Supported Route Table Entries
Cumulus Linux advertises the maximum number of route table entries supported on the switch, including:
- Layer 3 IPv4 LPM (longest prefix match) 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, use either the NCLU net show system asic
command or cl-resource-query
.
Supported Route Entries
Cumulus Linux provides several generalized profiles, described below. These profiles work only with layer 2 and layer 3 unicast forwarding.
The following tables list the number of MAC addresses, layer 3 neighbors, and LPM routes validated for each forwarding table profile. If you do not specify any profiles as described below, the switch uses the default values.
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 have been validated as part of the 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-2 and Spectrum-3
Profile | MAC Addresses | Layer 3 Neighbors | Longest Prefix Match (LPM) |
---|---|---|---|
default | 50k | 41k (IPv4) and 20k (IPv6) | 82k (IPv4), 74k (IPv6-long), 10K (IPv4-Mcast) |
l2-heavy | 115k | 74k (IPv4) and 37k (IPv6) | 16k (IPv4), 24k (IPv6-long), 10K (IPv4-Mcast) |
l2-heavy-1 | 239K | 16k (IPv4) and 12k (IPv6) | 16k (IPv4), 16k (IPv6-long), 10K (IPv4-Mcast) |
v4-lpm-heavy | 16k | 41k (IPv4) and 24k (IPv6) | 124k (IPv4), 24k (IPv6-long), 10K (IPv4-Mcast) |
v4-lpm-heavy-1 | 16k | 16k (IPv4) and 4k (IPv6) | 256k (IPv4), 8k (IPv6-long), 10K (IPv4-Mcast) |
v6-lpm-heavy | 16k | 16k (IPv4) and 62k (IPv6) | 16k (IPv4), 99k (IPv6-long), 10K (IPv4-Mcast) |
lpm-balanced | 16k | 16k (IPv4) and 12k (IPv6) | 124k (IPv4), 124k (IPv6-long), 10K (IPv4-Mcast) |
ipmc-heavy | 57k | 41k (IPv4) and 20k (IPv6) | 82K (IPv4), 66K (IPv6), 8K (IPv4-Mcast) |
ipmc-max | 41K | 41k (IPv4) and 20k (IPv6) | 74K (IPv4), 66K (IPv6), 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.
Spectrum
Profile | MAC Addresses | Layer 3 Neighbors | Longest Prefix Match (LPM) |
---|---|---|---|
default | 40k | 32k (IPv4) and 16k (IPv6) | 64k (IPv4) and 28k (IPv6-long) |
l2-heavy | 88k | 48k (IPv4) and 40k (IPv6) | 8k (IPv4) and 8k (IPv6-long) |
l2-heavy-1 | 180K | 8k (IPv4) and 8k (IPv6) | 8k (IPv4) and 8k (IPv6-long) |
v4-lpm-heavy | 8k | 8k (IPv4) and 16k (IPv6) | 80k (IPv4) and 16k (IPv6-long) |
v4-lpm-heavy-1 | 8k | 8k (IPv4) and 2k (IPv6) | 176k (IPv4) and 2k (IPv6-long) |
rash-custom-profile1 | 20k | 8k (IPv4) and 1k (IPv6) | 8k (IPv4) and 1k (IPv6-long) |
v6-lpm-heavy | 40k | 8k (IPv4) and 40k (IPv6) | 8k (IPv4), 32k (IPv6-long) and 32K (IPv6/64) |
lpm-balanced | 8k | 8k (IPv4) and 8k (IPv6) | 60k (IPv4), 60k (IPv6-long) and 120k (IPv6/64) |
Forwarding Resource Profiles
You can configure forwarding resource allocations. Choose the profile that best suits your network architecture.
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 ipmc-max:
cumulus@switch:~$ sudo cat /etc/cumulus/datapath/traffic.conf
# Specify the forwarding table resource allocation profile, applicable
# only on platforms that support universal forwarding resources.
#
# /usr/cumulus/sbin/cl-resource-query reports the allocated table sizes
# based on the profile setting.
#
# Values: one of { *** Common ***
# 'default', 'l2-heavy', 'l2-heavy-1', 'l2-heavy-2',
# 'v4-lpm-heavy', 'v4-lpm-heavy-1', 'v6-lpm-heavy',
# 'rash-v4-lpm-heavy', 'rash-custom-profile1',
# 'rash-custom-profile2', 'lpm-balanced'
#
# *** Mellanox Spectrum2+ only platforms ***
# 'ipmc-heavy', 'ipmc-max'
#
# }
#
# Default value: 'default'
# Notes: some devices may support more modes, please consult user
# guide for more details
#
forwarding_table.profile = ipmc-max
After you specify a different profile, restart switchd
for the change to take effect.
Spectrum TCAM Profiles
Specify the profile you want to use with the tcam_resource.profile
variable in the /etc/mlx/datapath/tcam_profile.conf
file. The following example specifies ipmc-max:
cumulus@switch:~$ cat /etc/mlx/datapath/tcam_profile.conf
#
# Default tcam_profile configuration for Mellanox Spectrum chip
# Copyright (C) 20xx-2021 NVIDIA Corporation. ALL RIGHTS RESERVED.
#
#TCAM resource forwarding profile
# Applicable for Spectrum-1 and Spectrum-A1 switches only
# Valid profiles -
# default, ipmc-heavy, acl-heavy, ipmc-max, ip-acl-heavy
tcam_resource.profile = ipmc-max
After you specify a different profile, restart switchd
for the change to take effect.
When you enable nonatomic updates (acl.non_atomic_update_mode
is TRUE
in the /etc/cumulus/switchd.conf
file), the maximum number of mroute and ACL entries for each profile are:
Profile | Mroute Entries | ACL Entries |
---|---|---|
default | 1000 | 500 (IPv6) or 1000 (IPv4) |
ipmc-heavy | 8500 | 1000 (IPv6) or 1500 (IPv4) |
acl-heavy | 450 | 2000 (IPv6) or 3500 (IPv4) |
ipmc-max | 13000 | 1000 (IPv6) or 2000 (IPv4) |
When you disable nonatomic updates (acl.non_atomic_update_mode
is FALSE
in the /etc/cumulus/switchd.conf
file), the maximum number of mroute and ACL entries for each profile are:
Profile | Mroute Entries | ACL Entries |
---|---|---|
default | 1000 | 250 (IPv6) or 500 (IPv4) |
ipmc-heavy | 8500 | 500 (IPv6) or 750 (IPv4) |
acl-heavy | 450 | 1000 (IPv6) or 1750 (IPv4) |
ipmc-max | 13000 | 500 (IPv6) or 1000 (IPv4) |