Hey HN,<p>I'll cut straight to the chase here, our AWS bill is creeping up faster than we can actually control our resources and I need some help. I work for compute intensive AI SaaS company and that had a pretty big spike in traffic last month so now our AWS costs are at least 3x. Currently we don't really have that manpower to manage the bill so I would appreciate any advice that you'll throw at me. (People to follow, tools, blogs, or just any sort of general tips)<p>What we considered:
-Spot instances, but that very short shutdown notice doesn't sound attractive to our workloads
-AWS savings plans, though we aren't keen on that long term commitment
-I've seen many people mention calling AWS support, so do we just bargain there to get better prices?
When I was in a similar position we switched to Digital Ocean (no affiliation) simply because of their predictable pricing model, and that worked very well for us. Linode would have worked the same.<p>Another service that I've used is using dedicated servers from OVH.<p>Of course all this is my personal opinion but the obscurity of AWS pricing (predict/calculate ec2 hours + predict/calculate network traffic costs + S3 storage class/tiers costs) is a major turnoff for me.<p>In some situations all that granularity could have costs benefits but in my experience it does tend to run and suddenly a we end up paying $600+/mo for a $20/mo instance because we hit some cap thresholds (true story)...<p>Good luck figuring this out!
[DISCLAIMER] I work for CAST AI<p>Hey there, I know clouds and their bills might be hectic.. Even though a lot of information in your case is lacking, here are a few tips for you to consider:
-Get to know the AWS instances, their types and find the ones that could do the the same job - but better. Here at CAST we benchmarked a lot of instances and select the best ones for you.
-Reconsider spot instances - as "advertised" they really do save up to 90% if you are lucky enough to bid low for an instance that is currently in not such a big demand.
-Are you a kubernetes user? If so, do you have HPA and VPA configured and running smoothly ?
-If you're an EKS / GKE /kOps user, Give us - CAST, a shot, I know it's a bold claim, but if you invest a minute to visit <a href="https://cast.ai" rel="nofollow">https://cast.ai</a> - the ROI can be massive. We've had companies in e-commerce, martech and development services save anywhere in the range from $50,000 to $1,700,000 per year with CAST AI.
To be completely truthful, some of them cut 40%+ on their bill with the information from our free report alone.
-And of course, go trough that AWS bill in detail, check and cross out what's a low hanging fruit. Also check whether some workloads do have to be running at night.
Disclaimer: I'm Co-Founder and CEO of an a16z backed company named <a href="http://vantage.sh/" rel="nofollow">http://vantage.sh/</a> -- Before this I used to work at both AWS and DigitalOcean so know this world well.<p>If you're up for it, you can connect your AWS account to Vantage. We automatically profile all of your AWS costs and provide cost recommendations for you based off of your account's actual usage. If you'd like, I can also personally hop on a call with you to walk through your AWS costs which you can book with me here: <a href="https://calendly.com/ben-vantage/30min" rel="nofollow">https://calendly.com/ben-vantage/30min</a><p>Depending on your level of AWS spend, we have a free tier that gives you up to $2,500 per month of costs tracked for free as well.<p>Lastly, we have a free website we host named the Cloud Cost Handbook which covers general concepts and best practices as it relates to cloud costs which may be helpful: <a href="https://handbook.vantage.sh/" rel="nofollow">https://handbook.vantage.sh/</a>
May be look into getting compute intensive Dedicated servers ? There are companies like hivelocity [0] that provide really good dedicated servers at a reasonable cost.<p>[0] <a href="https://www.hivelocity.net/dedicated-servers/" rel="nofollow">https://www.hivelocity.net/dedicated-servers/</a>
We use Hetzner for compute intense services. I think last time we checked a similar configuration was 10x more expensive on AWS (which we still use for other services)