Fragmented multicast on VLAN?

[camber_shill]

Configuration:

catalyst 3560 switch

5 vlans with 4 ports each, 1 vlan with 8 ports, 1 vlan with 16 ports

system mtu jumbo set to 9014

igmnp snooping on or off not making a difference

no routing between vlans

the 5 4port vlans consist of one multicast generator and two receivers plus a spare port

the other two vlans are just normal traffic with nominal bandwidth

when the multicast packet size is limited in software to less than 9000 bytes everything works fine but the performance is bad due to system overhead on the generators as i am actually wanting to send 1440000 bytes for each packet and am having to break up into 160+ packets. taking approx 24msec to transmit all packets

if i change to packet size to 64,000 bytes (less than 64K max) no errors are reported but very very few of the packets are received although the traffic is apparent on the network.  this must be a problem in the switch with handling the fragmented packets ( each frag 64000/MTU).  this was not an issue on our previous dlink switch (no longer available).  note my total transmit time drops to 16msec but is usless as the data is not properly received by the receivers.

what am i missing in the configuration?

[Laurent Aubert]

Hi,

The switch will drop any received packets higher than the configured jumbo MTU. Also the jumbo MTU applied only to GE interface. If your receivers are on 10/100 interfaces, the max MTU supported is 1998Bytes.

Jumbo frames can only be switched and not routed.

For more information: http://www.cisco.com/en/US/docs/switches/lan/catalyst3560/software/release/12.2_44_se/command/reference/cli3.html#wp1948947

HTH

Laurent.

[camber_shill]

I guess i was not clear.

all the NIC attached to the ports are gigabit.

the generator NIC/OS fragments the large 64,000 byte packets so they are smaller than the MTU (standard UDP fragmentation)

as stated there is no routing between VLANs.  all traffic is within each VLAN.  the jumbo frames work fine unless there has been fragmentation.

which is why i suspect some kind of multicast setting rather than being a jumbo frame issue.

I can run the exact same software on my generators and receivers when hooked to the dlink switch and everything works.  switch cable to the cisco switch and it no longer works.  I cannot find any settings that would explain this but most settings are "hidden" in the CLI for the cisco switch.  by hidden i mean they are not available from the network assistant.

i would continue with the dlink but it does not have enough ports and is continuing to have ports go bad requiring frequent reboots.

[Laurent Aubert]

Hi,

Did you notice any packets drop ? Can you check each receiver is receiving the right number of pps ?

You are saying it's working fine without any fragmentation so it makes me think the receiver may have difficulties to rebuild the original 64k PDU.

Thanks

Laurent.

[camber_shill]

bump.  anyone?

[camber_shill]

Turns out it only occurs when there is a NIC on the receiver that is configured to receive data

from more than one multicast group.  This screws up all the other NICs as well.  as long as none of the packets are fragmented then everything works.  packets on other NICs can be fragmented and all still works.  if any of the packets being received on the NIC from two multicast sources (226.0,0.201 and 226.0.0.202) are fragmented the whole system goes crazy.  I am rewriting my applications so i can configure the maximum packet size per multicast interface so i can improve the performance on those which have only one multicast group per NIC.