Hacker News Reader

Comment by napoleond | Hacker News Reader

People keep saying this, and I can see how it's easily true if we're talking about EC2 or one of Amazon's database services. But what's the hardware equivalent of S3 (the majority of the costs in this case) that's cheaper and "not much harder to manage" than S3? I feel like once you want to get past a couple TB of (redundant!) storage in colo/dedicated you're back to the land of sysops, no? I mean, I am decently comfortable in my ability to keep a server running, but I've never maintained a SAN (is that what I would need to replace S3?) and wouldn't trust myself to provide the level of resilience that Amazon is able to provide.

View on HN

GlusterFS on Supermicro Storage servers and loads of raid-grade harddrives, install once - scale forever. And it's not very hard to learn one additional software, and everyone knows that you need to setup a simple backup solution or system.

What I'm saying is that if you pay a cold $30k to AWS it is probably time to look at other solutions - or your COGS will keep growing out of hand.

I understand the greatness of AWS and Cloud systems when starting up, especially to keep CapEx down, but only to a certain point, you still have to manage systems (servers), security and storage/mount-points etc, so why not take the time to learn how to do it yourself and save a lot of money? Why not get an experienced DevOps instead of pure Developer?

And this company was already paying $300k to consultants, why not use a consultant or outside company to setup/manage their own hardware and service contracts as well? It would still be less than $30k and would still scale a lot better.

Have you tried running the equivalent of S3 using GlusterFS? Because I've seen mixed results with Gluster in practice and am not sure I'd feel comfortable going that route. Also, which one of their staff would have had to learn how to scale that? How many mistakes would they have made along the way and how much would that have cost them?

I think you're not understanding that you DONT have to eventually manage all of the systems on your own. S3 requires effectively 0 management. Dynamo requires effectively 0 management. The list goes on, and grows every quarter.

Sure, if you're just running EC2 nodes, there's some sysop requirement there. But if you run those on ElasticBeanstalk instead, you don't even have to manage that!

It's very hard for server and hardware-focused folks to 'get' the dramatic difference of running with AWS or GCE or other large cloud ecosystems. It's trivial for a, say, Java developer, to have a large, scalable system and have 0 access to each try to manage the actual machines and software that everything is running on.

You CANNOT replicate that in a private environment today, for anything less than a large infrastructure management cost.

Riak CS may be a simple answer for that. It even has S3-compatible API.

Reply on news.ycombinator.com