whoknowswhat11

I’m doing this on b650d4u with success. 1.5k per machine or so but the clock speed is fantastic, good Remote Desktop performance as well. I can’t find reasonable cost server setups that can match this

I had this exact type issue for a charity account. A bit higher impact given team size but NO WAY to talk to a human!! Or if we did they couldn’t do anything. I imagine this has since changed, but it does leave a sour taste

The point about lack of projects on AWS I agree with if that’s what you are talking a about. That said AWS is absolutely a cloud leader. We use azure and AWS now and there continue to be reasons AWS does well despite cost and quirks

Yes, AWS being patient with both customer fault (common) issues and even out of scope for support tier level is crazy - it’s got to be a purposeful cost center expenditure. I like some things about google w project model but we spend on AWS (not too much).

No kidding, especially for programs that aren’t that big.

Wait what - people should be able to unplug other peoples cars? Why? The car will still be parked in the charging spot

As someone who initially paid for a blue check the idea was appealing to increase signal vs noise, I detested all the scam / spam comments and posts.

I didn’t understand the rush to remove checks for existing notable accounts - and then forcing checks on folks and claiming they were paying is absolutely pathetic.

I share this because I was part of target market and was open to paying. If I cancelled mine I’ve got to imagine others will too - be interesting to get some data

I also got a lot of mileage out of a homelab.

I experimented with proxmox and ESXi on the homelab front. Now ESXi is at the business.

Same thing with different server setups (Dell / HP / SuperMicro). Gives you a good feel for updates, KVM options etc.

Same thing with networking - I'm wasting time on Mikrotik and it's pretty fun.

It can be handy as a pool of additional resources. I had a very built out homelab server (with new SSD drives etc). Leadtimes got long for some Dell Server configs, we needed something 1-2 days, just grabbed a box from home. With ESXi once the real machine comes in its not that hard to move stuff over.

The big win is experimenting with low stakes. I find I'm learning 2-3x faster in that setup. I can try something, reinstall worst case, reboot, KVM, and literally plug in very easily (I have a hardware KVM setup in addition to the iDRAC type stuff).

I've always wondered why services are not counted down more often. Is there some sliver of customers who have access to the management console for example?

An increase in error rates - no biggie, any large system is going to have errors. But when 80%+ of customers loads in the region are impacted (cross availability zones for whatever good those do) - that counts as down doesn't it? Error rates in one AZ - degraded. Multi-AZ failures - down?

Anyone understand why these services go down for so long?

That's the part I find interesting.

My Amazon order history showed no orders, but now is showing my orders again - so stuff seems to be getting either fixed or intermittent outages.

For those confused, you scan in your badge, that get's you through the door.

Then you check the monitor to figure out what job you will be doing (ie, pack or rebin)

If your station is taken for some reason, you should be able to just move to another station and log in there - you don't have to use station assigned.

You will also have a monitor which shoes your rate.

Managers probably won't even know your name, and if you are day one employee showing up late, will have no clue who you are.

Some of the safety rules (because amazon get's hammered on this) - SUPER annoying - no listening to music - are you kidding? The most boring job possible, and to keep folks safe no music, not even one earbud in.

I wasn't totally clear about the difference between total apathy, which I'm sure facebook had for a small country with a language not in system vs "determined effort".

You'll appreciate that HR is done through an app. If you want to quit you can literally do that on the app, no need to talk to anyone. Same with apply for the job, accepting your offer, etc. It's mostly computerized.

That said, it's not THAT hard to get through process, and something of the steps are just filter. We don't know the posters work history / capacity / etc. We do know amazon probably cares very little that this person made it only 3 minutes?

I did hear that for older workers they have a work hardening process during peak to try to harden up workers a bit more gradually to be able to handle full shifts.

There is plenty of info online about becoming a worker at AWS if you are interested.

https://hiring.amazon.com/hiring-process#/ https://hiring.amazon.com/hiring-process/application-guide#/

This guy did an insanely detailed step by step of the process.

https://youtu.be/mfKctyhTi_Y?t=60

Playlist here: https://www.youtube.com/playlist?list=PL8Ja-dm83mgqj8B_cP4Gt...

He has videos about picking jobs, application, first day etc.

Note that from what I've heard Amazon really leans on the tech (ie, Amazon A to Z). You can resign / quit etc even from the phone app you download.

Probably for the best for this person they didn't make it more than 3 minutes at the job - its not that easy in the end from what I've heard - much easier WFH type jobs these days if you can get them.

That does seem like a somewhat obvious solution to built a relatively simple pooler / more efficient listener in. Anyone working on something like this? Merge with an existing solution?

This was exactly our my case. I did 10 servers going flat out calling into one (beefy at the time) MySQL. I was amazed at how far you could go with that. Then we had replication at the time built in, and so did read replicas. For free and somewhat hacky software the scaling was solid.

When I started playing with postgresql I had no clue why folks liked it. My memory may be bad, but we used UPSERT and replication. On Postgresql at least through 8 you could not upsert or stream replication.

Here is a bit more detail on this determined role they took to spread the speech from reuters.

"To this day, the company continues to rely heavily on users reporting hate speech in part because its systems struggle to interpret Burmese text.

Even now, Facebook doesn’t have a single employee in the country of some 50 million people. Instead, it monitors hate speech from abroad. This is mainly done through a secretive operation in Kuala Lumpur that’s outsourced to Accenture, the professional services firm, and codenamed “Project Honey Badger.”

According to people familiar with the matter, the project, which handles many Asian countries, hired its first two Burmese speakers, who were based in Manila, just three years ago. As of June, Honey Badger had about 60 people reviewing reports of hate speech and other content posted by Myanmar’s 18 million active Facebook users. Facebook itself in April had three full-time Burmese speakers at a separate monitoring operation at its international headquarters in Dublin, according to a former employee."

I used MySQL ages ago. It was pretty fast out of the box. I never needed to vacuum. You could get up and running with Upsert and connections without trouble. It did feel hacky, but don't discount that it mostly worked for a lot of use cases. It always reminded me of software developed with use cases in mind (vs theory). That's actually a complement (ie, full joins are rare, left joins more common - so MySQL was better at left joins).

Going to postgre - much pickier. I found it slower out of box. You couldn't just throw tons of connections at it (ie, connection buildup / teardown felt slower). I had issues initially with quoting and capitalization etc.

That was a long time ago. Now I enjoy postgres and haven't touched MySQL, but there is a real history where MySQL was the database you could get going with pretty easily (I was pre-Oracle buyout).

Thanks - that's very interesting!

Looks like you bought the cards yourself. That makes it unlikely that there was a chargeback on any card I think right unless you had a payment issue somewhere along the line?

Regardless, given the relatively large activity on the account I'm surprised the human or algo isn't able to figure this out.

Gift card + chargeback + international IP + small / quiet account = reasonable flag.

But you don't have chargeback or payment failure it doesn't sound like and account isn't that small - so super super interesting!

This is kind of interesting, there isn't anything immediately obvious here as a clear trigger (lots of these complaints when you get into it you can see maybe what triggered).

Are we sure it was the gift cards? Obviously gift cards are often used in various scams etc, but they should have a way to allow folks to prove they purchased them.

My own guess would be maybe gift cards and overseas IP access if there was some of that, the only thing I can think of (that would be a somewhat high flag in the algo's potentially).

A reversed payment in chain somewhere?

Never even heard of these guys or seen their posts. I'm on here a bit.

This happens for a few reasons.

1) Inactive or low activity account starts selling (instead of buying) high value / high risk items. Think GPU's, Macbooks, etc.

2) An old account has activity (including someone trying passwords from various breaches some of which might match an old salt). Because old account takeover and then fraudulent sales is a pretty big issue (old accounts often have 100% positive ratings) this seems to trigger the bots.

3) Accounts without KYC - normally you are asked to update.

I'm not sure how old accounts have to be to have problems. I'm from early 1999 which was three or four years after I think they got going.

The fraud problem was getting pretty ridiculous on ebay so kind of glad they are cracking down but it's got to be a pretty imperfect hammer and so no doubt lots of false positives.

I've gone to selling my old iphones / ipads etc back through apple. No where near the price you get elsewhere, but no hassle either from my experience.

> No one said you can overdraw your account

False: The qz article I'm responding to says this.

"When you pull cash from an ATM or swipe your card at the coffee shop, the machine needs to determine the precise time that the transaction occurs to, for example, prevent it from being over-drawn."

Please please explain how this "over-drawn" situation occurs because of a GPS satellite.

> but if the clock is off on them vs the server they will fail, I have worked on a lot of payment systems.

False, CC terminal clocks do not need to be GPS time accurate. When the terminal connects to authenticate, they can update their time. There can be issues with authentication if clock skew is bad, but this requires normal levels of time sync (ie, NTP without GPS will be fine).

no question that imessage, email parsers, etc that do third party untrusted network type interactions SHOULD be memory safe. But of course they are not, and apple in particular LARDS these formats down with a million features.

The claim was about ATM withdrawals and credit card swipes. Ie, without a GPS clock there would be a "clock sync" error in the ATM machine, and no cash could be drawn.

I continue to remain EXTREMELY skeptical of this claim. It makes almost no sense.

Please note that GPS is not the source of time, NIST is, and you can get NIST time in a variety of ways including over the internet if needed. And as I have been describing, many of these described errors are around systems with one source of truth (ie, not google spanner type globally distributed databases). So they make even less sense.

Note I'm familiar with clock skew issues and errors in authentication and other systems. These issues show up when errors are in minutes and seconds not microseconds.

If you need it you can do something around -12+.

Is this needed for the clock drift in the CC hand terminal that creates these "clock sync" errors that make it so the bank can't figure out my balance / avoid an overdraft?

There is digital radio time code, there is the internet and NTP where you can pull directly from the atomic clock standard, there is GPS. You can get your own atomic clock and get by for a while even IF you need some kind of perfect sync so the credit card reader doesn't have a "clock sync" error.

The claim here is apparently that relatively insane levels of percision are needed so that normal database transactions remain atomic. That I can withdraw cash from an ATM. That I can use my credit card.

I remain skeptical. Withdrawing from an ATM depends on ONE system knowing what your balance is. Your banks. While it might be nice to be accurate to the microsecond, it's not a requirement. To the second is going to be fine. Similarly, when you swipe a credit card, the auth process does a live inquiry (balance, fraud control etc). Again, I'm extremely skeptical of the claims here that we need microsecond SYNCED time for any of this.

The entire point of a cesium clock is that it does NOT need to remain linked all the time. That is literally why you would get one in your rack.

So your statement is totally false. In fact, the BENEFIT of a cesium clock is the incredible stability in the absence of a sync.

Exactly.

You don't need a GPS atomic clock, you can actually buy a rackmount one.

If that doesn't work you can use the internet to connect to NIST.

NIST is actually where GPS get's it's time, so you can just connect, using the internet, there directly.

There are systems that can do a fallback to a digital radio timecode. These are extreme cases that DO NOT apply to just an ATM cash draw.

The Bank has a record of your balance, regardless of GPS. For all live network authenticated reductions of this balance, GPS time is not relevant, the bank still has access to the balance, can still decrease it, can still report an overdraft.