Hacker News Reader

Comment by elevensies | Hacker News Reader

Back of the envelope calculation:

- assume $100/TB for cloud data transfer

- assume one employee full time equivalent to manage colo'd servers ($10,000/mo), plus $30/TB data transfer

The break even point for the colo'd setup from a networking perspective is:

    10,000 + 30X = 100X
    X = 10,000/70 = 142 (TB/month)

At 1MB per "request" I believe this works out to about 50 requests per second average to reach this traffic level.

Weaknesses of this model:

- Data transfer only. Depending on what else you're doing you could also save a lot on compute and storage.

- I don't know that much about how colocated data transfer would be priced ... i.e. do you need overprovision to guarantee availability, etc.

- one employee to handle servers to replicate the Amazon AWS experience ... could be highly variable depending on what AWS features you are using.

View on HN

This calculation is an all or nothing affair though. The reality is that you can avoid most AWS bandwidth charges by direct connecting to an internet exchange. It's still only worth the effort if you're looking at a 6 figure bandwidth bill, but it doesn't require new employees.

I guess what I'd like to know is: at what point is it worth seriously evaluating alternatives to just using cloud services and paying the sticker price -- which means evaluating the labor cost as well as the cost of the service, where this submission and many others I've seen will quantify the cost of the services and say that bare metal is so much cheaper while ignoring the cost of managing it.

The general break even point is circa 200k USD per month, assuming the company has spare capital, and the ability to commit to using the same gear for 3 years.

It's useful to keep in mind that in this type of scenario the goal is not to replace a cloud provider but to leverage capital to get better bang for the buck on select services. ex. using in-house compute farms and leveraging s3 for storage

50 rps is a very modest box you obviously will want at least 2 but at any rate the difference between that and AWS is not a full time employee it's more like 10h per month. Why would colo data transfer cost $30/TB :)?

The $30 number comes from the original article -- If I were to make the colo data transfer free and change nothing else, break even point is 100TB/mo.

For the employee cost I agree if you just have a couple boxes pumping out data your labor cost will be quite low. I'm imagining what else you might have happening on your back end using AWS services that you'd also have to replicate in your colocated installation to preserve the efficiency, which would take labor. For example, if you want some extra machines to do some load testing. Or if you need to store a bunch of data and want an S3-like system. Or if you want to wipe a machine and reinstall from an "image". The backup system, and backups testing. RDS, Aurora, DynamoDB, etc. Most of that can be done with a couple clicks on AWS.

it's $20 at DigitalOcean which is already heavily marked up :) compared to colo. The sweet spot for AWS is low traffic project that uses a very large number of AWS services or something with extremely spiky workload. It's a legend that you don't need ops people for AWS. We have a fairly complicated large scale project on AWS we have 4 full time ops people. If we were in colo we would need 5-6 instead of 4 and a bit of remote hands services.

Reply on news.ycombinator.com