Comments by "Winnetou17" (@Winnetou17) on "Serverless: A Comprehensive Breakdown" video.

  1. Except he did the math wrong and forgot a 24 in the big seconds to month multiplications (I so hate how much he rushes) which would make the cost about $28k, when using the 500ms time. With, say, 100ms time, it would be $5k. I do wonder though, how would the serverless response times be. They would be bigger than 90ms, right ? Since there's spin-up time, well, at least for some of them. Is that spin-up time 10ms ? 100 ms ? 1 second ?
    3
  2. Here I place a formal request for Theo to STOP DOING MULTIPLE MULTIPLICATIONS AT ONCE. And in general to STOP RUSHING over delicate things/parts like that. The rather large and significant math mistake at the end would've been TRIVIALLY avoided by not rushing so much and doing the math in multiple steps. Like first stating how many requests are per minute, then per hour, per day then per month. Or at least something like 3600 seconds per hour, 720 hours per month to keep things still simple. I really mean it that it was really a rush, as what I requested above takes SECONDS extra to write. Can you NOT feel dumb for doing a big mistake because you wanted to save, say, 1 minute, out of a 74 minute video, kinda spoiling it at the end ? Also, Theo, please place the multiplication in the written part too. If you miss something, you AND US can see it there much more easily and not have to search in the calc app history. Not to mention that you switch things so fast there, it's very hard to keep up and spot the mistake right then (that is, without pausing), it's almost looking like you WANT to hide something (which I know you don't, I'm just stating how bad this is) After all, it was stated that serverless is more expensive and that the cost difference might not matter that much anyway, so the overall conclusion is still kind of there, but it's still so SOO frustrating to see (not the first time) such glaring mistakes that also might make someone less knowledgeable really question the video and conclusion. For something so easily avoided. STOP RUSHIIIIIIIII
    2
  3.  @t3dotgg  But you can do both (though I'm ok with just having the long videos, since I like understanding the context and nuances). That is, AFTER you do the initial video, with all the details, you can make a summary video, a more distilled take. There's things that can be taken out like the chat interaction and tangents and the time it takes to write and compute. And you can make corrections too. Now, if you have the time and mood for a summary video, that's another topic. But it is (at least theo-retically) possible and not pointless.
    2
  4. He used the average of 2625 req/s (aka half of the peak of 5250), but yeah, that still amounts to $28k, effectively 10 times more expensive.
    2
  5. Don't know if he switched to something else, but I think it still is Arc. If I'm not mistaken, there's also a new one called Zen that also has this feature baked in, and it's based on Firefox (Arc is based on Chromium)
    2
  6. It's easy, we don't have a life, so we can watch things like this when we eat, for example 😀
    1
  7. It's easy how he got there - 90ms mean response time, meaning you can do about 11 requests in a second. 10 if you round it. So, in order to have 0 extra latency, he just said that you can do 10 requests per second per core. It's true that in reality, that 90ms response time is more like 40ms CPU time and 50ms waiting for stuff to be loaded, like a database query response, or the SSD to return file data and so on. So a core could handle more than a request at a time, but it depends on the CPU time vs iowait time proportions just how much it can do. And after a point, it can do multiple at the same time, but there might be a bit of latency on all of them. Still, it's just easier to compute the way he did it, and also has some headroom, just in case.
    1
  8. I was thinking the same. If you have a base load, consolidate that on dedicated servers, but still have a way to use serverless for extra load and spikes. That would combine the best of both worlds. I wonder who's already doing that out there (surely there is somebody).
    1