Exactly my experience. And not only about Databases.
At a previous job I was able to cut our infrastructure expenses by a lot (70%) by essentially have our infra being 1 load-balancer, 2 beefy application servers and 1 (+1 failover) PostgreSQL database. Each of the application server costs 300~ Euros p/m , 24 cores and 256GB of RAM. I have not experienced the same cost to perf ratio anywhere else.
What was your salary, and the salary of additional people to increase the bus factor to an acceptable level? There is a lot more than raw perf numbers that go into these calculations for any responsible business owner.
FWIW I'm in a similar situation, running a moderately sized site (10 servers, ~2gbps) on leaseweb hosts, costing ~$1,500/mo - every year or so I look at cloud offerings, and see that the bandwidth alone would be something like $40,000/mo.
I also have ~10 smaller hobby projects (<100 monthly active users) which could easily fit inside the smallest cloud services, but the smallest on offer seems to be $5/mo, so $50/mo in total. For $25/mo I can get a pair of bare metal servers which host all of them with failover and a ton of capacity to spare.
Then my actual day-job is at the opposite end of the scale, where we build our own datacenters...
I am very confused by the movement towards using the cloud for everything, because I work at all sorts of scales, and can't find a cloud host who works well for any of my use-cases :S
> I am very confused by the movement towards using the cloud for everything, because I work at all sorts of scales, and can't find a cloud host who works well for any of my use-cases
Just a couple off the top of my head:
1 - Incredibly large infrequently run / ad-hoc workloads
2 - Companies with such a horrendous internal work environment that they can't retain decent technical talent, where managed services can substitute to some degree.
Kimsufi - AFAIK they get their low prices by using second-hand hardware, but I've not found that to be an issue (hard drive failures are a little more common than the leaseweb servers, but "more common" is still only "one failure per server per 5 years" - and that's easily mitigated by setting up a pair of servers with failover)
This misnomer is frequently brought up, but untrue. Whether you’re on prem or cloud, the skills are transferable and staff needed regardless unless you’re a dev who just knows API calls for cloud services (my condolences). Cloud doesn’t mean you don’t need to know the infrastructure, it’s workload management. You still need to know how the underlying fundamentals operate, or you get burned (either through excessive cost, unexpected performance degradation, or data loss).
Absolutely. Its not like you will have one less person in the cloud. You'll still need the same amount of people (maybe more!), but they'll just be doing different things.
At my current role we looked to moving from Rackspace to AWS or self hosted. The cost difference between AWS and Self Hosted was easily 2x as much for the former. And Self Hosted was faster! Plus you get better control.
Sure there are challenges, but when there are issues, we can figure it out. We don't have to wait for support.
At a previous job I was able to cut our infrastructure expenses by a lot (70%) by essentially have our infra being 1 load-balancer, 2 beefy application servers and 1 (+1 failover) PostgreSQL database. Each of the application server costs 300~ Euros p/m , 24 cores and 256GB of RAM. I have not experienced the same cost to perf ratio anywhere else.