Nsis_Imp July 2006

nsis_imp@informatik.uni-goettingen.de

11 participants
15 discussions

RE: [Nsis_imp] qos-nslp local repair bug - is my GIST set upincorrectly?
by Christian Dickmann 26 Jul '06

26 Jul '06

Hi, > I have a question again. > My test setup is composed of 3 nodes: > > B > / | > / | > / | > A | > \ | > x | > \ | > C > > I wished to test the local-repair feature of QoS-NSLP, > which, as I understand, roughly does the following: > > 1. initially A sends to C (cmode, downstream) > with direct route A-C > 2. signaling is OK, I got response to the reserve > from C (direct C-A route), etc., all is fine > 3. I pull the C-A wire, faking a failure Just to understand you correctly: You have GIST running on B, but initially B is not on the path, right? > 4. since my code at A does an "ip route show" every > 500 ms, it detects the route change correctly > (I use Quagga/ospfd) So the complete network has a valid route state now? Is a ping from A to C and back to A really working? > 5. local-repair at A sends a refresh reserve > immediately to C, but on the new route (A-B-C) So your own QoS-NSLP (not ours, right?) sends a new QoS message (RESERVE), right? As GIST does not know about the route change, it still assumes that C is the correct peer. > **** > 6. though C receives this new refresh reserve > somehow, and I see some Dmode messages at B, > my qos-nslp code running on B does not get > any recvmessage calls... > (also C sends a reply to A, but it does not > arrive neither to B, nor to A...) > **** That is strange. You see (using ethereal, oder how?) that C sends the response to A, but the message does not arrive at A? Doesn't it arrive at the NSLP, or not at the machine at all? I would expect, that C and A exchange message just like they did before the route change, because GIST does not now about the route change yet. Once GIST sends a refreshing GIST Query (if I remember correctly, the GIST-Query is sent every 30 seconds), B should become the new next peer. Maybe it helps to chat directly using skype? If you think so, please tell me your skypename and I will contact you. Christian Dickmann

1 0

qos-nslp local repair bug - is my GIST set up incorrectly?
by pfer 26 Jul '06

26 Jul '06

Hello! I have a question again. My test setup is composed of 3 nodes: B / | / | / | A | \ | x | \ | C I wished to test the local-repair feature of QoS-NSLP, which, as I understand, roughly does the following: 1. initially A sends to C (cmode, downstream) with direct route A-C 2. signaling is OK, I got response to the reserve from C (direct C-A route), etc., all is fine 3. I pull the C-A wire, faking a failure 4. since my code at A does an "ip route show" every 500 ms, it detects the route change correctly (I use Quagga/ospfd) 5. local-repair at A sends a refresh reserve immediately to C, but on the new route (A-B-C) **** 6. though C receives this new refresh reserve somehow, and I see some Dmode messages at B, my qos-nslp code running on B does not get any recvmessage calls... (also C sends a reply to A, but it does not arrive neither to B, nor to A...) **** I'm thinking about this being a problem of the GIST connection setup. Am I right about this? Suggestions? Thanks in advance, Ferenc Pintér __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

1 0

[Fwd: Regarding NSIS suite]
by Xiaoming Fu 21 Jul '06

21 Jul '06

hi niklas & others, can you help with the attached enquiry about natfw-nslp implementation? xiaoming -------- Original Message -------- Subject: Regarding NSIS suite Date: Wed, 19 Jul 2006 20:48:43 +0300 From: Dinesh Gurnani <ding.iitk(a)gmail.com> To: fu(a)cs.uni-goettingen.de Dear Sir, I am a research assistant at Helsinki University of Technology. I am trying to open pin holes in the firewall using the NATFW NSLP messages. There are 3 machines in my topology. One is external host i.e. outside firewall, one is firewall (no nat) and third one is internal host i.e. behind firewall. Following is the topology configuration which I am using: 100.1.1.5(external host) ----- 100.1.1.6/10.1.2.6(Firewall <http://100.1.1.6/10.1.2.6%28Firewall> without nat) ----- 10.1.2.7(internal host) I have started nsis with pingServer, natfwServer and diagServer daemons on all 3 machines and have opened the udp port 4 (as mentioned in the code) for the traffic on the firewall. But the firewall does not accept any connections. In fact, I have even tried with a machine configured as open for all network traffic but still it does not recognize NSLP messages and does not install specified rules in the Iptables. On scanning ports of the middle machine, I find that there is no process listening for NSLP messages although NSIS with mentioned daemons is running on the machine. I have also gone through the relevant portions of the NSIS documentation but could not clearly identify where is the mistake in my method. Sir, it would be very kind of you if you can suggest me where is the mistake in the way I am proceeding. Thanks in anticipation. Faithfully, Dinesh Gurnani Research Assistant, Telecommunications and Multimedia Laboratory, Helsinki University of Technology.

4 4

RE: [Nsis_imp] mri and directions question
by Christian Dickmann 21 Jul '06

21 Jul '06

Hi, > I'm working on an implementation of QoS-NSLP and its > RMD extension, they are near complete. > I was using them with the -nsis_0.3.0.tar.gz- package, > and found something I do not understand. > > Given a 3 machine mini testbed like this: > > A - B - C > > if I wish to send my nslp-data from A to C, I build an > mri with two ipv4 addresses: A as src and C as dst, > and with direction flag DOWNSTREAM. > Then I call api->SendMessage(). > Fine. > > When B receives this, it has to has to change the src > address in mri to its address (B), before sending to C > to work. > B leaves dst address and direction flag unchanged. > Okay. > > When the message arrives to C, and C needs to send > some response back to A, it changes the mri: > sets its local ip address as src_ip (C), sets ip of A > as destination address, *AND* needs to set direction > to > *DOWNSTREAM* again to make it work. That is not correct. The MRI defines the data flow you are signalling for. That means, that the MRI remains unchanged along the complete path. You only change the direction flag. In your example: 1. A constructs MRI with A as src and C as dst, direction DOWNSTREAM and sends it 2. B receives the message and does not change the MRI at all. The data flow is still from A to C and B is only a router on the path. As the next hop should be C, the direction is still DOWNSTREAM. 3. C receives the message, constructs a response and changes the direction of the MRI to UPSTREAM. But still, the flow is from A to C! Upstream means, that you are now "traveling from dst to src". 4. B receives the message, keeps UPSTREAM as direction 5. A receives the response. Please have a look at our current release 0.4.0. It contains a "skeleton NSLP", which is a NSLP template. It is simply a forwarder, where a message travels from the initiator (A) over the intermediate routers (B) to the receiver (C) and then back to the initiator (to A over B) Maybe this skeleton NSLP can help you as an example. > Sometimes I get GIST segfaults... I would be interested in seeing a backtrace created by GDB or some other debugger. I hope that helps. Christian Dickmann

1 0

mri and directions question
by pfer 21 Jul '06

21 Jul '06

Hi all! I'm working on an implementation of QoS-NSLP and its RMD extension, they are near complete. I was using them with the -nsis_0.3.0.tar.gz- package, and found something I do not understand. Given a 3 machine mini testbed like this: A - B - C if I wish to send my nslp-data from A to C, I build an mri with two ipv4 addresses: A as src and C as dst, and with direction flag DOWNSTREAM. Then I call api->SendMessage(). Fine. When B receives this, it has to has to change the src address in mri to its address (B), before sending to C to work. B leaves dst address and direction flag unchanged. Okay. When the message arrives to C, and C needs to send some response back to A, it changes the mri: sets its local ip address as src_ip (C), sets ip of A as destination address, *AND* needs to set direction to *DOWNSTREAM* again to make it work. I tried sending with flag UPSTREAM as I normally thought, but it did not work. This is for both cmode and dmode I believe. I know very little about NTLP internals, even less on how your implementation work, and I don't really have time to dive in, so please help me on this. Does the reverse-path state mean that sending back from C to A shall work with only setting the flag to UPSTREAM? Is it right what I do? It works this way, but I bet I mess up GIST states and internals quite heavily. Sometimes I get GIST segfaults... What is the right solution? Thanks in advance, Ferenc Pintér PS: I'd be quite happy if that interop I read on the nsis list would realize itself this time. __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

1 0

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Nsis_Imp July 2006