Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 1 | .. _scale: |
| 2 | |
| 3 | Scale |
| 4 | ----- |
| 5 | |
| 6 | The only limiting factor on FIB scale is the amount of memory |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 7 | allocated to each heap the FIB uses, and there are 2: |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 8 | |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 9 | * The main heap |
| 10 | * The stats heap |
| 11 | |
| 12 | |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 13 | Main Heap |
| 14 | ^^^^^^^^^ |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 15 | |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 16 | The main heap is used to allocate all memory needed for the FIB |
| 17 | data-structures. Each table, created by the user, i.e. with; |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 18 | |
| 19 | .. code-block:: console |
| 20 | |
| 21 | $ ip table add 1 |
| 22 | |
| 23 | or the default table, comprises 2 *ip4_fib_t* objects. |
| 24 | The 'non-forwarding' *ip4_fib_t* contains all the entries in the table |
| 25 | and, the 'forwarding' contains the entries that are matched against in |
| 26 | the data-plane. The difference between the two sets are the entries |
| 27 | that should not be matched in the data-plane. |
| 28 | Each *ip4_fib_t* comprises an mtrie (for fast lookup in the data-plane) |
| 29 | and a hash table per-prefix length (for lookup in the control plane). |
| 30 | |
| 31 | To see the amount of memory consumed by the IPv4 tables use: |
| 32 | |
| 33 | .. code-block:: console |
| 34 | |
| 35 | vpp# sh ip fib mem |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 36 | ipv4-VRF:0 mtrie:335744 hash:4663 |
| 37 | ipv4-VRF:1 mtrie:333056 hash:3499 |
| 38 | totals: mtrie:668800 hash:8162 all:676962 |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 39 | |
| 40 | this output shows two 'empty' (i.e. no added routes) tables. Each |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 41 | mtrie uses about 150k of memory, so each table about 300k. |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 42 | |
| 43 | |
| 44 | Below the output having added 1M, 2M and 4M routes respectively: |
| 45 | |
| 46 | .. code-block:: console |
| 47 | |
| 48 | vpp# sh ip fib mem |
| 49 | ipv4-VRF:0 mtrie:335744 hash:4695 |
| 50 | totals: mtrie:335744 hash:4695 all:340439 |
| 51 | |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 52 | .. code-block:: console |
| 53 | |
| 54 | vpp# sh ip fib mem |
| 55 | ipv4-VRF:0 mtrie:5414720 hash:41177579 |
| 56 | totals: mtrie:5414720 hash:41177579 all:46592299 |
| 57 | |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 58 | .. code-block:: console |
| 59 | |
| 60 | vpp# sh ip fib mem |
| 61 | ipv4-VRF:0 mtrie:22452608 hash:168544508 |
| 62 | totals: mtrie:22452608 hash:168544508 all:190997116 |
| 63 | |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 64 | |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 65 | IPv6 also has the concept of forwarding and non-forwarding entries, |
| 66 | however for IPv6 all the forwarding entries are stored in a single |
| 67 | hash table (same goes for the non-forwarding). The key to the hash |
| 68 | table includes the IPv6 table-id. |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 69 | |
| 70 | To see the amount of memory consumed by the IPv4 tables use: |
| 71 | |
| 72 | .. code-block:: console |
| 73 | |
| 74 | vpp# sh ip6 fib mem |
| 75 | IPv6 Non-Forwarding Hash Table: |
| 76 | Hash table ip6 FIB non-fwding table |
| 77 | 7 active elements 7 active buckets |
| 78 | 1 free lists |
| 79 | 0 linear search buckets |
| 80 | arena: base 7f2fe28bf000, next 803c0 |
| 81 | used 525248 b (0 Mbytes) of 33554432 b (32 Mbytes) |
| 82 | |
| 83 | IPv6 Forwarding Hash Table: |
| 84 | Hash table ip6 FIB fwding table |
| 85 | 7 active elements 7 active buckets |
| 86 | 1 free lists |
| 87 | 0 linear search buckets |
| 88 | arena: base 7f2fe48bf000, next 803c0 |
| 89 | used 525248 b (0 Mbytes) of 33554432 b (32 Mbytes) |
| 90 | |
| 91 | as we scale to 128k IPv6 entries: |
| 92 | |
| 93 | .. code-block:: console |
| 94 | |
| 95 | vpp# sh ip6 fib mem |
| 96 | IPv6 Non-Forwarding Hash Table: |
| 97 | Hash table ip6 FIB non-fwding table |
| 98 | 131079 active elements 32773 active buckets |
| 99 | 2 free lists |
| 100 | [len 1] 2 free elts |
| 101 | 0 linear search buckets |
| 102 | arena: base 7fed7a514000, next 4805c0 |
| 103 | used 4720064 b (4 Mbytes) of 1073741824 b (1024 Mbytes) |
| 104 | |
| 105 | IPv6 Forwarding Hash Table: |
| 106 | Hash table ip6 FIB fwding table |
| 107 | 131079 active elements 32773 active buckets |
| 108 | 2 free lists |
| 109 | [len 1] 2 free elts |
| 110 | 0 linear search buckets |
| 111 | arena: base 7fedba514000, next 4805c0 |
| 112 | used 4720064 b (4 Mbytes) of 1073741824 b (1024 Mbytes) |
| 113 | |
| 114 | and 256k: |
| 115 | |
| 116 | .. code-block:: console |
| 117 | |
| 118 | vpp# sh ip6 fib mem |
| 119 | IPv6 Non-Forwarding Hash Table: |
| 120 | Hash table ip6 FIB non-fwding table |
| 121 | 262151 active elements 65536 active buckets |
| 122 | 2 free lists |
| 123 | [len 1] 6 free elts |
| 124 | 0 linear search buckets |
| 125 | arena: base 7fed7a514000, next 880840 |
| 126 | used 8915008 b (8 Mbytes) of 1073741824 b (1024 Mbytes) |
| 127 | |
| 128 | IPv6 Forwarding Hash Table: |
| 129 | Hash table ip6 FIB fwding table |
| 130 | 262151 active elements 65536 active buckets |
| 131 | 2 free lists |
| 132 | [len 1] 6 free elts |
| 133 | 0 linear search buckets |
| 134 | arena: base 7fedba514000, next 880840 |
| 135 | used 8915008 b (8 Mbytes) of 1073741824 b (1024 Mbytes) |
| 136 | |
| 137 | and 1M: |
| 138 | |
| 139 | .. code-block:: console |
| 140 | |
| 141 | vpp# sh ip6 fib mem |
| 142 | IPv6 Non-Forwarding Hash Table: |
| 143 | Hash table ip6 FIB non-fwding table |
| 144 | 1048583 active elements 65536 active buckets |
| 145 | 4 free lists |
| 146 | [len 1] 65533 free elts |
| 147 | [len 2] 65531 free elts |
| 148 | [len 4] 9 free elts |
| 149 | 0 linear search buckets |
| 150 | arena: base 7fed7a514000, next 3882740 |
| 151 | used 59254592 b (56 Mbytes) of 1073741824 b (1024 Mbytes) |
| 152 | |
| 153 | IPv6 Forwarding Hash Table: |
| 154 | Hash table ip6 FIB fwding table |
| 155 | 1048583 active elements 65536 active buckets |
| 156 | 4 free lists |
| 157 | [len 1] 65533 free elts |
| 158 | [len 2] 65531 free elts |
| 159 | [len 4] 9 free elts |
| 160 | 0 linear search buckets |
| 161 | arena: base 7fedba514000, next 3882740 |
| 162 | used 59254592 b (56 Mbytes) of 1073741824 b (1024 Mbytes) |
| 163 | |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 164 | as can be seen from the output the IPv6 hash-table in this case was scaled |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 165 | to 1GB and 1million prefixes has used 56MB of it. |
| 166 | |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 167 | The main heap is also used to allocate objects that represent the FIB |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 168 | entries in the control and data plane (see :ref:`controlplane` and |
| 169 | :ref:`dataplane`) such as *fib_entry_t* and *load_balance_t*. These come |
| 170 | from the main heap because they are not protocol specific |
| 171 | (i.e. they are used to represent either IPv4, IPv6 or MPLS |
| 172 | entries). |
| 173 | |
| 174 | With 1M prefixes allocated the memory usage is: |
| 175 | |
| 176 | .. code-block:: console |
| 177 | |
| 178 | vpp# sh fib mem |
| 179 | FIB memory |
| 180 | Tables: |
| 181 | SAFI Number Bytes |
| 182 | IPv4 unicast 1 33619968 |
| 183 | IPv6 unicast 2 118502784 |
| 184 | MPLS 0 0 |
| 185 | IPv4 multicast 1 1175 |
| 186 | IPv6 multicast 1 525312 |
| 187 | Nodes: |
| 188 | Name Size in-use /allocated totals |
| 189 | Entry 72 1048589/ 1048589 75498408/75498408 |
| 190 | Entry Source 40 1048589/ 1048589 41943560/41943560 |
| 191 | Entry Path-Extensions 76 0 / 0 0/0 |
| 192 | multicast-Entry 192 6 / 6 1152/1152 |
| 193 | Path-list 40 18 / 18 720/720 |
| 194 | uRPF-list 16 14 / 14 224/224 |
| 195 | Path 72 22 / 22 1584/1584 |
| 196 | Node-list elements 20 1048602/ 1048602 20972040/20972040 |
| 197 | Node-list heads 8 24 / 24 192/192 |
| 198 | |
| 199 | and with 2M |
| 200 | |
| 201 | .. code-block:: console |
| 202 | |
| 203 | vpp# sh fib mem |
| 204 | FIB memory |
| 205 | Tables: |
| 206 | SAFI Number Bytes |
| 207 | IPv4 unicast 1 33619968 |
| 208 | IPv6 unicast 2 252743040 |
| 209 | MPLS 0 0 |
| 210 | IPv4 multicast 1 1175 |
| 211 | IPv6 multicast 1 525312 |
| 212 | Nodes: |
| 213 | Name Size in-use /allocated totals |
| 214 | Entry 72 2097165/ 2097165 150995880/150995880 |
| 215 | Entry Source 40 2097165/ 2097165 83886600/83886600 |
| 216 | Entry Path-Extensions 76 0 / 0 0/0 |
| 217 | multicast-Entry 192 6 / 6 1152/1152 |
| 218 | Path-list 40 18 / 19 720/760 |
| 219 | uRPF-list 16 18 / 18 288/288 |
| 220 | Path 72 22 / 23 1584/1656 |
| 221 | Node-list elements 20 2097178/ 2097178 41943560/41943560 |
| 222 | Node-list heads 8 24 / 24 192/192 |
| 223 | |
| 224 | However, the situation is not a simple as that. All of the 1M prefixes |
| 225 | added above were reachable via the same next-hop, so the path-list |
| 226 | (and path) they use is shared. As prefixes are added that use |
| 227 | different (sets of) next-hops, the number of path-lists and paths |
| 228 | requires will increase. |
| 229 | |
| 230 | |
| 231 | Stats Heap |
Neale Ranns | dfd3954 | 2020-11-09 10:09:42 +0000 | [diff] [blame] | 232 | ^^^^^^^^^^ |
Neale Ranns | 05cac30 | 2019-05-28 11:09:40 +0000 | [diff] [blame] | 233 | |
| 234 | VPP collects statistics for each route. For each route VPP collects |
| 235 | byte and packet counters for packets sent to the prefix (i.e. the |
| 236 | route was matched in the data-plane) and packets sent via the prefix (i.e. the |
| 237 | matching prefix is reachable through it - like a BGP peer). This |
| 238 | requires 4 counters per route in the stats segment. |
| 239 | |
| 240 | Below shows the size of the stats segment with 1M, 2M and 4M routes. |
| 241 | |
| 242 | .. code-block:: console |
| 243 | |
| 244 | total: 1023.99M, used: 127.89M, free: 896.10M, trimmable: 830.94M |
| 245 | total: 1023.99M, used: 234.14M, free: 789.85M, trimmable: 668.15M |
| 246 | total: 1023.99M, used: 456.83M, free: 567.17M, trimmable: 388.91M |
| 247 | |