Re: [magma] Igmp Group-Source specific Retransmission

[Bharat Joshi <[e-mail]>]

Deepak,

     I think your interpretation is perfect and I think that is exactly how one should implement it on router side.

     BTW, Just one question. Do you see any issue with this?

Thanks,
Bharat
________________________________________
From: Kudachi, Deepak S [[e-mail]]
Sent: Thursday, June 03, 2010 3:15 PM
To: Bharat Joshi; [e-mail]; [e-mail]
Cc: Sankaran, Balaji (HP Procurve)
Subject: RE: [magma]  Igmp Group-Source specific Retransmission

Hi Bharat,

I am still not clear on the part that you said all the queries have different src addresses.

May be I can give an example for this case.

Let us assume following values.


Robustness = 2
LMQI = 10 sec

Group = G1 = 224.0.2.1
Srouces
        S1 = 10.10.10.10
        S2 = 10.10.10.20
        S3 = 10.10.10.30
        S4 = 10.10.10.40
        S5 = 10.10.10.50

Our interface's current state is INCLUDE for group G1.

So
INCLUDE{S1,S2,S3,S4} for G1

Now we received Block message for the same group on same interface

Consider the message received is
BLOCK {S3}.

now according to RFC 3376 sec 6.4.2
INCLUDE (A)    BLOCK (B)    INCLUDE (A)             Send Q(G,A*B)

We need to send a group-source specific query for A*B
which is nothing but send a query for source S3.

So query will be sent at Time = T1
Send Q(G1, S3)
We have started the timer (LMQI)10 sec with repeat 1 so that
When timer expires we need to send one more query for this G1,S3.
( query is send out for group G1= 224.0.1.2 ,S3 = 10.10.10.30.)

now at time  interval T2 < (T1+LMQI)
we received TO_EXC{ S1 , S2, S5 } on same interface for the same group G1.

Now according to RFC following function need to be performed.
INCLUDE (A)    TO_EX (B)    EXCLUDE (A*B,B-A)       (B-A)=0
                                                    Delete (A-B)
                                                    Send Q(G,A*B)
                                                    Group Timer=GMI


Interface moves to exclude mode
EXCLUDE { (S1,S2),(S5)} for group G1.

Now immediate query has to be sent for S1,S2.

Since S3 retransmission is still pending we need to merge S3 with S1,S3 in this query.
Now Q(G1,S1,S2,S3) will be sent at interval T2.

Now for S3 we have sent two queries and its query count is set to 0 and will be
removed from retransmission list.

For S3 we sent out second query much before T1+LMQI and no more queries will be sent for S3.

This is what we infer from the RFC and just want to know others view on this.

(please Note: I am only considering router side processing).


Thanks and Regards
Deepak

-----Original Message-----
From: Bharat Joshi [mailto:[e-mail]]
Sent: Wednesday, June 02, 2010 6:07 PM
To: Kudachi, Deepak S; [e-mail]; [e-mail]
Subject: RE: [magma] Igmp Group-Source specific Retransmission

Deepak,

      It would be better if you handle the example with IP addresses. Also consider how the state in the hosts are and what final state should be there after things stabilize.

      For the below example, I think you should attach the robustness count as query count with the source addresses and then as and when timer fires, you prepare a group-source specific query, send it out and reduce the query count for each source address you include in the query. Once the query count for a source address reaches zero, it is removed from the list and when all sources are removed from the list, we do not restart the timer.

      So we will end up sending three queries. But all the three queries will carry different sources in the list.

      I hope this helps.

Regards,
Bharat
________________________________________
From: [e-mail] [[e-mail]] On Behalf Of Kudachi, Deepak S [[e-mail]]
Sent: Wednesday, June 02, 2010 5:50 PM
To: [e-mail]; [e-mail]
Subject: [magma]  Igmp Group-Source specific Retransmission

Hi All,

I was going through the IGMPv3 RFC 3376  and was not clear on the below
Mentioned section.

/*---------------------------------------------------------------------------------------------------------*/
6.6.3.2. Building and Sending Group and Source Specific Queries

   When a table action "Send Q(G,X)" is encountered by a querier in the
   table in section 6.4.2, the following actions must be performed for
   each of the sources in X of group G, with source timer larger than
   LMQT:

   o Set number of retransmissions for each source to [Last Member Query
     Count].

   o Lower source timer to LMQT.

   The router must then immediately send a group and source specific
   query as well as schedule [Last Member Query Count - 1] query
   retransmissions to be sent every [Last Member Query Interval] over
   [Last Member Query Time].  The contents of these queries are
   calculated as follows.
/*-----------------------------------------------------------------*/


This section talks about how the group-source specific queries are build and
Send out.

Suppose our robustness variable is set to 2(default value).
And Last Member Query Interval is 10 Sec.

Any group-source specific query need to be transmitted robustness number of times
That is 2 times in our case and separated by LMQI which 10 sec.
So 2 queries has to be sent across Last Member Query Time which is 20 sec.


Now suppose we already sent Q(G,X) query for a group at T1 with set of sources. And we received
A report which causes new Q(G,(X+Y)) needs to be sent and T2( before T1+LMQI). Where Y is new set of sources to be
Queried. New Query will be sent with all sources merged from previous and new(x + y).

So the sources x already two queries will be sent and not separated by LMQI in this scenario.

My question is whether we need to send one more query after LMQI ( at T1+LMQI) for x
Set of sources for this group?

If we need to then we will be sending 3 queries for x set of sources as follows.

Time interval T1
Time interval T2
Time interval T1+LMQI

Which is not correct.

Let me know whether we should send query for x set of sources for this group
At T1+LMQI interval.

Thanks and Regards
Deepak