 10.10.0.2 out
$) ipfw pipe 1 config delay 100ms

Setting up the xcp capacity:
  XCP determines the unused bandwith by subtracting the current usage 
by the CAPACITY. This variable is tied to the interface. To set it:
$) ifconfig em1 xcp 12207 
where capacity = 12207 KB/s (100Mb/s)

  To view the xcp statistics 
$) ifconfig em1 xcpstats

XCP SYSCTLS:

1) net.inet.xcp.do_xcp_feedback: Whether xcp feedback calculations are
to be done or not. If using XCP the value should be 1.

2) net.inet.xcp.aggressive: A sysctl to control how aggressive XCP 
requests are if true, each packet contains a request to fill the 
window, if false that number is divided by the number of full sized 
packets in the window.

3) net.inet.xcp.xcp_metered_output: To avoid sending bursts of
packets. (Not tested) 

4) Trace sysctls:
net.inet.xcp.trace_fwd_toggle
net.inet.xcp.trace_teo_toggle
net.inet.xcp.trace_tqo_toggle
net.inet.xcp.trace_rcv_toggle
net.inet.xcp.trace_snd_toggle

TCP SYSCTLS:

Currently to avoid drops we make the buffers large (b/w * delay)
The sysctls to modified are

At the end points:
sysctl -w net.inet.tcp.sendspace=<value>
sysctl -w net.inet.tcp.recvspace=<value>
sysctl -w kern.ipc.maxsockbuf=<value>

At the router:
sysctl -w kern.ipc.maxsockbuf=<value>