Ahh, I was only looking at raw hosting costs e.g. $29 × 12 × 3 = $1,044. Totally makes sense now, thanks!
Forgive me if this is asked and answered, but what software are you running to handle fallover? You’ve got your 3 colocated nodes… are you using an ELB? Nginx on one of the 3 nodes? What does that part look like? If one node dies… what happens? Are you just running 3 identical nodes with db + application on each?
I assume you are not kubering all the netes with these puppies.
1 Like
Automatic backups are sent to Amazon S3 regularly, so it is more of a “hot spare” kind of situation. Recovery means downloading the S3 backup, restore to the hot spare. Probably an hour of downtime.
Maybe the cost of S3 backup should be included too 
1 Like
It is tiny, on the order of $1 per month, if that.
How about a P2P cloud? We’re stable with subutai (and plan to create a Blueprint to deploy Discourse to p2p cloud environments soon).
While doing these tests on various DO droplets and comparing with a local NUC, I have managed to boot up a 4 vCPU CPU-Optimized droplet that was slower than a 4 vCPU Standard(starter) droplet:
sysbench cpu --cpu-max-prime=20000 run
3801 vs 3947sysbench cpu --cpu-max-prime=40000 --num-threads=4 run
5671 vs 6053CPUs
Xeon(R) CPU E5-2697A v4 @ 2.60GHz (Q1’16) vs Xeon(R) Gold 6140 CPU @ 2.30GHz (Q3’17)Price
$80/mo vs $40/mo
The result was very surprising especially given the CPU Optimized droplet is twice as expensive.
After some digging and inquiring DO tech support I found out that with CPU Optimized droplets you’re not actually paying for better “optimized” CPUs but for having the vCPUs threads dedicated to your droplet instead of sharing them with other droplets on the same server. This has been confirmed by DO support.
So… when comparing the cost of a scooter computer with virtual droplets from DigitalOcean, one should use the CPU Optimized prices.
1 Like
I had it in my head that you were using these as dev machines that you remote into somehow and was wondering how you do the client end… but upon closer inspection you mean server hosting things like people typically use the cloud for. That’d be cool though!
Gives me an idea though! My NAS that I hate running at home, but don’t move to the cloud because storage is expensive and the data isn’t important.
Is there any thing like endoffice on the west coast?
1 Like
So i’m not sure if I’m missing something here. But wouldn’t it be $29 per mini pc to colocate?
So instead it’s:
$29 per month * 3 PCs * 12 months * 3 years = $3132
Then tack on the machine cost + taxes/shipping/misc coming out to $4132.
Not sure what the discount for multiple machines is…
You forget the internet price (in my country ADSL with 20Mb down 1Mb up cost 10$/month (and fibre infrastructure isn’t ready yet)
1 Like
I forgot to mention there is a reasonable(ish) priced aftermarket KVM-over-IP, the Lantronix Spider SLS200.
It is USB powered which is nice, and everything runs in the browser, no software required though there is a Java runtime dependency for the browser console.
I found these used on eBay for $200 - $300 and they work fine to retrofit, but it’s a big price jump over the base box. You still won’t have power control, but EndOffice (for example) provides a managed power rail as part of their package already.
To be honest, if you really need a remote admin backdoor, that implies your setup and needs are sufficiently complex that you should be looking at entry level “real” servers instead which tend to have this built in.
Thanks for the pointers Jeff. I recently deployed a scooter at endoffice.com and have been very happy with the service at Endoffice and the scooter box itself.
Also, recently I’ve found this box: http://go.antsle.com/privatecloudserver21 which looks very interesting if you want to go containers all the way.
1 Like
I’ve been really happy with ASRock DeskMini 310 as a powerful desktop replacement.
i7-9700 with 65W TDP it’s an absolute beast compared to much noisier laptops (I’ve put a Noctua NH-L9i CPU cooler in there). It’s volume is just under 2L and it would be amazing as a colo server I’d imagine.
A beefier i9-9900 (non K version with 65W TDP) would offer 4 additional megs of L3 cache and hyperthreading compared to 12 MB L3 and no HT of 9700.
1 Like
I totally agree with you. Most sys admins today don’t care to know (study) about hardware and they run in herds to the cloud without a proper cost analysis… that is where CFOs with an engineering background should enter and say no, learn to do an actual cost projection and we’ll see.
Specially today that hardware is “modern” longer than ever due to the fact that Moore’s curve is long gone.
Anyone that says you need a cloud for mundane tasks (almost everything you do today) should ask themselves why doesn’t Google sell its data centers to Amazon or Azure and contracts back cloud computing services (They must be dumb).
1 Like