A BGP issue?

Tue Mar 8 15:47:18 UTC 2011

On Mar 8, 2011, at 8:52 AM, Greg Ihnen wrote:
> On Mar 7, 2011, at 10:19 PM, Patrick W. Gilmore wrote:
>> On Mar 7, 2011, at 14:27, Greg Ihnen <os10rules at gmail.com> wrote:
>> 
>>> I run a small network on a mission base in the Amazon jungle which is fed by a satellite internet connection. We had an outage from Feb 25th to the 28th where we had no connectivity with email, http/s, ftp, Skype would indicate it's connected but even chatting failed, basically everything stopped working except for ICMP. I could ping everywhere just fine. I started doing traceroutes and they all were very odd, all not reaching their destination and some hopping all over creation before dying. But if I did traceroute with ICMP it worked fine. Does this indicate our upstream (Bantel.net) had a BGP issue? Bantel blamed Hughesnet which is the service they resell. I'm wondering what kind of problem would let ping work fine but not any of the other protocols. It also seems odd that I could traceroute via UDP part way to a destination but then it would fail if the problem was my own provider. Thanks.
>>> 
>>> If this is the wrong forum for this post I'm sorry and please just hit delete. If this is the wrong forum but you'd be kind enough to share your expertise please reply off-list. Thanks!
>> 
>> Honestly, I would rate this as one of the most on-topic posts in a while.
>> 
>> BGP only handles reachability, not higher level protocols.  (Of course, you can h4x0r anything to do jus about anything, but we are talking the general case here.)
>> 
>> If you can ping, BGP is working.  If you can ping and cannot use TCP, then something other than BGP is at fault. 
>> 
>> I've seen strange things like someone enabling TCP compression (common on very small or very expensive links) one side but not the other, which then allowed ICMP and UDP but not TCP.  It is a great way to annoy someone.  "See, I can ping, it must be your side!"
>> 
>> Have you tried TCP traceroute?  Or telnetting to port 80?

> 	I did try TCP traceroute and it failed too. I didn't have a machine to telnet to on port 80 but I did try an ssh tunnel on port 9999 and it failed too.

Sure you do.  Any web server will allow you to telnet to port 80.

	TiggerBook-Air3:~ patrick$ telnet www.yahoo.com 80
	Trying 67.195.160.76...
	Connected to any-fp.wa1.b.yahoo.com.
	Escape character is '^]'.
	GET GET
	<HEAD><TITLE>Not Found</TITLE></HEAD>
	<BODY BGCOLOR="white" FGCOLOR="black">
	<FONT FACE="Helvetica,Arial"><B>
	 Your requested URL was not found.</B></FONT>

	<!-- default "Not Found" response (404) -->
	</BODY>
	Connection closed by foreign host.

[In case it wasn't clear, I typed "GET GET" myself, just to have the web server respond with something.]

> 	From what everyone is saying it sounds like it was the satellite internet provider's compression scheme that was having trouble or some kind of an MTU issue.
> 
> 	What I don't understand is why when using traceroute UDP/TCP/GRE I could get replies from some routers but not all routers to the destination, and why some routes were bizarre. If it was a failure of the sat internet provider's compression scheme or an MTU issue wouldn't traceroute UDP/TCP/GRE fail completely? What could have happened to my packets that would make them go only part way or go the wrong way?

It was likely not MTU if you can traceroute to some places, but not others.  Traceroute doesn't send or receive big packets.

And I didn't really see anything terribly unusual in the traces you sent, other than some not completing.  If you are talking about the Cogent one, with many routers per hop, that's just standard load balancing.

-- 
TTFN,
patrick