zten

Markus Winand's write-up at https://use-the-index-luke.com/blog/2016-07-29/on-ubers-choi... really drives this point home. InnoDB has favorable characteristics for what they're trying to do.

I've recently gone through the eGPU enclosure drama after reading an article linked here. I'm using one of the 2019 refreshes of a MacBook Pro, with a Radeon Pro Vega 20. I bought a Radeon RX 5700 XT at launch this week, paired with a Razer Core X, like you've got. (I can go into more detail, but AMD dGPU + eGPU is a ridiculous battle of knowing which obscure config setting to add to drivers to get them to install [Google for EnableFalcon=true]).

rEFInd successfully gets Windows to find the Intel UHD integrated graphics, so that's good. That let me uninstall the Vega 20's drivers and disable it entirely.

Even then, I still don't boot with the enclosure attached - it seems to ignore its presence? Unplugging and plugging it back in triggers Windows to go find all of the hardware, and then I have to wait 30 seconds while things grind to a halt as devices install. I'm eagerly awaiting AMD Navi 10 driver support in OS X, which I suspect will come around since people have found the names in kernel extension strings earlier this year.

I don't think it's that big of a reach! I agree that it might be a bit unnecessary in this context, but in math classes it was always enormously valuable to have more examples than fewer ones. The pattern would become easier to identify. This helped out quite a bit if an example mixed in some other concept that I was quite a bit weaker at, because additional fully-worked examples would separate out all of the components and help me figure out what was blocking my full understanding of the problem.

> The thing with Left turn Option 2 is if you get proficient & comfortable with it you will have VERY few times car drivers yell at you. Because you're behaving the way they do.

If they see you / you don't screw up... otherwise, you get https://www.outsideonline.com/2392955/lets-talk-about-driver... written about you.

(edit: the article's situation is different, because it is describing a high speed rural road and not a city intersection.)

Mobile GPUs perform absolutely miserably; no surprise given the design constraints. Using your link as a comparison, its nearest neighbors are top-of-the-line parts from nine years ago, and budget parts from four years ago.

I don't think it's speed enforcement in most cases. It's usually people not wanting to deal with changing lanes to deal with the slowest traffic.

For a while I would come up across a large number of vehicles inexplicably stacked in the left lane, with a huge gap on the right lane. Younger me would zoom up the gap and inevitably find a semi truck and someone passing it very slowly, and the left lane would close ranks because they thought I was a dick. Older me just stays in the left lane.

edit: I'll move to the right when it's so obviously clear that I won't have to move back to the left. But it's really tedious to be weaving across the road with regularity to maintain a strict adherence to the keep right law.

You could pull it off by buying a Civic Type-R, which I think fits the theme.

For someone with no experience with SF and just looking at a map: a huge amount of the East Bay is closer time-wise to the busiest weekday BART station, Montgomery, than the western part of SF.

Even in the Sunset and the Richmond, 1 bedroom rents are still expensive. I poked around Craigslist recently and found plenty in the $2500 range. If you go across the bay, it's a little better. And, if you can deal with BART turning into a pumpkin at midnight, you might enjoy your work commute more.

Edit: this post assumes you take a job in SF. But, big companies are building more and more office campuses on the peninsula, away from reasonable public transit options. If I recall correctly, office rents in the city are actually on the rise, considerably, and it is driving larger employers out of the city under the guise of "putting our employees together to collaborate better."

Who's running model training 24/7 to justify reserving this instance or co-locating your own hardware? (Apologies in advance for not being very imaginative)

Their ImageNet timing fits within the bounds of a Spot Duration workload, so in the most optimistic scenario, you can subtract 70% from the price - assuming spot availability for this instance type. (Of course, there are many more model training exercises that don't even remotely fit inside 6 hours.)

Parallels is great. I like it so much that I bought the subscription, and I don't even use my Windows VM all that much.

However, sadly, third-party software that spins up VMs tends to avoid supporting Parallels. I recall Vagrant having a sub-par Parallels provider (compared to the paid VMware Fusion support), and even minikube went with VMware Fusion instead. https://github.com/kubernetes/minikube/issues/220

Are you trying to solve the problems caused by an insufficiently sized ops group? The other posts in this sub-thread imply devs are going to make a destructive mess of your whole system architecture, but I find myself blocked on more mundane things like being unable to fix deployments or builds, or toy around with a new Amazon service in dev because every new service requires new IAM roles and guess who exclusively owns that ability?

> Going westbound on Golden Gate st around Fillmore is one of the weirdest thing I've ever seen. You can see entire lines of double parked cars with no drivers inside (sometimes on both sides of the road!)

That one's probably because of the McDonalds.

1 cyclist was killed by a criminal driving recklessly beyond speed limit on JFK, and was near a T-shaped intersection and was riding in the parking/shoulder lane.

1 cyclist was killed in a SOMA intersection by a DUI operator through no fault of her own with legal right-of-way.

The law as written means you have to yield on a bicycle in exactly the same way you would a car. There’s no room for China-level tolerances, as you describe.

People may also be picturing different scenarios: NYC couriers and delivery people swerving around them inspires scorn, as do cars driving around you at 30mph in a crosswalk even if no contact will be made. Edit: but the former scenario should happen more often than the latter simply because the car is often too big to pull the same stunt without actually causing physical injury.

I love the picture of a 10 minute traffic meeting, though.

Oddly, this is a cyclist behavior that is very widely admonished as being pedestrian-hostile.

Ahh, ok, I didn't understand that part from your post. I haven't used Java on Windows in a long time!

Looking at an example of a library that does native binding: https://mvnrepository.com/artifact/com.github.fommil.netlib/...

It appears the artifact is a DLL, and not a JAR. Thus your original question stands.

Plus, you can strip down the Java bytecode even further using ProGuard, but convincing it to not delete code that's actually used is an error-prone dark art.

> How are .so, .dll files handled during Java packing?

Native libraries used by JNI can be shipped in the jar files and are read from the classpath, in addition to reading them from the ordinary linker path.

(edit: turns out I'm wrong about this on Windows)

> Urban Freeways are better at connecting neighborhoods than arterial streets are, because the freeway is off grade (either way above, or way below pedestrians), and therefore there is nothing present to 'divide' anything. Pedestrians and bicycles can freely move around anytime without any waiting for anything and with 100% safety at all times.

Can you show an example? 280 in San Francisco comes to mind. The division is the underpass or the overpass, and the frontage roads required to support it. Any big grade change and long distance between buildings suppresses non-car use. It's not limited to freeways, either -- you can achieve a similar effect with grade-separated railways.

Take a quick look at Google Maps and you'll see I-15 is a very short distance away. It's not like they decided to go full San Francisco and run major six-lane highway routes at surface level. The people using this sort of route are probably all locals. All of the through traffic is already on the freeway.

Freeways divide neighborhoods; we can't plop them down everywhere, and definitely not in this situation. It takes a massive amount of room to move vehicles at a rapid rate, and it discourages pedestrian-friendly, dense land use (then again, it looks like they already made that decision with a bunch of single-family house subdivisions). Notice nearby that the stuff you can access from I-15 is a bunch of big roads, malls, car dealerships, and an Air Force base. Nobody really likes living next to it, either: there's measurably worse air pollution and much higher noise.

Plus, as we see in any majorly dense area, the freeways turn into buffers for inadequate local traffic capacity as they very quickly turn into parking lots. Low volume exits onto slow surface streets with tons of intersections jams everything up.

Here's my guesses...

> Why isn't there a pedestrian bridge?

Signals that you don't really care about people walking or people with disabilities and want them to drive.

Ok, that was flippant, I realize, but the only time I've seen anyone use prefer overpasses and such is the Vegas strip and that's because it is outright hostile to walk next to or across.

> Why is there a major thoroughfare in front of a school?

Everybody is driven to school. Look at the land use around the school.

Gotcha. Just some confusion around what you meant by in-memory OLAP engine. It seems that many storage format libraries get integrated into other platforms that bring the other supporting features (and the other undesirable heavyweight things you didn't want) with them.

With a little bit more poking: Apache Calcite?

I would try Apache Spark. It's fairly easy to get it running locally, and does not require Hadoop. It provides the query engine.

If you're in a city that mostly moves people by bus, and has incredibly high construction costs, of course this is the solution they're going to chase. The bus service suffers with more competing vehicles on the road. People prefer rail and dedicated right-of-ways, but they also don't want their neighborhood destroyed while it's built. The transit advocacy groups end up hamstrung by these competing desires.

Uber and Lyft have given everyone a taste of the self-driving car dream (on-demand comfortable private transit) -- they just need to figure out how to get rid of the drivers. It's already producing political pressure to stall or end hugely expensive transit projects because of uncertainty about what a city might really need in 20-30 years.

Hopefully someone's smart enough to show that we'll all still be stuck in the same dumb traffic of our own making, regardless of a human or computer driver, because we really need to be solving for space and land use by minimizing the distance and frequency of vehicle trips instead, public transit included.

I've always suspected I have a problem like this but couldn't ever get a doctor to help.

Prior to treatment, did your sinuses open up completely during hard aerobic exercise and then feel like they slam shut as soon as you were done?

Only tangentially related to your post, but, two new traffic lights are being installed on the Wiggle. Haight is getting nine new lights, two of which are at Scott and Pierce. Maybe people will wiggle slightly differently to avoid the left turns at the lights.

This is a good article with sensible rationale for raising the speed limit on freeways. It starts to touch on why local safety advocacy groups push for lower speed limits, but it only really addresses speed in the context of other motorized traffic.

That is definitely a quote that only applies to racing. I agree that better fitness absolutely does make the easy stuff easier. But if you want to make going fast feel easy, you can't escape training for it.

I can see why you'd want to keep it non-Java and simple. I will admit that I am biased since I work with Spark all day, but I think you'll eventually want to write a query where you wish you had Spark SQL or could manipulate it as a Spark RDD.

Spark actually isn't too painful for this because you can run it locally in a single process. It is not exactly as elegant as the UNIX way you outline in your second example, but it isn't as horrifying as submitting a YARN job or spinning up a cluster.

In spark-shell:

    // Skip this conversion if you're just running one query. Do it if you're running many.
    spark.read.json(input paths).write.parquet("/target/path")  
    spark.read.parquet("/target/path").createOrReplaceTempView("logs")
    spark.sql("select count(distinct src) from logs where dport = 23").show()

The incremental conversion of your JSON data set to Parquet will be a little bit more annoying to write in Scala than the above example, but is very much doable. There is also a small amount of overhead with the first spark.read.parquet, but it's faster on a local data source than it is against something like S3.

Naming and formatting it differently is an easy way to get away from the REST dogma, though.