puzzle

40 ms·

391.
by puzzle 9y ago
Apple took a long time to get this cloud thing right and had to play catch-up. No wonder they flew a helicopter over Google's nearby data center when they were working on their first in NC.
392.
by puzzle 9y ago
A hefty box is a SPOF, so that's an apple to oranges comparison. How do you upgrade that machine to a newer Postgres version?
393.
by puzzle 9y ago
You can make educated guesses about which range the compute/storage ratio is in if you look at e.g. public information from talks like http://www.pdsw.org/pdsw-discs17/slides/PDSW-DISCS-Google-Ke... (it has o
394.
by puzzle 9y ago
It's not outrageously expensive if you have hit the limits of of Postgres or MySQL or need multi-region replication, but yes, for small amounts of data there's too much overhead from all the redundancy. Using it for Vault makes mo
395.
by puzzle 9y ago
Apple definitely does not use GCP for 100% of anything, but it could be done. A migration like that couldn't happen overnight anyway, as I'm sure you know. Google hit the first 1EB of raw disk space a long time ago. Or you could
396.
by puzzle 9y ago
Making it available outside would require hiring more people and doing more work that they don't need right now. For a while, it's very likely that Google will be the main user of these, so there's still plenty of incentive f
397.
by puzzle 9y ago
Yeah, I don't know what kind of leaders he was thinking of when he said the next generation would come up in two years. At least at the very top, it might take decades.
398.
by puzzle 9y ago
Interesting. Google treated remote access from China or employees based in China differently even before Aurora: http://fortune.com/2011/04/15/inside-googles-china-misfortun... Maybe in ten years or more some
399.
by puzzle 9y ago
Well, not public, but eventually accessible by those holding the guns and running the prisons. Which is still not great.
400.
by puzzle 9y ago
Is there going to be an updated paper on performance per Watt, now that TPUv2 is public and V100 has been preannounced on the Google blog?
401.
by puzzle 9y ago
Google probably got back a lot of the engineering costs before it even rented out the first TPU, simply by virtue of running its own workloads, without having to buy tons of CPUs or GPUs. They're also very, very good at reducing comput
402.
by puzzle 9y ago
Yes, T-Mobile has been doing it for four years now. See slides here: https://conference.apnic.net/data/37/464xlat-apricot-2014_13...
403.
by puzzle 9y ago
I'm surprised nobody has brought up the scrollbars yet. The original post mentions the bespoke UI, but only faults it for using MDI. The non-native scrollbars were smart and nifty, but, at the end of the day, they were jarring to use.
404.
by puzzle 9y ago
Wave was in Sydney, the only Google engineering office in Australia. I walked through their space before they had even announced the project internally. All that almost everyone in the company knew was that Lars and co. were busy on somethi
405.
by puzzle 9y ago
What about the OT code and the comments?
406.
by puzzle 9y ago
If you run them side by side, you will often find that Google Maps does not try as hard as Waze to save seconds. It prefers routes that are easier to learn, as you yourself like them, with fewer turns (and thus fewer interruptions). It cate
407.
by puzzle 9y ago
Is Waze's traffic really updated more frequently? Depending on the area, Google Maps can be down to roughly once per minute. (Open a view at a busy time of the day and figure how often the traffic data refreshes.) Perhaps it's a m
408.
by puzzle 9y ago
"You know you have a large storage system when you get paged at 1 AM because you only have a few petabytes of storage left." (2010) https://cloud.google.com/files/storage_architecture_and_chal... No, they&#x
409.
by puzzle 9y ago
Why are proprietary drivers a blocker? As long as you expose the same gRPC interface, your customers don't need to know what happens behind the scenes. You could have an FPGA or a Beowulf cluster of Raspberry Pis hiding.
410.
by puzzle 9y ago
This is really Tensorflow as a service. You get an IP address and a port you send gRPC requests to: https://github.com/tensorflow/tpu/blob/master/tools/diagnost... Presumably, there's a whole s
411.
by puzzle 9y ago
Where's this TOS you are speaking of?
412.
by puzzle 9y ago
No. New clients use vector data, but even the web version will go back to bitmap tiles if your browser does not support WebGL, the GPU is slow or busy, etc. There was a whole MapReduce (which Jeff Dean used for years in public talks to expl
413.
by puzzle 9y ago
That's cool. Thanks! I'd still argue that you want to provision N+2 capacity, so you can withstand a planned event and an unplanned one at the same time, without having to go through manual tweaking. I'd leave that for truly
414.
by puzzle 9y ago
Even that might be too rigid. A better way is to have a time-based setting. For x < N seconds, ignore the missing node. N could be 600 or 900. During that time, you have degraded performance, but the system is still serving. If you excee
415.
by puzzle 9y ago
You can book either Uber or even Lyft from Google Maps (assuming they are available in your location). Once the ride starts, you can follow progress from either Google Maps or the Uber app. You can still book and track the whole thing from
416.
by puzzle 9y ago
There is a telescope on the roof that tracks or is supposed to track the sun, though.
417.
by puzzle 9y ago
I can see how it's more likely on prem, but at my job, we run Kubernetes in production on AWS and most traffic is pod to pod, without NAT involved.
418.
by puzzle 9y ago
Are you referring to the service cluster IPs? Those are great for short lived or low volume connections. If you want to balance load over long lived connections or have high volume, you really want to know the addresses of all your backends
419.
by puzzle 9y ago
What's the destination? If it's the outside world, yes, you need NAT for state tracking and address rewriting, since the rest of the AWS infrastructure knows nothing about the pod CIDR (I guess you could set up a subnet for it and
420.
by puzzle 9y ago
That's inbound traffic coming from the outside world. You need NAT because the load balancer only knows about nodes, not individual pods (perhaps you can pull it off with e.g. ELBv2, but definitely not with v1). There's more iptab

More ›