Conversation

Notices

1. Embed this notice
   Tito Swineflu (tito_swineflu@sfba.social)'s status on Tuesday, 21-May-2024 01:33:16 JST Tito Swineflu

   Somehow, the world has been persuaded to put their code inside little container boxes that are then put inside virtual machines that run on cloud hardware that costs 3x as much as owning your own hardware.

   • Embed this notice
     Tito Swineflu (tito_swineflu@sfba.social)'s status on Tuesday, 21-May-2024 01:33:14 JST Tito Swineflu

     in reply to

     • edgeworth

     @edgeworth in my experience, there still needs to be sysops / sres for both cases, and the frequency of failure, be it hardware or human error, seems the same.
     I recently tried to pitch some old coworkers who run a few WordPress clients on this. They spend 30k/yr on cloud rental to get what could be supported on 10k/yr of owned and hosted hardware. I told them to pay me 10k/yr to manage it and keep the 10k/yr in savings. There's no way doing that would take more than 100hrs/yr of my time. I think those numbers are reasonable and scale to the favor of the company buying their own hardware and paying a sysadmin.

   • Embed this notice
     Phil in SF (kingrat@sfba.social)'s status on Tuesday, 21-May-2024 01:33:14 JST Phil in SF

     in reply to

     • edgeworth

     @tito_swineflu @edgeworth

     It really depends on the systems one is attempting to deliver. Trying to do what I do with on prem stuff would be a nightmare.

   • Embed this notice
     edgeworth (edgeworth@sfba.social)'s status on Tuesday, 21-May-2024 01:33:15 JST edgeworth

     in reply to

     @tito_swineflu I always wonder whether such assertions include the cost of engineers who can manage running their own hardware plus the extra costs and challenges of redundancy

   • Embed this notice
     Tito Swineflu (tito_swineflu@sfba.social)'s status on Tuesday, 21-May-2024 01:37:53 JST Tito Swineflu

     in reply to

     • Phil in SF
     • edgeworth

     @kingrat @edgeworth I've also managed a system with 200-300 physical machines. The amount we would have had to pay to get the same data throughput on the cloud would have been crippling compared to the cost of the hardware and the three sysadmins we had.

   • Embed this notice
     Phil in SF (kingrat@sfba.social)'s status on Tuesday, 21-May-2024 01:49:58 JST Phil in SF

     in reply to

     • edgeworth

     @tito_swineflu @edgeworth I'm not just talking the amount of compute. I'm talking about the security needs, the ease of managing services, the ability to change what's used quickly, etc. My employer currently has about 30 people and our cloud bill is about 1 person's salary. I would need 3 to 5 more people to manage the services we use, were we to do them on prem or in our own datacenter cage. And those systems would sit idle a lot of the time, because what we do is very periodic.

   • Embed this notice
     Tito Swineflu (tito_swineflu@sfba.social)'s status on Tuesday, 21-May-2024 01:53:11 JST Tito Swineflu

     in reply to

     • Phil in SF
     • edgeworth

     @kingrat @edgeworth Periodic stuff is one of the cases for which I appreciate the cloud. Also slow storage. I put tons of stuff on S3 or other slow edge servers that are cheap. Compute and especially fast IO speed never seems economical on cloud. I also experience a serious data deletion or major outage failure on cloud about once every 2 years, which is on par with physical stuff.

   • Embed this notice
     edgeworth (edgeworth@sfba.social)'s status on Tuesday, 21-May-2024 02:10:56 JST edgeworth

     in reply to

     • Phil in SF

     @kingrat Yeah, good point regarding the work being done as a huge factor in the tradeoffs—there are definitely some cases where own-hardware is the better option, but it seems to me challenging to make blanket statements that are accurate.

     @tito_swineflu In my experience, getting location redundancy with decent security is the more challenging aspect of managing one's own hardware.

     In AWS I can easily spin up multiple datacenters in a region, and relatively easily start up additional redundancy in another region altogether without needing to do all the contract review and negotiation that would be necessary for doing the same with my own hardware.