Skip to content
Register

First full table from my basement router

Networking by LichunLars 23 replies 3.5K views
LichunLars

Member OP

10 #1

First full table from my basement. 935,472 prefixes. 10G switch was $45 on marketplace, server is an old Dell with 32GB RAM running BIRD 2.0.12.

Cable management is, uh, a work in progress. Photo attached (warning: NSFW for network engineers).

PING 1.1.1.1: 0.8ms avg
PING 8.8.8.8: 1.2ms avg
Route via Contabo (https://contabo.com) adds 12ms
Route via KnownHost adds 18ms

Next: actual upstream instead of tunnel. Baby steps.

1ms or I don't want it
mediaaustin

Member

Deliverability Nerd

#2

That cable photo gave me physical pain. For what it's worth, I've seen worse in email infrastructure closets, but those usually hide behind locked doors.

Your SPF record for the domain in your signature is misconfigured, by the way. V=spf1 with no mechanism after the IP range. You're going to hit blocklists if you send from that server.

Congratulations on the table though. Full routes from a basement is a genuine achievement.

SPF, DKIM, DMARC — holy trinity ✉️
PA

Member

#3

Grabs popcorn

Following

SamAlvi

Member

Self-Host Everything

1 #4
LichunLars said:
Next: actual upstream instead of tunnel. Baby steps.

You've got BIRD running already. Add a docker compose for your looking glass, maybe a reverse proxy for secure access. I run my entire stack — DNS, mail, monitoring — from two Time4VPS VMs and a closet server just like yours.

The cable mess is temporary motivation. You'll fix it when you add the second upstream and actually need to trace which fiber goes where. Ask me how I know.

my cloud. my rules. my 3AM alerts.
GuruDeutschlag

Member

#5

Ja the Achievement is impressive nein I cannot approve the CableDisaster but the Border gateway protocol configuration appears functional ja you should consider a Patch panel installation for the Fiber optic cables nein do not use Plastic cable ties on glass ja very risky for the BendRadius nein I have seen worse in a RackNerd Datacenter once

DA

Member

#6

What's the Dell model. CPU matters for BIRD at this scale.

LichunLars

Member OP

4 #7
daviddude3 said:
What's the Dell model. CPU matters for BIRD at this scale.

R720, dual E5-2680v2. 20 cores at 2.8GHz, BIRD sits around 15% during a full refresh. RAM is overkill but I had the DIMMs lying around from a decommissioned job.

The real bottleneck was the 10G NIC -- Intel X520, $22 used. Without SR-IOV I was losing packets during the initial table dump.

1ms or I don't want it
DA

Member

#8
LichunLars said:
R720, dual E5-2680v2

Solid. I've got a T620 running similar, BIRD 2.0.12 as well. Your CPU can handle multiple full tables if you ever want to multihome properly.

The X520 is a workhorse. Just watch the temperature in a basement -- mine thermal-throttled last summer until I added a floor fan. Not elegant, but effective.

SamAlvi

Member

Self-Host Everything

6 #9
LichunLars said:
Without SR-IOV I was losing packets

This is why I run Proxmox on my closet box and pass the whole NIC through to the BIRD VM. No hypervisor scheduling games. My compose stack lives on a separate VM with a cheap Realtek for management.

Your 0.8ms to Cloudflare is impressive for a tunnel. What provider are you using for the endpoint?

my cloud. my rules. my 3AM alerts.
LichunLars

Member OP

#10
SamAlvi said:
What provider are you using for the endpoint?

Mullvad exit in Frankfurt. I know, I know -- VPN for BGP is a crime against networking. It's temporary until I get a cross-connect to Netnod or at least a friendly eyeball network in Stockholm.

I've talked to Bahnhof and Arelion. The pricing for even a 1G peering port makes me want to keep the tunnel forever. I used https://www.peeringdb.com to check.

1ms or I don't want it

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft