Cloud Computing: Let's kick this off

It'd be need to see the figures that go into what you saw for that (normalized to $1 for on-prem, of course), Massamune.
 
It'd be need to see the figures that go into what you saw for that (normalized to $1 for on-prem, of course), Massamune.
Can't share the exact data, but our model was pretty simple. We took the total all-in cost of our DC, software licenses, equipment and maintenance over a 5 year period, accounted for depreciation, and compared to a cloud environment that had the same # of VM's running. Even accounting for shutting off dev environment off hours etc, cloud was about $1.18 to $1 on prem for us when i last ran the numbers.
 
Can't share the exact data, but our model was pretty simple. We took the total all-in cost of our DC, software licenses, equipment and maintenance over a 5 year period, accounted for depreciation, and compared to a cloud environment that had the same # of VM's running. Even accounting for shutting off dev environment off hours etc, cloud was about $1.18 to $1 on prem for us when i last ran the numbers.

I am going to absolutely agree with you. We have drastically changed our approach to cloud over the last year, even since the initiative from my initial post here was out. We are going full on hybrid to a burst model and only using it for that and certain DR aspects. The cost of 1:1 forklifting got way too out of hand versus on-prem to make it a permanent viable option. We are still in the cloud space for sure, but not in the scale we initially envisioned. Don't let any cloud provider tell you they can save money. The only way they are running those numbers to make themselves look competitive is by claiming they can reduce FTE hours and hardware costs to skew it in their favor. Unless you are born in the cloud or your business/application/server model supports how one should utilize it, it isn't going to save you any money.

Just as a thumb in the air estimate we were figuring we could buy everything in our on-prem datacenter 2x a year if we forklifted completely to a cloud provider. And the man hours just managing costs of cloud use was at least 1-2 FTE's. It was definitely a good "failed" beginning to an experiment for us, as we learned a ton as a company on what to do and what to not do.
 
like dropbox running away from AWS?

The Epic Story of Dropbox’s Exodus From the Amazon Cloud Empire

sorry but I don't really buy the "enterprise" cloud... the cloud is great for companies that don't have the talent to run their own infrastructure

Workers who can maintain a HA cluster can be expensive. But I did calculations that showed that you can save at least 50% by doing things inhouse compared to cloud services or offsite virtualisation. The problem with offsite is that there is literally no guarantee on service levels. If the host of your instance is bogged down, all you can do is keep monitoring and try to migrate in hopes of finding a non congested host.
 
mikeblas

I think you're dead-on with your analysis of cloud computing, at least the compute aspect of it. Grimdaria is 100% correct, if you're using cloud compute 24/7 you'll go broke. However, if you have a) unpredictable workloads or b) predictable burst for a very short period of time, cloud compute makes 100% sense. Before I knew about cheap VPSes, I rented an AWS instance for a couple months, cost me $90/month (1 CPU, 1.7GB RAM, 10GB disk); needless to say that instance wasn't active for much longer.

4saken

With your hybrid, how do you make spun-up cloud instances addressable? Do you use scripts to update load balancers and DNS servers to make them aware of the cloud instances as they come online?
 
mikeblas

I think you're dead-on with your analysis of cloud computing, at least the compute aspect of it. Grimdaria is 100% correct, if you're using cloud compute 24/7 you'll go broke. However, if you have a) unpredictable workloads or b) predictable burst for a very short period of time, cloud compute makes 100% sense. Before I knew about cheap VPSes, I rented an AWS instance for a couple months, cost me $90/month (1 CPU, 1.7GB RAM, 10GB disk); needless to say that instance wasn't active for much longer.

4saken

With your hybrid, how do you make spun-up cloud instances addressable? Do you use scripts to update load balancers and DNS servers to make them aware of the cloud instances as they come online?

We are using an in house developed discovery service so we can burst out to multiple locations and register back to the discovery service as to their purpose. We also have an in house developed orchestration system and utilize chef for config mgmt as well. But at a very high level thats what we are doing.
 
4saken

Cool. :) Unfortunately, I don't have much use for cloud stuff so playing with it to any depth is out of the window - at least for the moment. :)
 
Its really no different than any other lease-buy comparison. Unless Amazon and MS and others become stupid, they will always price their service just slightly higher than the cost for an efficient data center operator to build/operate it themselves over a reasonable equipment replacement interval. Yes - I said "slightly higher".

It comes out slightly higher because (1) just counting the direct costs ignores the cost of money - both borrowing costs and opportunity cost. Also, if you are the owner, you have to buy ahead of demand (instead of on-demand) and have a huge cost of walking away if things don't work out. These intangible costs represent, From an economists perspective, Amazon's margin opportunity. In fact their opportunity is better than this because they have enough scale to command pricing power with the hardware manufacturers that very few of us get to enjoy. Life's not fair.

Neither model is "better". It is situational. If you are a business with really certain forecasts and low cost of capital (e.g., telco) then you'll land on the side of "own" almost every time. If you are a startup with a speculative business outlook and a high cost of capital then Cloud appears to be manna from heaven. Most of us land somewhere in between these extremes, which is why discussions like this one are lively and never end with a "correct" answer.
 
Cool. :) Unfortunately, I don't have much use for cloud stuff so playing with it to any depth is out of the window - at least for the moment. :)
That has me questioning your anecdote about not wanting to pay for an AWS instance. If you don't need it, it doesn't matter if it costs $9 per year or $90,000 per day. You don't need it.
 
That has me questioning your anecdote about not wanting to pay for an AWS instance. If you don't need it, it doesn't matter if it costs $9 per year or $90,000 per day. You don't need it.

At the time (i.e. 2007), I was using it to run an always-on VM. As I said, it was before I became aware of low-cost VPS instances - Amazon Web Services was the only brand I knew at the time (dedis were retardedly expensive; I knew nothing of Kimsufi either), so I jumped for that. As others have said, cloud compute is built and priced primarily for on/off utility use not 24/7 operation, yet 24/7 cloud might work out better depending on individual circumstances. For me, it didn't.
 
die-cut-stickers.png

I am going to have to send this to my ex-boss. He thinks the cloud is some kind of magical fairyland.
 
Back
Top