Decommissioned servers for actual server use

[Littlejd97]

Hey guys. As we all know, we've seen Linus find old server hardware and try to game on it for cheap. Of course, gaming on an old server doesn't really make sense, as that is far from its intended use.

 

But what do you guys think about getting old server hardware and using as an actual server for production environments. This probably isn't the best forum, but we've been running our server infrastructure on AWS and our bill is going to be around $4000. If we spend $1,500 on old server hardware (such as 3 Dell R610 with dual 6-core 3.07GHz Xeon X5675 and 32GB of ram), and colocation, does anyone see an issue with that?

[AshleyAshes]

Well, a critical thing to consider is how much reliability do you need?  Amazon can migrate your instances to new hardware in the event that their hardware fails.  If your collocated hardware has a failure what will be Plan B and how much downtime is acceptable to you to implement Plan B?

[Littlejd97]

We will be using a docker swarm or kubernetes which will quickly be able to relaunch the containers on the remaining hardware assuming one died, in which case we’d then go buy another and slap it back in. 

 

The “Plan B” is to overprovision. 

 

We could also attach EC2 instances to our swarm in theory

[Erkel]

$4000 per month or per year?

 

Really come down to the economics of it.  AWS is about infrastructure as a service,  they make all the physical layer problems go away.  

 

If only $4000 per year, you are probably better to stay on Amazon.  If $4000 per month I would look at local hardware, but bear in mind that you have to maintain it and ensure you have good redundancy, backup and DR processes.

 

You do not want to get into the trap of going down the path of thinking that doing something your self is always going to be cheaper than paying someone who is an expert at it that does it at scale.  When are you are a company.  I have been in places like that.  Spending $2-$3 in labour / staff time to save a $1 does not make sense.

I run almost all my infrastructure in house using second hand or clearance sale hardware, and it would be uneconomical for us to move to Amazon.  But you need to be a reasonably large scale before it makes sense to take on the overhead of dealing with everything hardware and network wise, vs AWS providing it as a service.

[Razor Blade]

Depending on how far the colo building is, you might want to consider the travel time to it anytime you need to do maintenance or make any physical changes to your server.

[AbsoluteFool]

On 19.5.2018 at 12:25 AM, Littlejd97 said:

Hey guys. As we all know, we've seen Linus find old server hardware and try to game on it for cheap. Of course, gaming on an old server doesn't really make sense, as that is far from its intended use.

 

But what do you guys think about getting old server hardware and using as an actual server for production environments. This probably isn't the best forum, but we've been running our server infrastructure on AWS and our bill is going to be around $4000. If we spend $1,500 on old server hardware (such as 3 Dell R610 with dual 6-core 3.07GHz Xeon X5675 and 32GB of ram), and colocation, does anyone see an issue with that?

From my knowledge componentes for the R610 is very easy to come accrss. However i would think the R710 is a better option. Moving to owned infrastructure you should also have a quick-spinup if the system gets problems. There is many things to think about, but i also see that serval companies are pulling their services back down from the cloud because of the cost. 

 

The thing i normaly tell companies pulling down their services is.

1. Do you have personal with the infrastructure knowledge?
2. Do you have a backup plan if plan A fails?
3. Do you have UPS and nettwork security?
4. Have you checked the availability for spare parts for your choosen system?
5. What exactly are the savings running it localy instead of in the cloud? (E.g, Power, internet, cooling, the infrastructure itself)

When you know all of these you are on good way of knowing if it actually will be worth the time and effort.

[Littlejd97]

Thanks for the input. I was thinking R610 because we don’t really need much disk, just ram and CPU mostly currently. Any other reason for a R710? I don’t want to waste rack space. 

 

The AWS bill is going to be $4,000/month. Due to the nature of our product, we are actually giving everyone silo’d environments (separate database, application code instances, etc) using docker. 

 

1. Well, we are a team of like 5 people currently. I’d like to say I’m a pretty smart guy, however no I’ve never run a data center. But I do manage all of our AWS infrastructure, I’ve built

several PCs. I actually simulated the physical environment on my computer with virtualization including network configuration with pfsense. 

 

2. Backup plan is probably move back to AWS

 

3. Our network security probably isn’t the best, but I don’t think it’s too bad either. In a physical environment, I was thinking I’d use pfsense for the router with two vlans, one for the servers main nic, and another for the DRAC. The drac would have no internet access in or out. Only via VPN. And I would basically do load balanced port forwarding to all the servers in the other VLAN for network (80,443) ports. All other access would be over a VPN.  And we might grab a UPS

 

4. There are so many R610 and R410s on eBay for cheap, spare parts plan is to just dump the old one and swap in a new. 

 

5. We got a quote for collocation for $500/month with 50mbps internet and 21Us local in Austin (short drive time) 

[AbsoluteFool]

On 21.5.2018 at 5:27 PM, Littlejd97 said:

Thanks for the input. I was thinking R610 because we don’t really need much disk, just ram and CPU mostly currently. Any other reason for a R710? I don’t want to waste rack space. 

 

The AWS bill is going to be $4,000/month. Due to the nature of our product, we are actually giving everyone silo’d environments (separate database, application code instances, etc) using docker. 

 

1. Well, we are a team of like 5 people currently. I’d like to say I’m a pretty smart guy, however no I’ve never run a data center. But I do manage all of our AWS infrastructure, I’ve built

several PCs. I actually simulated the physical environment on my computer with virtualization including network configuration with pfsense. 

 

2. Backup plan is probably move back to AWS

 

3. Our network security probably isn’t the best, but I don’t think it’s too bad either. In a physical environment, I was thinking I’d use pfsense for the router with two vlans, one for the servers main nic, and another for the DRAC. The drac would have no internet access in or out. Only via VPN. And I would basically do load balanced port forwarding to all the servers in the other VLAN for network (80,443) ports. All other access would be over a VPN.  And we might grab a UPS

 

4. There are so many R610 and R410s on eBay for cheap, spare parts plan is to just dump the old one and swap in a new. 

 

5. We got a quote for collocation for $500/month with 50mbps internet and 21Us local in Austin (short drive time) 

It seems to me that you have dedication to make self hosted work. However i will warn you that server infrastructure is alot harder to deal with than AWS and consumer desktops.

 

I will also say that Pfsense is probably a good choice if you have knowledge with it. However it can be an idea to run a "dry test" before you hook up your product to it.

 

Other than that i'll just say good luck.