2022-11-03

• The Elon takeover of Twitter has been pretty interesting, as a window into how people make sense of things going on around them. The big thing seems to be that people see around them the things they want to see. And they are, often, completely blind to things that would challenge their worldview. The result is that they construct their own reality, in many ways. To make it more concrete, the Elonification of Twitter has resulted in many strong opinions being shared on Twitter. You’ll see some people claiming that Elon is a genius. They’ll point out that charging power users for services they get value from is a good business idea. They’ll remark on how he is bringing a rapid pace of change to an organization known for its stability/long product cycles (some would say the product cycle is on the order of years). On the other hand, people will point out that Elon’s moves actually show that he is totally clueless. A rich guy who is incredibly out of touch. He bought a company for ~$40bn when the market would have valued it at ~$10bn. He is posting internal company chat logs on Twitter to humiliate and attack the previous owners/employees who enforced the contract that he signed. He’s negotiating product pricing changes in public Tweet exchanges with Stephen King. They might note that if Twitter itself is a clown car that fell into a gold mine, Musk is the perfect “clown born in a gold mine” to drive it. Whew, ok. What do we make of this? Well, as usual with particularly intense opinions (though, not always ofc), it’s fairly easy to point out that the people expressing the opinion are exhibiting motivated reasoning. The world looks this way to them because they want it to look this way. Seeing all of these opinions in the same timeline of Tweets is actually helpful in noticing that each of the opinions has some elements of truth to it, but misses other important information. The true parts gives the opinion some sense of legitimacy, and the strident language makes the ideas feel compelling. But the fact that so many of the views are contradictory highlights that these opinions fail to center other important information. And the choice about what information is excluded seems to neatly line up with which information challenges vs preserves a particular overarching narrative/worldview. The tweets are ultimately subservient to the reality that the tweeter wants to construct. With this in mind, we can expand outwards from this particular situation. We can notice the same phenomena with SJW types, who see systemic racism everywhere they look, or anti-semites, who see nefarious jewish stringpulling all around them. And we can expand it to include ourselves. When someone cuts us off in traffic, they must be an asshole, and not, say, a frantic parent trying to get their child to the hospital. When we fall in love, we see in our partner the perfect fit for us, someone who truly understands our every need and desire. We don’t see the story of love that we hold onto that requires “perfect fits” for us to feel excited. This gets at, for me, what is really beautiful and profound about Twitter. It shows us a window for looking at other people, which ends up also being a window for looking at ourselves.
  Weather report
  
• next 24, minor impacts to training:
  • Low 51 from night til 9am, starts to warm, gets to a High of 67
    
  • Partly cloudy
    
  • 0% chance precipitation
    
  • Wind from the south ~10mph. Could impact M4 quals for longer distance targets/unsupported positions
    
  • Light conditions
    • BMNT 0639
      
    • Sunrise 0743
      
    • Sunset 1751
      
    • EENT 1855
      
• next 48:
  • similar to Fri
    
  • low 60, high 74 in the afternoon
    
  • partly cloudy
    
  • 5% chance of rain, rain more expected Sat night
    NYC Marathon
    
• Main disruption will be associated with F Co during Soldiers travel back home post release. Verrazano Bridge will be closed until 4pm. Main body will likely be arriving back at home station around that time, with some admin tasks to knock out. But expect traffic if we release soldiers around 17/1800.
  
• Race runs north along the east side of manhattan
  
• !Pasted image 20221103171100.png
  
• Gets as high as 138th
  
• Harlem Armory is quite close, at 142nd and 5th ave
  
• Road closures will be nearby, likely to have minor impact on bus/tac vic transportation
  • Majority of route is unaffected. Only last 5 minutes of travel time may need to be updated
    
• Road closures IVO Harlem armory in effect from 8:00am to 6:00pm
  
• Mitigation is that we are coming from the North, race is coming up from the South. And we are exiting FDR prior to any actual closures. Also, if we are arriving later in the day, maybe traffic/road closures will be less extreme.
  
• A contingency in cases where Harlem River Drive is congested would be to have buses and TAC vics exit earlier and travel down St. Nicholas Dr to 142, then head east to the Armory
  
• Travel for Fox Co should not be affected. Planned route would not go through Manhattan in any case. Stays sufficiently far from the marathon.
  Notes
  
• Google File System
  • One of the big things that this system seems to do well is managing complexity. Most things attempt to be simple, do one job well, the job that is needed, and nothing else. Master nodes don’t keep track of too much state. They don’t have a complicated persistence system. Lots of data is kept in memory. Keeping things simple allows for the system to end up doing a lot of stuff without becoming overly complex. The pieces compose themselves together nicely.
    
  • The system is built from components that are expected to fail. Common design assumption in distributed systems.
    
  • Access patterns/system requirements seem to be pretty well understood at the outset of the design for GFS (e.g. overall disk space needed, average file size, how clients access files, etc.)
    
  • GFS is optimized for workloads that involve appending data to large files and reading that data sequentially. Other operations (e.g. random reads) are possible, but not expected to perform well.
    
  • Throughput is more important than latency as a metric
    
  • GFS is built to support data processing jobs, not online transaction processing
    
  • Chunk sizes in GFS are 64MB, much larger than typical page in file system. Linux page is 4KB, for example.
    
  • Interesting that GFS mapping from file namespace to chunk handle reveals that the tree metaphor of file spaces is just an abstraction. The data is not stored in some sort of tree. It’s just chunks distributed variously (hopefully to ensure load balancing). The data is flat. The filesystem makes it look stacked.
    
  • How does GFS only store file/chunk namespaces in memory? Doesn’t this risk important data loss in case of crash?
    • GFS doesn’t only store file/chunk namespace metadata in memory. It persists this info in a log on disk that is replicated.
      
  • When is chunk data rebalanced after a chunkserver failure?
    • When the failure is detected after a certain number of missed heartbeats

---

---

---