Forwarding Table Size and Profiles
Cumulus Linux advertises the maximum number of forwarding table entries supported on the switch, including:
- Layer 3 IPv4 LPM entries that have a mask less than /32
- Layer 3 IPv6 LPM entries that have a mask of /64 or less
- Layer 3 IPv6 LPM entries that have a mask greater than /64
- Layer 3 IPv4 neighbor (or host) entries that are the next hops seen in
ip neighbor
- Layer 3 IPv6 neighbor entries that are the next hops seen in
ip -6 neighbor
- ECMP next hops, which are IP address entries in the routing table that specify the next closest or most optimal router in its routing path
- MAC addresses
To determine the current table sizes on a switch, run the NVUE nv show platform asic resource
command or the Linux cl-resource-query
command.
Each switching architecture has specific resources available for forwarding table entries. Cumulus Linux stores:
- Forwarding table resources in a KVD.
- ACL table entries and other switching functions in a fast memory area called the TCAM on Spectrum 1, and the ATCAM on Spectrum-2 and later.
Cumulus Linux provides various general profiles for forwarding table resources, and, based on your network design, you might need to adjust various switch parameters to allocate resources, as needed.
The values provided in the profiles below are the maximum values that Cumulus Linux software allocates; the theoretical hardware limits might be higher. These limits refer to values that NVIDIA checks as part of unidimensional scale validation. If you try to achieve maximum scalability with multiple features enabled, results might differ from the values listed in this guide.
Spectrum 1
Forwarding resource profiles control unicast forwarding table entry allocations. On the Spectrum 1 switch, TCAM profiles control multicast forwarding table entry allocations. For more information about multicast route entry limitations, refer to Hardware Limitations for ACL Rules.
Profile | MAC Addresses | Layer 3 Neighbors | LPM |
---|---|---|---|
default | 40k | 32k (IPv4) and 8k (IPv6) | 64k (IPv4) and 22k (IPv6-long) |
l2-heavy | 88k | 48k (IPv4) and 18k (IPv6) | 8k (IPv4) and 8k (IPv6-long) |
l2-heavy-1 | 176k | 4k (IPv4) and 2k (IPv6) | 4k (IPv4) and 2k (IPv6-long) |
l2-heavy-2 | 86k | 86k (IPv4) and 4k (IPv6) | 8k (IPv4), 4k (IPv6-long) |
v4-lpm-heavy | 8k | 8k (IPv4) and 16k (IPv6) | 80k (IPv4) and 16k (IPv6-long) |
v4-lpm-heavy-1 | 6k | 6k (IPv4) and 2k (IPv6) | 176k (IPv4) and 2k (IPv6-long) |
v6-lpm-heavy | 27k | 8k (IPv4) and 36k (IPv6) | 8k (IPv4), 32k (IPv6-long) and 32k (IPv6/64) |
lpm-balanced | 6k | 4k (IPv4) and 3k (IPv6) | 60k (IPv4), 60k (IPv6-long) and 120k (IPv6/64) |
ecmp-nh-heavy | 20K | 32k (IPv4) and 4k (IPv6) | 50k (IPv4) 4k (IPv6-long) |
The ecmp-nh-heavy
profile does not support warm restart mode.
Spectrum-2 and Later
On Spectrum-2 and later, forwarding resource profiles control both unicast and multicast forwarding table entry allocations.
Profile | MAC Addresses | Layer 3 Neighbors | LPM |
---|---|---|---|
default | 50k | 41k (IPv4) and 20k (IPv6) | 82k (IPv4), 74k (IPv6-short), 1k (IPv4-Mcast) |
l2-heavy | 115k | 74k (IPv4) and 37k (IPv6) | 16k (IPv4), 24k (IPv6-short), 1k (IPv4-Mcast) |
l2-heavy-1 | 239k | 16k (IPv4) and 12k (IPv6) | 16k (IPv4), 16k (IPv6-short), 1k (IPv4-Mcast) |
l2-heavy-2 | 124k | 132k (IPv4) and 12k (IPv6) | 16k (IPv4), 16k (IPv6-short), 1k (IPv4-Mcast) |
l2-heavy-3 | 107k | 90k (IPv4) and 80k (IPv6) | 25k (IPv4), 10k (IPv6-short), 1k (IPv4-Mcast) |
v4-lpm-heavy | 16k | 41k (IPv4) and 24k (IPv6) | 124k (IPv4), 24k (IPv6-short), 1k (IPv4-Mcast) |
v4-lpm-heavy-1 | 16k | 16k (IPv4) and 4k (IPv6) | 256k (IPv4), 8k (IPv6-short), 1k (IPv4-Mcast) |
v6-lpm-heavy | 16k | 16k (IPv4) and 62k (IPv6) | 16k (IPv4), 99k (IPv6-short), 1k (IPv4-Mcast) |
v6-lpm-heavy-1 | 5k | 4k (IPv4) and 4k (IPv6) | 90k (IPv4), 235k (IPv6-short), 1k (IPv4-Mcast) |
lpm-balanced | 16k | 16k (IPv4) and 12k (IPv6) | 124k (IPv4), 124k (IPv6-short), 1k (IPv4-Mcast) |
ipmc-heavy | 57k | 41k (IPv4) and 20k (IPv6) | 82k (IPv4), 66k (IPv6-short), 8k (IPv4-Mcast) |
ipmc-max | 41K | 41k (IPv4) and 20k (IPv6) | 74k (IPv4), 66k (IPv6-short), 15k (IPv4-Mcast) |
The IPv6 number corresponds to the /64 IPv6 prefix. The /128 IPv6 prefix number is half of the /64 IPv6 prefix number.
For the ipmc-max profile, the cl-resource-query
command output displays 33K instead of 15K as the maximum number of IPv4 multicast routes in switchd
. 15K is the supported and validated value. You can use the higher value of 33K to test higher multicast scale in non-production environments.
Change Forwarding Resource Profiles
You can set the profile that best suits your network architecture.
Run the nv set system forwarding profile <profile-name>
command to specify the profile you want to use.
The following example command sets the l2-heavy
profile:
cumulus@switch:~$ nv set system forwarding profile l2-heavy
cumulus@switch:~$ nv config apply
To set the profile back to the default:
cumulus@switch:~$ nv unset system forwarding profile l2-heavy
cumulus@switch:~$ nv config apply
Instead of the above command, you can run the nv set system forwarding profile default
command to set the profile back to the default.
Specify the profile you want to use with the forwarding_table.profile
variable in the /etc/cumulus/datapath/traffic.conf
file. The following example specifies l2-heavy
:
cumulus@switch:~$ sudo cat /etc/cumulus/datapath/traffic.conf
...
forwarding_table.profile = l2-heavy
After you specify a different profile, restart switchd
with the sudo systemctl restart switchd.service
command.
To show the different forwarding profiles that your switch supports and the MAC address, layer 3 neighbor, and LPM scale availability for each forwarding profile, run the nv show system forwarding profile-option
command.
ACL and VLAN Memory Resources
In addition to forwarding table memory resources, there are limitations on other memory resources for ACLs and VLAN interfaces; refer to Hardware Limitations for ACL Rules.