a petabye of ssds is probably cheaper than a petabye of hdds when you account for rack costs, electricity costs, and maintenance.
(Justin)
Tech nerd from Sweden
a petabye of ssds is probably cheaper than a petabye of hdds when you account for rack costs, electricity costs, and maintenance.
Hard drive density has stagnated. There haven’t been any major technology breakthroughs since 750GB PMR drives came out in 2006. Most of the capacity improvements since then have come from minor materials improvements and stacking increasing amounts of platters per drive, which has reached its limit. The best drives we have, 24tb, have 10 platters, when drives in the 2000’s only had 1-4 platters.
Meanwhile, semiconductors have been releasing new manufacturing processes every few years and haven’t stopped.
Moore’s Law somewhat held for hard drives up until 2010, but since then it has only been growing at a quarter of the rate.
Right now there are only 24TB HDDs, with 28TB enterprise options available with SMR. The big breakthrough maybe coming next year is HAMR, which would allow for 30tb drives. Meanwhile, 60TB 2.5"/e3.s SSDs are now pretty common in the enterprise space, with some niche 100TB ssds also available in that form factor.
I think if HAMR doesn’t catch on fast enough, SSDs will start to outcompete HDDs on price per terabyte. We will likely see 16TB M.2 Ssds very soon. Street prices for m.2 drives are currently $45/TB compared to $14/TB for HDDs. Only a 3:1 advantage, or less than 4 years in Moore’s Law terms.
Many enterprise customers have already switched over to SSDs after considering speed, density, and power, so if HDDs don’t keep up on price, there won’t be any reason to choose them over SSDs.
sources: https://youtu.be/3l2lCsWr39A https://www.tomshardware.com/pc-components/hdds/seagates-mozaic-3-hamr-platform-targets-30tb-hdds-and-beyond
Openshift is also a good competitor product if you’re interested in containers.
Nixos’ weakness is definitely it’s documentation. There’s often configuration snippets you can copy and paste, though. If you go with NixOS, make sure to come back with questions, the community is very helpful.
r2modman works natively on Linux
Sure, but people probably don’t expect ram slots on a smart watch. Just a replaceable screen, housing, mobo, dials, and battery would do a lot.
Right, that’s probably true. Video encoding hardware and storage is incredibly cheap, but we get talks from netflix engineers where they’re talking about how they’re limited by dram bandwidth on their servers.
Some napkin math:
Youtube has ~7M average concurrent viewers.
https://streamscharts.com/overview?platform=youtube
A 1080p av1 stream is roughly 2-3mbits, maybe 5mbits for 60fps. You could serve all of those users with 14tbps of bandwidth, then.
Stockholm peering pricing for 14tbps (rough ballpark at this scale tbf) over 43x 400gbit ports at a Stockholm Internet eXchange, would cost about 240k EUR/month, with a 25% volume discount.
https://www.netnod.se/ix/netnod-ix-pricing
For comparison, Mastodon’s monthly donations are about 30k EUR/month, and lemmy.world receives about 2k EUR/month.
Super rough calculations, but there’s probably enough of a base in the fediverse for us to take over like 5% of Youtube’s viewer base, funded through donations. Not as cheap as wikipedia, but still doable with a committed open-source community. Beyond that, and a netflix/spotify/nebula subscription model would allow to fund further market share.
It’s notable to see though that Nebula seems to have millions in monthly revenue, but only about 700k subscribers (aka barely 100k concurrent streams). However I believe the majority of their expenses are going towards their creators and towards marketing for future growth.
But yeah, I think network effect is a bigger barrier than cost here.
The fact that they can do expensive, on-the-fly video processing like this, and still make a profit, proves that video hosting costs are not an insurmountable barrier for the open-source internet. We need to make hardware accelerated peertube ubiquitous, and get creators to move over.
Intel a310 is the best $/perf transcoding card, but if P40 supports nvenc, it might work for both transcode and stable diffusion.
Bluesky has opt-in federation with the fediverse: https://techcrunch.com/2024/06/05/bluesky-and-mastodon-users-can-now-talk-to-each-other-with-bridgy-fed/
Unraid is bad at NAS and bad at docker. Go with a separate Nas and application server.
overclock.net lives on
It’s implied that it was a decision by management. If I had to guess, it’s related to money and/or the redundancy cause by the same parent company owning both Tom’s Hardware and Anandtech.
Kind of unfortunate, since I always thought Anandtech had the better articles, but I guess this also preserves Anandtech’s legacy in some ways.
Yeah this is definitely a brand merger in some ways.
I imagine it might be due to profitability, too. I think the rate of articles has slowed down in the last 5 years, and I think losing Ian Cutress’s analysis was also tough for their articles.
It feels like a lot of the hardware journalism these days has moved to YouTube, like Gamers Nexus, Hardware Unboxed, TechTechPotato, Moore’s Law Is Dead, etc.
I think Chips and Cheese seems to be the biggest site for detailed hardware analysis these days.
Don’t jump to the complex right away
It’s more complex to have 10 different ways to do the same thing. Like, just take a week to teach your ops team how to use Docker and Kubernetes, so everything can simplified to just one Kubernetes cluster instead of 20 bespoke EC2 instances.
Cloud Native development isn’t about making systems unnecessarily complex. It’s about simplifying tools down to common, scalable components, and reusing code as often as possible.
For example We use kubernetes to run code, because kubernetes is the only platform to run code that can be automated with simple HTTP apis. It is a common platform for computing, much simpler to use than the mess of EC2 instances, cron jobs, and shell scripts that the industry used to rely on. Of course, it is a higher level abstraction than programming everything yourself in Assembly, but that’s the point.
Is there a way for me to be “notified” if shell access of any form is gained by someone?
Falco is a very powerful tool for this.
It’s a safety issue
Yeah, that’s true. But if we consider “the Fediverse” to mean “internet forums that support Activity Pub”, then Mastodon is unforenot the biggest pleyer on the Fediverse in terms of user count, public impact, and funding.
Of course, Threads can go fuck themselves. Open source communities have no obligation to play megacorps’ games.
Sometes the prices go up, but they steadily go down over time.
This chart is really good for seeing storage prices
https://en.m.wikipedia.org/wiki/History_of_hard_disk_drives#/media/File%3AHistorical_cost_of_computer_memory_and_storage.svg