I am recently studying hpc configs and the docs I follow most (amd,suse) recommends blocking thp by default which I was able to replicate performance impact myself as well, on the otherside standard hugepages brings lot of advantage, I do setup during kernel boot but didnt test properly, currently using 4GB. If you really wanna bypass page latency I recommend buying a cheap persistence memory, not all filesystems support yet , (ext4 and xfs ?) but with -o dax mount option you dont use pages, if data is temporary tmpfs with hugepages as far as I can see quite common usage but I also found kernel memblock argument by simulating non-existent persistent memory and mounting a filesystem with dax brings similar performance or even faster most of my cases yet still you wanna script each time boot mkfs/mount
Can you add some details about what are the most complex components in your system? Also what parts you implemented from scratch? I see you still use torch
Korean reactor reached 7 times hotter than sun, when I was reading it made me laugh that humans said fuck the dyson sphere and skipped a step on Kardashev scale yet still I find it ironic that this one built with military support.
Well quick google search tells me there is about 50 to 100 year fossil resources left, thats about a lifetime for a currently newborn or more optimistic next 2 generation
I understand your point and you are completely right, but on their perspective your experience from multiple aspects of codebase is still valuable contribution and making a difference isnt it.
"showing porn to people against their consent is a form of sexual abuse" is more descriptive then your quotes I believe, at least this kind of precautions adds an alarm/notification options between kids in social media and twisted adults
My gf not good with computers yet still she is on fedora since about 4 year but she is only using for browsing, she doesnt mind to use work computer (windows) when needed for excel etc... But if she needs one particular win app I would install back without question.
There are plenty of leader election/replication libraries, shouldnt be problem if you pick one and give a shot if you are not confident about your self implementation or just benchmark and see how it goes.
Also amd avx2 machines faster than intel one but the current benchmarks probably didnt have hardware on hand, googles highway numbers also intel based, I wanted to test also, i got machine but having laziest days :)
I dont see someone to have a reason refuse a reliable software solution which would speed up their existing application which could be done just by an software upgrade and reduce their bills at no cost, most benefical I think would be the tunning part for an average user. Its also compliment with many existing architectures today including cloud/container orchestration and the baremetal itself. But probably the key is still community/vendor support and standards , I dont know how it would affect maintainer life
Your list plus cellular aging, my other categories:
os: hpc/highly tuned single purpose oses/unikernel machines will become more dominant
filesystem: cxl on the horizon dax supported systems will become dominant then userspace+vendor shipped software will take over if they can catch up a common standard (intel already does good job)
fantasy:
hardware cooling: I believe we have a lot of room on having control on the heat, wanted to check alternative applications especially vacuum