| 5 Jun 2025 |
 Mr. Defenestrator | Yea I saw '<5k' and thought typical prosumer stuff. Some ubiquiti home networks hit above that price range lmao | 13:39:49 |
| Mr. Defenestrator | Yea the mlxsw supported switches start at 25Gbe I think, its a more recent ASIC, so that also explains the price. | 13:40:28 |
 emily | well I'm open to being memed into buying something horrendously excessive but beyond three digits is a fairly hard sell :P | 13:42:06 |
| emily | I'd be up for Ubiquiti-level prices if it means mainline Linux / no blobs | 13:42:44 |
| Mr. Defenestrator | In reply to @emilazy:matrix.org
when you say "Linux control plane" though, are we talking mainline? because it seems these Celestica things are Broadcom, and my understanding was that the Broadcom switch stuff was basically totally useless on mainline without a pile of proprietary blobs. Yea a recent kernel at least. But it will need blobs, of course | 13:44:33 |
| Mr. Defenestrator | In reply to @raitobezarius:matrix.org
yeah and transreceivers are not in the same price range as well It's more about the fiber tbh. I have dozens of 40G transceivers that all require MPO. Bought for $4/ea. Any cable will cost $50 minimum. The CWDM optics are just hard to find. | 13:46:01 |
 raitobezarius | i have the opposite problem | 13:46:39 |
| raitobezarius | i have stupid amount of CWDM 100G optics almost for free | 13:46:46 |
| raitobezarius | but long range is not cheap | 13:47:11 |
 hexa | uhh, I think we tried one out in 2018 | 13:49:24 |
| hexa | sn2100 possibly? 🤔 | 13:49:55 |
| emily | right. I think both the cheapo Realtek stuff and mlxsw at least don't need userspace blobs | 13:50:03 |
| emily | AIUI from reading LKML posts, the Broadcom switch situation is that there's an "upstream" driver that basically just acts as a very dumb pipe between proprietary userspace tools and the switch ASIC | 13:50:24 |
| emily | so it's not quite what I'd consider a native Linux networking experience | 13:50:40 |
| hexa | or more like an sn2010 | 13:50:47 |
| hexa | so not exactly new, but still not worth abandoning 🙂 | 13:51:37 |
| hexa | we installed some debian to it and ran bird 1.6 probably | 13:51:50 |
| hexa | it was limited to 30k routes in hardware | 13:52:03 |
| Mr. Defenestrator | Yea that's kinda the case with broadcom ASICs. Mellanox switches have the best open source module imo. There may be some opportunity with MikroTik whatever in that price range, but I don't touch that stuff. | 14:10:10 |
| Mr. Defenestrator | You should buy not only the switch, but get a colo and start a NixOS based network shop ;) | 14:11:46 |
| Mr. Defenestrator | In reply to @hexa:lossy.network
it was limited to 30k routes in hardware 30k is plenty for that tier of switch IMO. If you have a Linux based BGP already, you could even do route filtering based off your flows and run a "digested" full table. I do that in production with Arista SRD | 14:16:03 |
| Mr. Defenestrator | If you really like extracting maximal hardware value. | 14:16:05 |
| emily | I thought people liked MiktroTik for running the actual vendor stuff but that mainline Linux wasn't really an option. could be wrong though | 15:27:44 |
 K900 | Usually the case yeah | 15:28:06 |
| emily | feel like I've fallen down enough of a rabbit hole already for now :) | 15:28:08 |
| K900 | Though Mikrotik hardware is also fairly commodity | 15:28:13 |
| emily | what kind of switch chips do they put in them? | 15:28:44 |
| emily | I thought it was custom stuff but maybe I'm wrong | 15:28:54 |
| K900 | Usually Marvell or Realtek stuff | 15:29:58 |
| emily | hm. | 15:30:34 |
