Is there really that much room for improvement by changing packet size? You're avoiding a little bit of overhead for the 40 byte packet header and you're also increasing the risk of that packet failing CRC and getting discarded which is a lot more expensive.
There is much improvement to be gained by changing packet size from the current de-facto limit of 1500 byte MTU, but requiring 32-bits is likely unnecessary as just increasing it to the current 16-bit limit of 65535 bytes would be adequate for any reasonably designed protocol.
Given current compute, you would be hard-pressed to control plane more than ~100 million packets per second per core which is ~1 Tb/core-s at a 1500 byte MTU. Increasing this to just 65535 is a ~40x increase to ~40 Tb/core-s (~6.5 TB/core-s). Clocking at 5 Ghz and issuing 2 cache-line sized stores per clock you can still only issue ~640 GB/core-s (~4 Tb/core-s) of memory traffic from a core (assuming you do not get memory bandwidth limited earlier than that). So, at a 1500 byte MTU you need a 1:4 data to control core ratio, but at 65535 you flip that to a 10:1 core ratio. This should be reasonable for basically anything for the foreseeable future.
The real problem is the packet arrival rate on today’s high-speed interfaces.
At 1.6Tb/sec, with 1500 byte packets, that leaves very little time for a router to make a forwarding decision and route each packet before the next arrives. It’s kind of wild we have the same max MTU on interfaces that speed as we had with 10Mbit Ethernet.
So far we’ve kept up with some very clever engineering, but I do wonder if it can continue indefinitely.
I mean over the standard 1500 byte packet, for sure. I'd assume that you'd probably find an optimal spot at 1 to 8Mb on super fast networks. Also networks at these speeds tend to have very low retry rates. It's the buffer sizes that start to get crazy though.
So it probably can't kill you with blunt force trauma but it could drop something toxic into your drink while you're in the other room. Which, you know, assumes some sort of hostile intent that a robot can't have on its own lol. So it can kill you but probably not accidentally.
If your adversary wants to murder you that badly, I doubt they're going to give up when they find out you don't have a Stringman wired up in your kitchen.
Yeah this is one of my biggest questions from a lot of the high profile departures during the Trump admin early days (both times probably). Sure, you save face a little maybe, and you register your protest, but then you are replaced with a yes man almost certainly, which is significantly worse.
But also, then, I know exactly how I would feel about continuing to work if my purpose is now to just be a roadblock. It must be pretty demoralizing and can you really ask that of someone else or be upset if they aren't up for that? I would certainly be tempted to jump ship for a chance to get paid better doing something I enjoy.
I get the sentiment but basically you say that people need to join corrupt fucked up organisations to make them better.
Like if you don’t join and improve things, some kind of psychopath will join in your stead and make things worse.
I don’t think that’s a helpful way to think about it.
You’re not required to improve things anywhere and you can leave a toxic (work)relationship if that’s what you need.
The problem is not that power corrupts; it’s more that power attracts corruptible people
I mean literally like if the person's password was "potato##!" the customer service person would read back "potato##!". They weren't reading the encrypted contents of a pw field.
their point is it could have been stored encrypted and then decrypted for the customer service person to read back to you. Access to plain text != storing it as plain text.
Some of the concerns around generics and why it took so long were for the users as well. One of the biggest draws to Go has always been that you get the performance of a compiled language and yet compile times are so low that it can feel like you're developing with an interpreted language. The design of generics needed to maintain the compile time advantage or else it wouldn't feel like Go any more.
Languages like CLU, Ada, Delphi, Standard ML, OCaml, D, were having Go like fast compilation times, with generics, some of them decades before Go was created on much weaker hardware.
Command and Conquer was almost certainly my first exposure to him. I guess it's time to go replay Red Alert 3 and enjoy those cutscenes all over again.
A regular yearly replay is in order for me, too. Already done with SC2... just SC1/WC3/DoW(2)/RA2(mental/flippedmissions/etc)/RA3(uprising)/HOMM3/etc/etc/etc. Ah well.
Yeah, I built mine in 3090 times and told myself I'd upgrade every 2-3 GPU generations. I do have 128G ram and I'm thankful for that but I'm definitely not on the same upgrade schedule any more.
I have 32GB and an aging 5800X3D and I have not had any issues running games yet. Honestly would love to know what games need more powerful RAM/CPU outside of stuff like Flight Simulator, or running at 160fps without framegen.
Honestly I have a steam deck and that has gotten me accustomed to not sweating the lower frame rates. That and the ridiculous number of great indie games that aren't heavy on the graphics, and I'm not too worried about upgrades as much as I am worried about replacements if/when the time comes.
reply