smueller1234

You're right and it did get used for that. Unfortunately that's nowhere large enough a use case to support the entire product/business/r&d.

Oh hilarious! The business logic for getting optimal routing for faxes was... quirky. Picking the right source country/line/carrier to get the best quality line while getting the lowest cost and balancing capacity. In the old days all held in a complex queue in a MySQL db (sigh). I recall getting paged out of bed a number of times as an escalation to get it all working again after we hit yet another lock contention issue in MySQL.

Good times!

Booking is the largest and thus a juicy target for both attackers and articles. Having worked there, I know that competitors had the same problems though.

Nice catch! That was indeed something we found about a decade ago when I worked there. I don't recall what the security folks tried to do about it.

Generally, though, more and more hotels shifted from faxes to API integrations via channel managers. But that's a whole extra step in sophistication.

And even though I've not worked there in many years, I can tell you with extreme certainty that booking doesn't do that. They're also (hopefully obviously!) PCI compliant.

IIRC when I worked at booking until 2017, they had already stopped the practice to share customer data via email for privacy+security reasons years ago.

Back in 2005 the researchers who worked on EUV lensing (mirrors, really) were very cynical and thought that it would never pan out. Coffee breaks were hilarious. It ultimately took a whole lot of time indeed.

My recollection from the Pierre Auger telescopes was 1-2 degrees uncertainty. TA is likely a similar order of magnitude. So matching a single event against candidate sources might not be the most reliable analysis. (Caveat: I didn't dig up papers to check my memory from 15 years ago.)

We've had many more events within an order of magnitude of these events' measured/reconstructed primary energy. The spectrum is a power law. While theoretically something completely different could be happening in the atmosphere of you add 2-10x the energy, there's really no indication thereof. Auger also has high energy event detected by multiple detection methods, not just the "standard" water-cherenkov or scintillator ground stations.

Several of them use multiple as-independent-as-possible detection methods. The most common are ground stations in a grid that measure high energy particles passing through, and then correlate them in time, followed by ultra fast (in terms of data acquisition - eg 20M samples/s) telescopes pointed at the atmosphere (not the sky), which can see the slight trace of fluorescence light left behind by the air shower from up to tens of kilometers away.

What's common to all of the detection methods I know of is the atmosphere. Back at Pierre Auger Observatory, we used a combination of LIDAR scans and weather balloons to constantly monitor it. The atmosphere is basically a calorimeter for these detectors.

The fluorescence detectors (FD) are generally considered the most direct measurement of both primary particle energy because the fluorescence mechanism is relatively simple (somebody is bound to criticize me for saying that), with a proportionality whose constant can be measured in the lab[1]. But there's still models and simulations that go into it. And these fluorescence telescopes can only be operated in moonless nights, so have a duty cycle of only about 10%.

Eg. Auger combines FD with surface detectors (SD) to use simultaneously measured events to calibrate the more indirect energy measurement of the SD and thus make use of their near 100% duty cycle.

[1] Other methods have been tried to measure this fluorescence yield. It's primarily done in the lab: https://arxiv.org/abs/1210.1319 For something completely different: my own masters thesis from a long time ago was an attempt to determine this yield indirectly because we know from detailed bottom up simulations that air shower shape is near universal given primary and energy. If we knew the event geometry (direction) well enough, we could use the ratio of Cherenkov to fluorescence light along the recorded track and fit our longitudinal shower size model using the fluorescence yield as the free parameter. In hindsight that was a lot of fun. I would enjoy that type of work a great deal more today, now that I feel that have less to prove. :)

Edit: formatting

And the largest cosmic ray observatory, Pierre Auger Observatory in Argentina clocks in at around 3000 km^2. https://en.wikipedia.org/wiki/Pierre_Auger_Observatory

A very different and really neat concept that hasn't become real yet is JEM EUSO, a telescope that would be mounted on a space station, pointed at Earth, would detect air showers via fluorescence like Auger's fluorescence telescopes do on the ground. This could theoretically cover a much larger area than traditional CR observatories. https://en.wikipedia.org/wiki/JEM-EUSO

To our best understanding, particles of such high energy do not originate in our own galaxy. We don't have a good understanding of what exact process might be putting this much energy into a single particle, but the commonly accepted acceleration mechanisms of, eg., supernovae don't reach anywhere near these energies.

GP also said "nearby source (in astrophysical terms)" which in this case is code for "could be over 100 million light years".

Yes, it's the same Spanner that's used internally. But obviously it's progressed hugely since 2013.

That's nonsense. Booking.com with it's subsidiaries was about 18000 or so people when I left in 2017. About 1800 of that were in engineering The vast majority of staff were working either on the supply or customer service side.

Not with GCS, you don't.

https://cloud.google.com/storage/docs/availability-durabilit...

Are you thinking of Persistent Disk (PD) and Replicated PD?

(I work on storage at Google.)

While we're at sharing neat vim configs, here's the one I was quite happy with at the time: I got tired of indent style arguments. I don't care. So I wrote a little heuristic to detect indent style, then configured vim with my preferences, and added this to .vimrc:

> map <F5> <Esc> :perl use Text::FindIndent;VIM::DoCommand($_) for Text::FindIndent->to_vim_commands(join "\n", $curbuf->Get(1..$curbuf->Count()));<CR>

Text::FindIndent is the aforementioned heuristic as a Perl module. Sorry for the mangled quote, original at https://metacpan.org/pod/Text::FindIndent

When I work on existing code, I hit F5 and work within the author's preferences.

Since there's a ton of Google employees here - reminder that Google does gift matching for this charity!

Some of us enjoy helping other people. First bit of advice I give to new or would-be managers is to learn indirect gratification. Most common failure mode for new engineering managers IMO is the struggle to feel productive given that impact is achieved chiefly indirectly.

For what is worth, I'm a fairly senior manager (a few hundred engineers) and I firmly believe and vocally point out the opposite. I'm sure the kind of self dealing you describe is not unheard of, but it's also pretty cynical to assume that as a default. If the culture in your work place has gone that far down the drain for real, maybe take some time to look for a new gig or reality check it with a manager you trust?

The entire story isn't about laptops. It's about the workstations that may be used to work on code locally (or also build it locally). These machines already aren't allowed outside of premises.

It is the case at Google and none of them are actually air gapped in the real meaning of "air gapped".

Yes, though not exclusively anymore.

There's a reason the profs love Landau/Lifshitz. The books have a certain elegance that, once you have a decent understanding of the subject already, is easy to appreciate. But as a first year theory student, that's just not what you need.

Went through the same experience you describe. But enjoyed them when I picked them up again a few years later.

You're right. It's not untested at all. It's just not instantaneous, unfortunately. :)

That's correct.

Long term holdbacks actually are way way harder to implement and maintain in a fast moving product than meets the eye. We had passionate and ultimately unfruitful conversations about how to do that when I was there in 2016.

The only thing we had at the time was continued tracking against the experiment groups after flipping that switch.

I think the problem underneath is quantitative fallacy: https://en.m.wikipedia.org/wiki/McNamara_fallacy

So one thing we did was for (a/b style) experiments to continue to collect data even after taking a decision to enable or disable[1]. No new sessions/devices would be added to the experiment, but those already exposed would still be reflected. So if everyone who decided to book because of an experiment ends up cancelling or consistently providing more negative feedback over the months that follow, our automation could still flag that.

Whatever we may say about the company and its practices, the engineering team working on the experimentation tooling had their heart in the right place. They worked extremely hard and with tremendous care on trying to make sure the data presented would cause the best possible decisions.

[1] Caveat: if an experiment was disabled fully (doesn't work or buggy) or enabled fully, obviously the experiment measurement is now no longer clean because folks in one variant will now see the other. But for trailing metrics like anything that affected past booking decisions, this type of analysis may not be perfect but still carries some meaning as a health check.

Edit: formatting fix

Ah, that would explain, thank you. No way they would have been able to make that change on hotels.

I worked at booking until late 2017. I was responsible for all their infrastructure and tooling development until the start of that year.

We all hated the urgency messaging internally. Hated it. We felt bad. Many of us were reasonably proud of our product and felt this tarnished our ability to deliver value to customers. We argued with the product org and the CPO on a regular basis.

The reality is that the urgency messaging drove significant incremental business that they felt they could not afford to pass up. The logic that the CPO applied at the time was some pretty sporty mental gymnastics: "we optimize for conversion (looker to booker) because that's a proxy to actually delivering value to the customer." The idea was that if a customer books and stays, it means they found what they were looking for. Being stuck in a perpetual, frustrating decision paralysis sucks after all. While I didn't agree with that framing, it's a touch less preposterous than it sounds. At the time, almost all bookings were fully refundable. So if you booked and stayed (you could have cancelled any time before the day at no cost, for most bookings), there's indeed something that went sort of right. Of course we also tried to measure CSAT in a number of ways (like net promoter scores) but those measurements are tricky at best.

In any case, one of the counter proposals, which I tried to co-champion unsuccessfully, was to add a checkbox to the account area to opt out of urgent messaging. The idea was that if you were annoyed enough by them to find that flag (or to be told where to find it) then surely it would also be better for the company to give you that and retain you as a customer? It was IIRC never tried. I was sad. It would've been a tricky to measure experiment, that would have had to be run for a long time. Loyalty measurement is hard.

Wow really? When did that change? It was 12% as recently (hah) as 2018.