3

I am benchmarking multicast throughput of my LAN using an application that considers all the servers as group members.

The network consists of homogeneous servers (S0-S9: 192.168.21.70-79) inter-connected by gigabit switch (DELL powerconnect 5448) with multicast ip: 239.0.0.1

Everything was working well until I decided to split the LAN into two independent LANs: S0-S4 on the first switch with multicast ip: 239.0.0.10 and S5-S9 on the second one with multicast ip: 239.0.0.11 where the two switches are connected via an uplink cable. After running the application on each group simultaneously, the throughput of each group drops down (almost to the half).

Noting that switches are upgraded and configured correctly. The two groups are now using IGMP snooping and querier and filtering bridge multicast messages. One of them is using IGMPv3. From servers side of view, non-concerned servers aren't receiving messages at all.

As a conclusion, switches are forwarding multicast messages to all ports interiorly, while non-concerned ports are blocking them from being transmitted over cables. Am i right? Is there something I have missed?

9
  • Bridge multicast messages are link-specific and will never be replicated to other ports. IGMP is a protocol between a host and a multicast router. Switches can use IGMP snooping to listen to the IGMP conversation, and use that information to forward or block multicast messages on specific ports. Is your switch using IGMP snooping, or do you have it configured as a layer-3 switch?
    – Ron Maupin
    Jan 24 '16 at 16:47
  • Also, you should edit your question to list the two multicast groups you are using since the multicast range can alter the multicast behavior.
    – Ron Maupin
    Jan 24 '16 at 16:53
  • Please check the update. It is layer two switch having the ability of IGMP snooping.
    – Newbie
    Jan 24 '16 at 17:26
  • Your multicast groups are reserved by IANA, You should really use multicast groups in the Organization-Local Scope range of 239.0.0.0-239.255.255.255. A drawing would be really helpful. Is the link between the switches the same speed as the server ports? Are the servers on one switch the sources and the servers on the other switch the destinations, or where are the sources. If the source is the same for both multicast groups, then you should expect the throughput to be cut in half since the bandwidth from the source must be split between traffic for the two multicast groups.
    – Ron Maupin
    Jan 24 '16 at 18:17
  • It is a homogeneous network, all the cables and NICs are 1Gb/s. Groups are totally independent, so the sources are from the group itself. By the way, even if they have the same source, as far as I know, the message will be duplicated inside the switch not from the source.
    – Newbie
    Jan 24 '16 at 18:32