We have come across an issue with the way C3's handle IGMP leaves and
wonder if anyone else has ?.
It seems that when we have several devices receiving a multicast
feed and one of these sends a IGMP leave the multicast feed stops to
all ports not just the one that
actually sent the IGMP leave. The other devices then see that the feed
has stopped and immediately send out an IGMP join and the multicast
feed is then resumed. So
when the multicast feeds are TV signals this is seen as a frame freeze
or pixelation then the TV signal resumes. In the case of multicast
ghosting PC images this causes a
bit more grief as the imaging process see's this data loss and
restarts from its last checkpoint causing the imaging process of many
workstations to take much longer that is
usually required.
Now the C3 passes the IGMP leave to the CISCO who then removes the
entry for that IP address-to-multicast group address from its IGMP
group membership table entry,
however because the C3 runs a IGMP group cache only ever one
membership join is maintained and seen on the CISCO's router
interface. Thereby the CISCO stops the
multicast feed to the only entry it knows about and then only restarts
the multicast feed when all the devices notice a hlt and start
resending IGMP joins.
This does not happen on an N3 as the its code prevents the IGMP leave
from being forwarded to the router, and simply stops the feed to that
specific port by removing
the entry from its own IGMP group membership table, and if the C3 did
this as well then there would be no issue.
Typically Layer 2 switches should only snoop IGMP traffic and not
participate in the IGMP process as it is a layer 3 protocol, and thats
why we do not see the issue in
our older HP switch architecture.
The workaround we have developed is as follows
1. Created a ingress policy on each C3 switch port that denies IGMP
leaves (224.0.0.2)
set policy rule 5 ipdestsocket 224.0.0.2 mask 32 drop
2. Changed the IGMP Group Membership Interval on the C3 to 30 seconds
from 260
set igmpsnooping groupmembershipinterval 30
3. Changed the Group Membership Interval on the CISCO's router
interface to 10 seconds (uses 3xn as total interval timeout - default
60)
Changing these IGMP timers is so that if the end device changes to
another multicast group then the group it left timeouts a bit quicker.
This does not affect a PC ghosting
application but it can be an issue if someone starts to TV channel
"surf".
What I can also say is in the following two scenarios it works
1. Cisco router ----> N3 ------> C3 -------> n x set top boxes
2. Cisco router ----> N3 -------> n x set top boxes
the problem we are having is in the following configurations
3. Cisco router ----> C3 -------> n x set top boxes
4. Cisco router ----> C3 ------> C3 -------> n x set top boxes
We are also running the latest code base 1.02.02.0009 but have been
experiencing this on previous releases as well.
Cheers
Darren
---------------------------------------------------------------------------------------
Darren Coleman
Senior Network Engineer/Data Network Manager
Networks & Communications
Division of Information
Leonard Huxley Building
The Australian National University
Canberra ACT 0200 Australia
T: +61 2 6125 4627
F: +61 2 6125 8199
M: 0405 153 815
CRICOS Provider #00120C
---
To unsubscribe from enterasys, send email to [EMAIL PROTECTED] with the body:
unsubscribe enterasys [EMAIL PROTECTED]
