I want to start with - this is incredibly lean, and is super impressive to see such an extensive breakdown. I think that even the most savvy and frugal people would struggle to make large savings here.
A few thoughts/ideas still - have you asked Sentry [0] or Grafana (no link) for support? They might front you higher tiers of your current offerings for $0. AWS [1] also published a while back that they provide free credits for OSS projects, and in my experience they're generous and want to help. I don't know if they're still offering this though. My experience with AWS was that along with the credits came a healthy amount of Support and advice on ways to do things. These were often more expensive than the naive way of doing things in the name of redundancy. But given you're already all-in on AWS it seems like it might be no harm. If you were open to not being all-in on AWS, I've only got great things to say about pairing cloudfront offerings with AWS, but again yuo're talking minor savings for large amounts of work and complexity.
The obvious one is the EC2 instances for the actual compilation. Have you considered a hybrid model with always-on instances for even 30% of your linux capacity? A handwavy number we came up with at $PREV_PROJ was that it was cheaper to pre-allocate about 70% capacity on bare metal machines (with OVH and the likes), and burst for the last 30%.
[0] https://sentry.io/for/open-source/ [1] https://aws.amazon.com/blogs/opensource/aws-promotional-cred...
Thanks so much! I've not had too much success chasing down no-strings-attached stuff. Grafana and Sentry are pretty cheap compared to the AWS, so I didn't spend too much time...
Twice I've had folks from AWS approach me about helping, but it fizzled out. The best I've had is a pal who was a salesperson used to slip me a few "$100 new customer vouchers" every now and then I could redeem.
The pre-allocation means some kind of commitment: every time I've looked I've baulked at dropping many thousands of dollars in one go (and it screws up my terrible book-keeping :D ). I should probably revisit it. It's really nice to know I'm not paying for things I don't "need", but it's unlikely CE usage will drop _that much_ in the next year or two (say).
I'll take a look at the OSS stuff you linked: thank you!
> I'll take a look at the OSS stuff you linked: thank you!
Hope it fares out! I didn't message on your behalf because I didn't know if you had agreements already with them. But $20 or $70/mo saved on outgoings for a few emails certainly worth a shot :)
> The pre-allocation means some kind of commitment: every time I've looked I've baulked at dropping many thousands of dollars in one go (and it screws up my terrible book-keeping :D ). I should probably revisit it. It's really nice to know I'm not paying for things I don't "need", but it's unlikely CE usage will drop _that much_ in the next year or two (say).
There's quite a few ways to do this, and if you do manage to get some info from AWS they'd almost certainly be happy to help you out. Sticking with AWS, you're already using Spot Instances but you mentioned you have:
GPU Production: 2-4x g4dn.xlarge/g4dn.2xlarge with 1 on-demand base + spot instances
Well; I emailed the address at the AWS link and got: "The recipient's mailbox is full and can't accept messages now. Please try resending your message later, or contact the recipient directly." :)
Despite the bounce a nice person replied and is helping me apply! Thanks for bringing this up!! :)