Thanks Google AI Overview - UK Rail penalty fare speed-run

The constant wide eyed look of a heron is always bemusing, though they are likely just eyeing up food in the water

Reminder that being #1 or #2 on hacker news is about 1.5 page loads per second. If your side explodes when it hits HN, something has gone horribly wrong on your end

BGP handling bug causes widespread internet routing instability

On May 20th 2025 a BGP message was propagated that triggered some surprising (to many) behaviors with two major BGP implementations that are often used for carrying internet traffic.

In a new blog post, I will dissect what that message was, and my thoughts on how it happened:

https://blog.benjojo.co.uk/post/bgp-attr-40-junos-arista-session-reset-incident

WTF happened here I wonder?

Why did a considerable amount of people see a random /17 from Yahoo in their google results for a few days

Fancy mandarin duck

This "We simulated what it would be like if LLMs ran a vending machine biz" paper ends up with the same kind of insane emails that I sometimes get in my inbox, maybe they have emulated the true mind of the average vending machine operator after all

Found some Bees! Love the difference between the 300mm on the camera body I like, vs the 300mm on the body I was pretty meh on

Also, all of these pics are miracles, amazed the bees could hold on in the wind, for getting photos of them was a challenge as they were swishing around

Normal internet infra/computer posting will resume soon btw, I'm just having a lot of fun with the new lens on my Fuji and I have good reasons to go to parks at the moment

London paraqeet !

Also enjoy some crows, who are a real pain to take good pictures of because my camera does not like metering on a black bird in bright sunlight

Squirrel pics!

(the 300mm was a great call, lots of good photos without being in the danger zone of animals)

I wish there was a easy way to write a fedi bot that posts all of the slightly deep fried AliExpress marketing copy

Some lunch time bird pics on a very grey day

TIL there are PCIe Sound cards (such as the one pictured, a ASUS XONAR_AE) that are actually two devices glued together, a PCIe USB card, and a USB Sound card.

I cannot tell if this is incredibly stupid (this is obviously sub-optimal), or incredibly smart (this will not have any driver issues)

Thinking about that time the robot called google "smart reply/compose" was trying to tell other people that it was not a robot

This RIPE Atlas probe is soon going to be old enough to have mailing list arguments with, and boasts a industry leading "one nine" uptime over those years

(I'm pretty sure a lot of the time it spent down was when it ate a USB disk and I didn't notice for a few weeks)

DZ: Ben being upset at the state of windows

I boot windows 10 in a VM like once every week for basically just microsoft word (for contract review stuff / blog post edits) and excel (Excel still does better in some areas than gdocs)

but holy fuck, the situation just keeps getting worse. I rebooted this time to get:

A) A new splash screen full of crap I don't want like news tickers etc
B) A intro "Welcome to Microsoft 365 with Copilot" that would not go away

How on earth does the rest of the world deal with this, I'm not trying to get on the desktop linux user superiority complex or whatever, but it's such a user hostile environment that is just constantly trying to test my patience with the amount of crap it's going to throw directly into my face every time i'm just trying to use my goddamn computer to do something

Cant believe I even paid for this bloody license

We have to do something about the product managers

(do not reply with, "[Google|Open|Libre]Office can replace MS office", because I assure you it does not, especially in spreadsheets)

A special good afternoon to this crow who seems to be dead set on getting into this apartment

The faces of someone who now has to host the next Euro Vision, vs the faces of someone who does not

I would like to congratulate AS49450 for having the longest RIPE org-name, causing bgp.tools wanting to display it's name as:

Federal State Budget Institution NATIONAL MEDICAL RESEARCH CENTER FOR OBSTETRICS, GYNECOLOGY AND PERINATOLOGY named after academician V. I. Kulakov of the Ministry of Healthcare of the Russian Federation

(But it gets crushed down to just 56 chars to prevent a CSS explosion)

Rolls right off the tongue right?

fun fact, it is impossible to locate hard disk screws when you need them

Signs you have been AFK for a week doing conference stuff

good god Lisbon airport is very active on GitHub

This might win the award for most unsubtle google maps censorship in 2025

Introducing the "Tuscolo" Certificate Transparency logs, a new thing that @filippo and I am operating:

https://groups.google.com/a/chromium.org/g/ct-policy/c/KCzYEIIZSxg

For a while the certificate transparency ecosystem has been struggling to keep up with correctness (basically never roll back) and reliability (99% uptime) requirements, to address this there is a updated standard that @filippo worked on, Since he also worked on the first serious implementation of this new standard "sunlight", Port 179 LTD (me) and Geomys (Filippo and friends) are now running a log running this, ensuring there is "skin in the game" for this spec.

This log will be different from the traditional set up of CT logs that involve large MySQL or Cassandra clusters, and instead we just have a single reasonably low cost "bare metal" AMD machine. We expect this log to be around 50 times cheaper to operate than the established CT logs based in the "hyperscalers" (AWS/GCP/Azure/etc).

Tuscolo is currently receiving all Lets Encrypt certificates (as they are issued), hopefully there will be more CA's to come once we have full acceptance in the web browsers (we will likely be the first for a sunlight/new spec log to be accepted)

Today's fun debug adventure, on one of the bgp.tools remote IX collector boxes in Iraq, all DNS packets appear to be ACL'd now. See the difference in mtr's for port 53 vs 54

Not too much of a problem, as just flipping the switch on systemd-resolved to use DNS Over TLS "fixed" the problem. I guess systemd-resolved is good for something then!

Since I've been using my Nikon 300mm zoom leans for more than just reading the model numbers of cell sites recently, I've bought a Tamron 18-300mm F/3.5-6.3 Di III-A VC VXD for the camera body I actually like to use!

Walked around today after getting lunch for some first attempts, the AF is so much faster (to be fair I think my Nikon 300mm is slightly broken anyway)

CW: Ben stares directly at you jump-scare

Why when I launch the front facing camera on my phone does my face slowly distort for the first few seconds? what cursed things are going on in this device?

Oh cool, Victron's WebUI thing now has a mode that does not involve Websockets+VNC, but instead a WASM thing that talks MQTT... written in C++ & QT with WebGL?

Cool I guess, slightly cursed, but it works beautifully, no obvious jank that I would have expected from such a setup.

Much love to the embedded software engineer that tried out wasm I guess!

I think another Digital Realty (data center) Madrid site went dark at 14:00 UTC, based on DE-CIX Madrid Route Server prefix graphs

@benjojo I mean, they're supposed to be uninterruptible, not unexplodable

The uninterruptible power supply, a fine type of machine, apart from when they interrupt themselves by exploding

The internet routing table has lost about 5,000 IPv4+IPv6 prefixes after grid power Spain/Portugal was lost, and the number is still slowly going down as I assume batteries/generators run out.

I assume most prefixes are still announced while their final destinations are unreachable

(Graph is from a one of the sessions bgp.tools has with a T1)

Various Iberian Internet Exchanges have either seen at least 1/2 of traffic (the remaining traffic will be machine to machine, or international), or in some cases 1/4, GigaPix seems to have entirely gone (or the monitoring infra has died)

📉📈

I guess "spot the fault and then clearing" today, Spain and Portugal being the primary loser

One of the most "benidorm brit" possible things to complain about when a area of 60M people had their grid desync and collapse

switch:~# ethtool -m swp12
	Identifier                                : 0x03 (SFP)
	Extended identifier                       : 0x04 (GBIC/SFP defined by 2-wire interface ID)
	Connector                                 : 0x07 (LC)
[...]
	Length (SMF,km)                           : 0km
	Length (SMF)                              : 0m
	Length (50um)                             : 0m
	Length (62.5um)                           : 0m
	Length (Copper)                           : 10m
	Length (OM3)                              : 70m
	Laser wavelength                          : 850nm

Hmm, a the 850nm optic can do 10 meters over copper I see... Sure I guess

Trying to bring up a box without any BaseT ports, went into the spare optics box, picked something that looks vaguely 10G SFP+ stuck it in the workstation and uh, dmesg was not happy

System basically said

this optic is ass, interface terminated

no seriously, even reloading the driver wont bring that port back >:(

A Lot of Questions Already Answered By the Company Name

So, I'm not saying there's a coincidence here but... there is a little bit of one right?

root@blah:~# sensors
...

power_meter-acpi-0
Adapter: ACPI interface
power1:        4.29 MW (interval = 300.00 s)

Hmm yeah, Thanks HP, I am sure that the thing under my desk is 0.015% of the total grid consumption in the UK...

I mean weird place to put the databases, but sure I guess

Flying panda trying to get the last bit out of this lavender plant

Hmm, Last night (Euro time) there was a huge flood of BGP updates into the internet, looking deeper it appears that CERNET3 went down and up for a moment.

This would normally be a non issue, the BGP table sees a steady stream of updates per second of people mostly doing the same thing (your average full routing table has about 10 changes per second). Except CERNET3 is a bit weird. It has 4096 IPv6 /32's, and ASNs... CERNET3 calls this "slicing", and I've never really figured out what benifit this gives them to export 4096 prefixes, let alone 4096 ASNs.

Anyway, whatever happened hit all of the "slices". Causing a huge flood of updates to flood around the internet.

Not necessarily fatal (though this did cause a noticeable CPU spike, the only reason I noticed it), it does feel like poor behavior here to design a network where this impact is this visible from the outside!

A damp walk in a forest!

"liiiiick, oh ew"

I suppose I am glad that stripe has me covered so when I start making sixteen figures ( quadrillion? ) worth of revenue in a month

I bought this cheap ESP32 Dev board for something and IT'S SO TINY!!

tiny oled!!