[Openais] Are variants of multicast supported in corosync?

[Sampathkumar, Kishore]

Hi,

On Wednesday, June 23, 2010 9:08 PM Steven Dake wrote:

+ On 05/25/2010 12:53 AM, Sampathkumar, Kishore wrote:
+ > Hi,
+ >
+ > Does corosync support, or are there plans to support, the following:
+ >
+ > (a) Source-Specific Multicast for IP, as specified in rfc4607
+ >
+ > (b) Administratively Scoped IP Multicast, as specified in rfc2365
+ >
+ > The group multicast addresses used today in corosync will reach all
+ > nodes in the network. This may be undesirable, both in terms of
+ > performance and security, as well as for purposes of administering a
+ > data center.
+
+ Honza brought up that he had read one of these RFCs and gave me a brief 
+ explanation atleast of SSM.  I am uncertain how SSM differs from IGMP + 
+ the multlicast used today.  Wouldn't the igmp filter only limit 
+ multicast to those nodes which are subscribed to the IGMP group?

Yes, to my knowledge, the IGMP filter -does- limit multicast to those nodes
that are subscribed to the IGMP group. However, it does really nothing
much towards setting the "scope", and "limit" the extent to which multicast
can swamp your network. See details below about Multicast in IGMP:

The TTL (Time To Live) field in the IP header has a double significance in multicast. As always, it controls the live time of the datagram to avoid it being looped forever due to routing errors. Routers decrement the TTL of every datagram as it traverses from one network to another and when its value reaches 0 the packet is dropped. 

The TTL in IPv4 multicasting has also the meaning of "threshold". Its use becomes evident with an example: suppose you set a long, bandwidth consuming, video conference between all the hosts belonging to your department. You want that huge amount of traffic to remain in your LAN. Perhaps your department is big enough to have various LANs. In that case you want those hosts belonging to each of your LANs to attend the conference, but in any case you want to collapse the entire Internet with your multicast traffic. There is a need to limit how "long" multicast traffic will expand across routers. That's what the TTL is used for. Routers have a TTL threshold assigned to each of its interfaces, and only datagram with a TTL greater than the interface's threshold are forwarded. Note that when a datagram traverses a router with a certain threshold assigned, the datagram's TTL is not decremented by the value of the threshold. Only a comparison is made. (As before, the TTL is decremented by 1 each time a datagram passes across a router). 

A list of TTL thresholds and their associated scope follows: 
--------------------------------------------------------------------------------
TTL     Scope
----------------------------------------------------------------------
   0    Restricted to the same host. Won't be output by any interface.
   1    Restricted to the same subnet. Won't be forwarded by a router.
 <32         Restricted to the same site, organization or department.
 <64 Restricted to the same region.
<128 Restricted to the same continent.
<255 Unrestricted in scope. Global.
--------------------------------------------------------------------------------

The TTL-trick is not always flexible enough for all needs, especially when dealing with overlapping regions or trying to establish geographic, topologic and bandwidth limits simultaneously. To solve this problems, administratively scoped IPv4 multicast regions were established in 1994. (see D. Meyer's "Administratively Scoped IP Multicast" Internet draft). It does scoping based on multicast addresses rather than on TTLs. The range 239.0.0.0 to 239.255.255.255 is reserved for this administrative scoping. 

Regards,
- Kishore

-----Original Message-----
From: Steven Dake [mailto:sdake at redhat.com] 
Sent: Wednesday, June 23, 2010 9:08 PM
To: Sampathkumar, Kishore
Cc: openais at lists.linux-foundation.org
Subject: Re: [Openais] Are variants of multicast supported in corosync?

On 05/25/2010 12:53 AM, Sampathkumar, Kishore wrote:
> Hi,
>
> Does corosync support, or are there plans to support, the following:
>
> (a) Source-Specific Multicast for IP, as specified in rfc4607
>
> (b) Administratively Scoped IP Multicast, as specified in rfc2365
>
> The group multicast addresses used today in corosync will reach all
> nodes in the network. This may be undesirable, both in terms of
> performance and security, as well as for purposes of administering a
> data center.
>
> Thanks,
>
> - Kishore
>

Honza brought up that he had read one of these RFCs and gave me a brief 
explanation atleast of SSM.  I am uncertain how SSM differs from IGMP + 
the multlicast used today.  Wouldn't the igmp filter only limit 
multicast to those nodes which are subscribed to the IGMP group?

Thanks
-steve

> _______________________________________________
> Openais mailing list
> Openais at lists.linux-foundation.org
> https://lists.linux-foundation.org/mailman/listinfo/openais