Might be our different experiences come from us at iProxy working with networking devices (and not being able to control their mobile net quality). And USB limits don't really bug us at all. So CPU stuff becomes a big deal. Plus, our server load is way beyond 20 devices.
Sorry for follow-up questions, I mean.. One of my main router servers handles traffic of 500+ USB devices(as entry router) and here are the stats:
This is handling more than 35k connections.
ss -o state established | wc -l
36931
CPU is 14 year old L5640 @ 2.27GHz.
It runs 3proxy and node.js based HTTP proxy relay.
vnstat live reports avg 27-30k p/s, with 200mbits rx and 200mbits tx.
And as we can see from load averages, server isn't practically sleeping but there is no significant load either.
And we've got scraper users on our proxies, pushing the load even higher, so yeah, stress testing is just part of our routine lol.
Question, why scraping is considered something big? If the client is smart and avoids loading static contents over proxied devices, it should not be a problem at all, no? My example I shared above is something of scraping too, static files includes video streams, images and of course other resources.
What do you consider as stress test, what loads you get, what is considered as big load?
Since the original poster is thinking about setting up a USB farm, I'd say your insights are super valuable because you're in the USB zone.
But yeah, my two cents about clock speed might also give something to think about, especially for those planning to push their proxies hard, like with scraping and other automations.
Can you offer some data on why? Like some examples of your loads, connections etc. I'm super curious at this point.
Edit: aaaahh wait, I forgot - you connect your phone proxies via backconnect right? That would explain why you struggle with loads, as because the double lined traffic though phone, if the phone has bad connection it will slow down everything and everyone, including your servers but thats not where high CPU would help with either, even top notch CPU would struggle with enough load placed if there is such bottleneck. no use for AMD EPYC either as it would go bananas anyways and all cores would go full load at one point.