IP Multicast Hardware Lookup Modes

Extreme platforms support various hardware forwarding lookup modes by using a combination of L3 hash table and L2 (FDB (forwarding database)) table. Refer to Multicast Table Management for details on these tables. The scalability limits vary based on the lookup mode used.

Configuration Options

Configuration options allow you to choose the hardware forwarding lookup mode for multicast forwarding. Here is a list of options:
  • source-group-vlan -- Uses L3 hash table with S,G,V lookup. This is the default mode for all platforms.
  • group-vlan -- Uses L3 hash table with *,G,V lookup.
  • mac-vlan -- Uses L2 table with DMAC, VLAN (Virtual LAN) lookup.
  • mixed-mode -- Uses both L2 and L3 tables for multicast. In this mode, the following logic is applied on installing the cache entries in the hardware:
    • Multicast cache entries requiring forwarding across VLANs would be installed in the L3 IP multicast table. This includes PIM, MVR, and PVLAN cache entries.
    • Multicast cache entries requiring L2 forwarding within a VLAN are installed in the L2 table. This includes entries corresponding to IGMP (Internet Group Management Protocol) Snooping, PIM snooping, and MLD snooping.
    • Any IPv4/v6 reserved multicast addresses (for example, 224.0.0.x or IPv6 equivalent) are installed in the L3 IP multicast table as needed. These reserved addresses map to the following multicast MAC addresses: 01:00:5e:00:00:xx, 33:33:00:00:00:xx, 33:33:00:00:01:xx, or 33:33:ff:xx:xx:xx.
    Note

    Note

    Any change in the lookup key configuration causes all cache entries to be cleared, and traffic is temorarily dropped until the system re-learns the multicast caches and associated subscriptions.
    Note

    Note

    mac-vlan mode helps increase scaling. This mode is supported on all platforms.
    Note

    Note

    mac-vlan and mixed-mode are not supported prior to ExtremeXOS 15.3.1.
The EXOS multicast process continues to maintain the cache entries as "S,G,V", and interacts with HAL the same way as today. EXOS hardware abstraction layer (HAL) applies the logic explained above and installs the cache entries in the appropriate hardware table. If the cache entry needs to be installed in the L2 multicast table, HAL derives the MAC address based on the standard logic and installs the MAC entry in the L2 table.

The IP multicast address to MAC address mapping is not validated for the received/forwarded multicast packets in EXOS to date. If the lookup mode is configured either as "mac-vlan" or "mixed-mode", the multicast kernel module is modified to validate this mapping and, if a packet does not use the standard mapping, the packet is dropped.

IPMC Compression

In order to increase the scaling of multicast entries, EXOS implements a feature called IPMC compression which allows multiple <S,G,V> (or <*,G,V>) IP multicast FDB entries to utilize the same IP multicast group table entry when the associated egress port lists are the same. The base IP multicast compression implementation will be reused for achieving L2 multicast entry reuse. In this case, multiple <MAC,VLAN> multicast FDB entries can use a single L2MC index if the egress port lists of the cache entries are the same.

Interactions with Static FDBs

EXOS allows you to create FDB entries for multicast MAC address using:

create fdb mac_addr {vlan} vlan_name ports port_list

These entries also get installed in the L2 table and use the L2MC table for hardware forwarding. If there is a dynamic <MAC,VLAN> entry from MCMGR and a static entry from FDB manager, the static entry takes precedence and the dynamic entry would get deleted in hardware. Compression of L2MC indices is not supported on these types of entries. Each newly created static multicast FDB will cause the allocation of a new L2MC index.

Interactions with Dynamic FDBs

When IP multicast forwarding entries are utilizing the L2 MAC table, the multicast entries are installed as static in the hardware L2 table to avoid undesirable interactions with L2 protocol or user administered FDB flushing. These multicast L2 entries also take precedence over dynamic unicast L2 MAC entries. If there is a hash bucket collision upon inserting an L2 multicast entry, it will replace another dynamic unicast L2 entry if one exists in the same hash bucket.

Virtual Router Support

Current IPMC cache hardware entries stored as <S,G,V> additionally include the VRID associated with the ingress virtual router (VR). In this feature, <MAC, VLAN> cache entries are stored in the L2 table which does not additionally include the VRID. However, user VRs are still supported since the VLAN portion of the lookup key is unique across all VRs.

IPMC Cache Rate Limiting

Based on the number of cache entries supported on each platform, there is a software cache limiting implementation present in EXOS multicast. The HAL module informs MCMGR about the supported limit, MCMGR creates cache entries up to MAX supported limit, and the remaining traffic is dropped in software.

Supported Platforms

This feature is implemented on all platforms.

Limitations

The following limitations exist for the L2MC table feature:

  • When using the "mac-vlan" configuration option:
    • PIMv4/V6, MVR features cannot be used.
    • IGMPv3 should not be used in conjunction with this mode
    • Private VLAN multicast should not be used.
    • Issues with IP multicast address to MAC address mapping:

    All IPv4 multicast frames use multicast mac addresses starting with 01:00:5e:xx:xx:xx. The lower order 23 bits of the IP multicast address is used in the MAC address derivation. As only 23 bits of MAC addresses are available for mapping layer 3 IP multicast addresses to layer 2 MAC addresses, this mapping results in 32:1 address ambiguity.

    When traffic is received for 1 out of these 32 overlapping address, then the MAC, Vlan entry is installed in hardware based on the IGMP group membership of received traffic‘s destination multicast IP address. After this installation, traffic to any of the remaining 31 addresses is delivered based on the existing cache entry and the actual receiver list of the remaining 31 addresses will not be honored.

    IPv6 multicast streams use multicast MAC addresses in the form 33:33:xx:xx:xx:xx. The lower 24 bits of the IPv6 multicast address are used to derive the MAC address. So, the address ambiguity issue is also applicable to IPv6 with more severity. Given this condition, we do not recommend using overlapping IP multicast addresses with this mode.

    This limitation applies to "mixed-mode", too.

  • IPv4 multicast addresses consist of a block of addresses (224.0.0.x) used for network control traffic. Packets having IP destination addresses from the LNCB are always flooded to all ports of the VLAN. The address ambiguity issue discussed above is applicable for the addresses in this block too. For example, 224.0.0.5 (address used for OSPF (Open Shortest Path First)) and 225.0.0.5 would use the same MAC address 01:00:5e:00:00:05. If a mac based multicast FDB entry is installed on the hardware for 01:00:5e:00:00:05 based on the 225.0.0.5 join list, it would break OSPF functionality. Hence, MAC addresses mapping to the LNCB block will not be installed in the L2 table, resulting in software forwarding for those streams. We recommend that you avoid useing multicast addresses that map to the 01:00:5e:00:00:xx MAC address range.
  • As per RFC 3307, IANA assigned reserved IPv6 multicast addresses could be in the group Id range of 0x00000001 to 0x3FFFFFFF. As a result, EXOS switches flood traffic addressed to ff02::/98 to all ports of the VLAN. Since the lower 32 bits of IPv6 multicast addresses are mapped to the multicast mac address, not installing all of the addresses in this range would make it too restrictive. So, installing entries for 33:33:00:00:00:xx in hardware would be avoided, and the traffic would be software forwarded.

    In addition, the following important IPv6 multicast addresses cannot be installed as hardware forwarding entries:

    Therefore, the following multicast MAC addresses are not programmed in hardware, and the corresponding packets are handled in slowpath: 33:33:00:00:01:xx , 33:33:ff:xx:xx:xx
  • Given the issues with IP multicast address to MAC address mapping, no attempt is made to merge subscriber lists of multiple overlapping IP groups.
  • The following limitation regarding IPMC compression is also applicable for this feature, because this feature uses the same L2MC entry for multiple l2 multicast entries with same egress ports.
  • When IP multicast forwarding entries are installed as <MAC,VLAN>, IGMP or MLD packets which have a MAC-DA=<group> will cause the refresh of the IP multicast cache, preventing timely entry age-out.
  • The L2MC table is limited to 1K entries on all platforms. This means that only up to 1K unique port lists can be addressed from the <MAC,VLAN> IP multicast forwarding entries that are stored in the L2 table. Additionally, statically created multicast FDB entries do not perform L2MC index compression.