puzzle

12 ms·

61.
by puzzle 8y ago
The MFM encoding used 2 raw bits for each bit of data to be stored. As such, people figured how to use the Blitter to do just that. I'm not sure if the OS did it; it was probably not the case in the 1.x days.
62.
by puzzle 8y ago
Yes, it read the whole track, starting from wherever the head was, then the floppy device handler figured where each logical block was, based on the magic sync word $4489 (which is not supposed to be output by the default MFM encoding) and
63.
by puzzle 8y ago
I meant crawling your own logs
64.
by puzzle 8y ago
At Google you don't get to read random log files, sometimes even from your own project. There are entire pipelines for raw and sanitized logs, with expiration dates and corresponding access controls. Unless you're on the security
65.
by puzzle 8y ago
Do we know custom how? It might be the same chips with different microcode or whatever GPUs' equivalent is. Is it a custom spin like Intel has been doing for years for Google and Amazon?
66.
by puzzle 8y ago
You underestimate how much Google tries to squeeze out of all the machines in its fleet. That includes old ones, sometimes to comical effect. A colleague at my current job told me about utilization targets at Amazon, where he used to work.
67.
by puzzle 8y ago
Maybe it wasn't AMD, but they already had a massive fleet of GPUs. It wasn't running Dremel, either. Or maybe they found a way to do that, too, I don't know, but there are already enough workloads at Google to keep GPUs well
68.
by puzzle 8y ago
You can't do time slicing, no, but you can definitely reduce time to first frame in many ways. If you don't do that, you need to provision even more hardware. Packing is also part of the capacity planning phases of a service. The
69.
by puzzle 8y ago
Yeah, if you know where to look, they left clues about using MI25 hardware. (I haven't been an employee for years, this all unfolded afterwards and, ironically, it is just one search away.) I'm sure they got bulk/promotional
70.
by puzzle 8y ago
https://bilbosrandomthoughts.blogspot.com/2008/07/dietrologi...
71.
by puzzle 8y ago
It's bad enough that there's a word in the language (in English, it translates along the lines of "behindology") to refer to Italians' pseudoscience of looking for the actual story that hides BEHIND the apparent rea
72.
by puzzle 8y ago
There is an even better incentive: make the parents accept liability for any medical costs caused by their unvaccinated children. Because their principles matter more than money, right?
73.
by puzzle 8y ago
"Psychedelic Warlord" Well, at least he didn't turn out to be Satoshi. That would be one of the few ways to top this.
74.
by puzzle 8y ago
You're not misremembering. It's just that timing matters. And GC always got less attention than compactions. My rule of thumb was: if I have a change that will shrink tablets by 10% through e.g. compression and I apply it wholesal
75.
by puzzle 8y ago
Compactions happen on the tablet server. They only write data. By definition, they increase disk usage, right away. Only the garbage collector, which needs global state and thus runs on the master, can actually delete files in Colossus and
76.
by puzzle 8y ago
Compactions do NOT reduce disk usage. On the contrary, they rewrite the data, so they generate new and presumably smaller SSTables. Which means more disk usage. It's the garbage collector on the master that removes files. You can combi
77.
by puzzle 8y ago
Two straightforward changes in Bigtable or Spanner are changing tablet compression settings or tuning compaction more aggressively. Edit: given the mention of a job that got stopped, perhaps they had a tool that manually forced compaction o
78.
by puzzle 8y ago
Technically, it's not satellite, but aerial imagery from multiple angles. Geometry is reconstructed with pose estimation and SfM (the open source Ceres solver originated from this and other work).
79.
by puzzle 8y ago
Google did it even before that and it's closer to a P2P model than Kraken (all Borg nodes participate, IIRC). By 2010, quite a few services built on top of Borg/MPM with the feature enabled had already undergone SAS 70 (later SOC
80.
by puzzle 8y ago
Given that e.g. AWS charges you for cross-region ECR image pulling, this can make a difference for scrappy companies that push large images on green (=multiple times a day, with lots of cache misses) to multiple regions. That's even if
81.
by puzzle 8y ago
I was at Google when this became the standard more than a decade ago ( https://www.usenix.org/sites/default/files/conference/protec... ). For jobs where you had tens or hundreds of replicas, i.e. for servi
82.
by puzzle 8y ago
How? Based on VAT? I can think of multiple approaches, but each of them is convoluted or deviating from existing norms in its own way. Every time there's an announcement like this, it's often surprisingly low on actual details.
83.
by puzzle 8y ago
The money is collected by Google Ireland. So they're going to charge Google Ireland? It's probably safe to assume they're not charging Google LLC in Delaware.
84.
by puzzle 8y ago
This is not very helpful or informative. In the case of Google, who are they going to tax? Google Ireland? Google France? How exactly?
85.
by puzzle 8y ago
Speaking of the founders' involvement, I seem to remember that, almost ten years ago, picking up to ten "best ever" places among your Google Hotpot ratings was Sergey's idea. I don't think that lasted more than a ye
86.
by puzzle 8y ago
Yours is the first public mention of MindMeld I've seen. There were a bunch of other related projects. Some of that experience eventually percolated to Tensorflow. Anyway, good luck with Dgraph! It looks very useful.
87.
by puzzle 8y ago
I meant tenants in that building. "The Internap expansion represents approximately 1.3MW of additional critical power in a severely constrained downtown NY market.” https://www.inap.com/press-release/internap-conti
88.
by puzzle 8y ago
Some of the largest colo tenants like Internap already left years ago: https://www.datacenterknowledge.com/archives/2013/07/29/inte... At Google's densities, they would never run a datacenter in Man
89.
by puzzle 8y ago
I think that 7K is just employees: "employs more than...". That is a deliberate distinction from "workers". I wouldn't be surprised if the count passed 10K once you considered contractors. They're mostly in non
90.
by puzzle 8y ago
What prevents a botnet from sharing that same session cookie?

More ›