This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revisionNext revision | Previous revision | ||
| network_stuff:cumulus [2022/05/26 14:32] – jotasandoku | network_stuff:cumulus [2024/09/22 16:47] (current) – [HAL (for forwardig)] jotasandoku | ||
|---|---|---|---|
| Line 46: | Line 46: | ||
| Current ASICs: | Current ASICs: | ||
| - Broadcom | - Broadcom | ||
| - | - Tomahawk | + | - [[https:// |
| - | - Trident | + | - Trident |
| - | - Jericho2 | + | - Jericho2 |
| - | - OVM Teralinks | + | - buffer-oriented, more programmable |
| - | - Marvell-prestera | + | - Intel: |
| - | | + | |
| - | - Inte: | + | |
| - [[http:// | - [[http:// | ||
| + | - P4 | ||
| - Tofino | - Tofino | ||
| - Mellanox | - Mellanox | ||
| + | - OVM Teralinks | ||
| + | - Marvell-prestera | ||
| + | - Cavium | ||
| + | |||
| - Netronome | - Netronome | ||
| Ownership: NVidia > Mellanox > Cumulus (so cumulus cannot use broadcom anymore)\ | Ownership: NVidia > Mellanox > Cumulus (so cumulus cannot use broadcom anymore)\ | ||
| Nvidia: cpu less important. GPU, TPU (for tensors (AI)) | Nvidia: cpu less important. GPU, TPU (for tensors (AI)) | ||
| + | |||
| + | * ASIC [[https:// | ||
| + | * Buffer space | ||
| + | * Forwarding table size | ||
| + | * On-chip table vs On-chip TCAM | ||
| + | * Forwarding features | ||
| + | * performance | ||
| + | * Programmable forwarding pipeline | ||
| + | * P4 | ||
| + | * recirculation for vxlan? | ||
| ---- | ---- | ||
| Line 73: | Line 86: | ||
| __DISAGGREGATION__: | __DISAGGREGATION__: | ||
| - | HAL (for forwardig): | + | HAL |
| + | (for forwarding): | ||
| * SAI (ms, set of apis to talk to the fwd hardware. compatible but one step more) | * SAI (ms, set of apis to talk to the fwd hardware. compatible but one step more) | ||
| * switchdev (part of the linux kernel but currently (2022) only Mellanox hardware supports it) | * switchdev (part of the linux kernel but currently (2022) only Mellanox hardware supports it) | ||