| Charles Chan | 4d5ee19 | 2019-09-09 23:08:36 -0700 | [diff] [blame^] | 1 | Supported Hardware |
| 2 | ================== |
| 3 | Following is a list of hardware that people from the ONF community have tested over time in lab trials. |
| 4 | |
| 5 | Supported Switches |
| 6 | ------------------ |
| 7 | |
| 8 | 1G/10G models (with 40G uplinks) |
| 9 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 10 | - OCP Accepted™ EdgeCore AS5712-54X |
| 11 | - OCP Accepted™ EdgeCore AS5812-54X |
| 12 | - QCT QuantaMesh T3048-LY8 |
| 13 | - Delta AG7648 |
| 14 | - Inventec D6254 (verified by Inventec) |
| 15 | |
| 16 | 25G models (with 100G uplinks) |
| 17 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 18 | - QuantaMesh BMS T7032-IX1/IX1B (with 25G breakout cable) |
| 19 | - Inventec D7054Q28B (verified by Inventec) |
| 20 | |
| 21 | 40G models |
| 22 | ^^^^^^^^^^ |
| 23 | - OCP Accepted™ EdgeCore AS6712-32X |
| 24 | |
| 25 | 100G models |
| 26 | ^^^^^^^^^^^ |
| 27 | - OCP Accepted™ EdgeCore AS7712-32X |
| 28 | - QCT QuantaMesh BMS T7032-IX1/IX1B |
| 29 | - OCP Accepted™ Inventec D7032Q28B (verified by Inventec) |
| 30 | |
| 31 | Commercially supported models |
| 32 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 33 | Models listed here are not supported by ONF community. You should contact switch vendor for commercial support. |
| 34 | - OCP Accepted™ EdgeCore AS5912-54X |
| 35 | |
| 36 | |
| 37 | Supported Servers |
| 38 | ----------------- |
| 39 | ONOS can be run on any x86 servers, with the following notes on CPU and memory allocation **per ONOS instance**: |
| 40 | |
| 41 | - CPU (Recommended: **32 cores**) |
| 42 | There are not much requirement on CPU. |
| 43 | In practice, we avoid running other CPU intensive processes/containers/VMs on the same physical machine where ONOS runs. |
| 44 | |
| 45 | - Memory (Recommended: **65G RAM**) |
| 46 | The memory requirement highly depends on the deployment scale. |
| 47 | In general, we suggest that each ONOS instance should get **at least 16G RAM**. |
| 48 | For production deployment at scale, we will definitely need not only more memory but also some fine tuning on JVM garbage collection mechanism. |