stmonica
asked on
Network Problem - IGMPv2 Storm
Hi all,
We are currently suffering from an IGMPv2 storm on our network on a single VLAN across multiple HP switches. Wireshark records over 2000000 IGMP packets within a few seconds. Unsurprisingly clients are unable to get network access while this is happening. We have roughly 50 WAP's on this VLAN and some client PC's. I'm struggling to tell which WAP / client is causing the issue as the traffic all seems to relate to the clients responding to the IGMP packets. I don't know how to work out which one is causing them to all respond.
A snapshot of the wireshark log is attached. The destination address's appear to be multicast and can be 224.0.0.252, 224.0.0.2 or 224.0.0.113
I've tried physically unplugging switches one by one to try and work out which one has the offending client but I can't get a definitive answer. I assume because packets are moving throughout the network. When I do this the storm will "calm down" but be back within a few hours.
Any help in working out what's going on would be greatly appreciated.
Max.
Wireshark-Conversation-Snapshot.PNG
Wireshark-Snapshot-IGMP.PNG
We are currently suffering from an IGMPv2 storm on our network on a single VLAN across multiple HP switches. Wireshark records over 2000000 IGMP packets within a few seconds. Unsurprisingly clients are unable to get network access while this is happening. We have roughly 50 WAP's on this VLAN and some client PC's. I'm struggling to tell which WAP / client is causing the issue as the traffic all seems to relate to the clients responding to the IGMP packets. I don't know how to work out which one is causing them to all respond.
A snapshot of the wireshark log is attached. The destination address's appear to be multicast and can be 224.0.0.252, 224.0.0.2 or 224.0.0.113
I've tried physically unplugging switches one by one to try and work out which one has the offending client but I can't get a definitive answer. I assume because packets are moving throughout the network. When I do this the storm will "calm down" but be back within a few hours.
Any help in working out what's going on would be greatly appreciated.
Max.
Wireshark-Conversation-Snapshot.PNG
Wireshark-Snapshot-IGMP.PNG
Is IGMP snooping configured? It doesn't sound like it.
If you google for those IPs - it must be Cisco routers chatting.
IGMP is normal network traffic if you use multicast.
IGMP is normal network traffic if you use multicast.
It's PIM traffic. Can you post the switch configs?
Switches dont do PIM talk. It is routers.
Layer 3 switches do, and other network devices (such as routers) do. The configs will help us to understand what the network looks like.
Layer 3 is not switches. It is a router.
Layer 3 is a router. https://en.wikipedia.org/wiki/Network_switch#Layer_3
Destination is Multicast (ie subscribers).
The source address is where is comes from.
The source address is where is comes from.
Souce MAC should be traceable.
Gheist, why are you arguing? Have you never heard of a layer3 switch?
A layer 3 switch more or less is a router bolted onto a switch. And layer-3 is by definition where routing takes place.
So there is no need for arguing by any one. A L3 switch also does everything a L2 switch does, it has to as part of the stuff in networks is done on L2.
So there is no need for arguing by any one. A L3 switch also does everything a L2 switch does, it has to as part of the stuff in networks is done on L2.
Exactly my point, noci. I don't understand the confusion.
I've asked for configs to see whether the switches are configured for snooping, amongst other things. Can we see configs please, stmonica?
I've asked for configs to see whether the switches are configured for snooping, amongst other things. Can we see configs please, stmonica?
Now on IGMP....
An IGMP membership announcement is done (a few times to be sure, on a lossy protol like UDP) when a multicast port is opened on a computer (host), indicating to an upstream router (and possibly switches [ IGMP-Snooping ] ) that it wants to receive multicast traffic. The Leave is sent when the port is closed.
(A router that manages Multicast traffic can then selectively pass on Multicast traffic, same for switches that do IGMP-Snooping that can select to only forward to ports that need it in stead of all ports). So snooping is a performance option, where systems not interested in multicast will not get the traffic as well.
Is there software active that only runs for short times that happens to use multicast traffic? That can explain a lot of enter/leave group messages.
The other option is that you run a farm of systems that use multicast: to address 224.0.0.113?
At least the systems *.48, *.31, *.32, *.33, *.34 and *.35 are active on that channel.
The 224.0.0.2 traffic is an indication of the amount of announcements.. for IGMP, for the number of times ports are closed. The opens should be to the whole group all hosts.
Due to it's nature Multicast is meant for longer duration of massive amounts of traffic from one source to multiple subscribers.
(TV & Radio broadcast are equivalent technologies).
224.0.0.2 - is All Routers... (As such it can be used to announce a host that is interested on the network. So PIM in a router can take note of the registration).
224.0.0.113 used to be used by the ALLJoyn framework, but shouldn't be now. as it also uses the mDNS group for that.
An IGMP membership announcement is done (a few times to be sure, on a lossy protol like UDP) when a multicast port is opened on a computer (host), indicating to an upstream router (and possibly switches [ IGMP-Snooping ] ) that it wants to receive multicast traffic. The Leave is sent when the port is closed.
(A router that manages Multicast traffic can then selectively pass on Multicast traffic, same for switches that do IGMP-Snooping that can select to only forward to ports that need it in stead of all ports). So snooping is a performance option, where systems not interested in multicast will not get the traffic as well.
Is there software active that only runs for short times that happens to use multicast traffic? That can explain a lot of enter/leave group messages.
The other option is that you run a farm of systems that use multicast: to address 224.0.0.113?
At least the systems *.48, *.31, *.32, *.33, *.34 and *.35 are active on that channel.
The 224.0.0.2 traffic is an indication of the amount of announcements.. for IGMP, for the number of times ports are closed. The opens should be to the whole group all hosts.
Due to it's nature Multicast is meant for longer duration of massive amounts of traffic from one source to multiple subscribers.
(TV & Radio broadcast are equivalent technologies).
224.0.0.2 - is All Routers... (As such it can be used to announce a host that is interested on the network. So PIM in a router can take note of the registration).
224.0.0.113 used to be used by the ALLJoyn framework, but shouldn't be now. as it also uses the mDNS group for that.
btw. 1Mbps seems a lot, but is roughly 0.1% of a 1Gbps line.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
other comments did not resolve the issue.