NOTES

 - weird RR behaviors
 	- "A" behavior -- 	Cisco
		- MPLS enabled routers don't add RR for outgoing MPLS connections
	- "B" behavior --	"Juniper" ?
			- add outgoing interface even if packet is bounced
	- "stupid" behavior : don't bounce RR packets, bounce non-rr packets
		mark RR packets on outgoing packets
		examples: 216.24.186.{4,6,33} from uw to cornell
		OrgName:    National LambdaRail, IndianaU ?? http://www.nlr.net/
		explanation: firewall blocking icmp bounce > 48 bytes?
			- use id increment rate to verify
	- "flakey" behavior: sometime mark RR packets, sometimes not
		Cisco under load?
	- "DropsRR" 
		- won't let packets with RR pass by

- apply the "off by one" network alias connection rule


 - some planetlab systems calls take >1 sec to complete

 - some weird middle boxes uncheck DF bit, screws up optimization

#################################################################
Problematic Trace

ttl 	nRR entries	RRtype	Inference source
-----------------------------------------------
1	0		A	2 is not type B (#2:beta)
2	1		A	3 has +2 entries  (#1:alpha)
3	3		B	+2  --> always B (#0:alpha)
4	3		A	5 is +2: (#4:beta)
5 	5		B	+2  --> always B (#3:alpha)
6	5		N	+0,+0 --> (#5:gamma)
7	5		A/N
8	6		N/B
9	6		A	in/new hop off by 1: (#7:epsilon)
10 	7		A	in/new hop off by 1: (#6:epsilon)
11	8		A	10 is A (#8,beta)






Rules:
	Rule alpha: 	+2 from prev --> you are B and prev is A
	Rule beta : 	+1 from prev --> you are B and prev is N or B
				-OR- you are A or N and prev is A 
	Rule gamma: 	+0 from prev --> you are A or N and prev is N or B
	Rule delta: 	if you are +0 and next is +0, you are N
	Rule epsilon: 	if +1 and inhop ip and new RR hop ip are off by 1, then you are A and prev is A
	Rule zeta: 	if out of band ping shows RR, then not N
	Rule eta: 	are inhop and new RR hop alises?  yes, then B, no then A or N
	GUESS: 
		g = nRR entries + #N's upstream: 
		- if g == ttl, then B
		- if g == ttl-1

Further note:
	a sequence of 0,1,2,3,4,5,6,7,8,9 nRR's is indeterminant:
	if could be all A's  or  N,B,B,B...
-----------------------------------------------------------------------------------------------
Calc queue size as a function of min delay: 
	say average packet is 80 bytes, then if we are going at b Bytes/s,
	then max queue size should be b/80 or 500kbps --> 64000/80 == 800 packets
--------------------------------------------------------------------------------------------
MPLS stuff:

Cisco explanation:  `no mpls ip propagate-ttl` 
	http://www.cisco.com/en/US/tech/tk436/tk428/technologies_tech_note09186a008020a42a.shtml
Good MPLS tutorial:
	http://www.iec.org/online/tutorials/mpls/topic03.html
Can also get MPLS info from ping: 
"MPLS OAM Tools for Troubleshooting MPLS Networks" http://www.cisco.com/application/pdf/en/us/guest/netsol/ns172/c654/cdccont_0900aecd80272b7f.pdf