Hacker News Reader

> You can't even log into the instance at all for that price.

What? could you please explain how would you manage the instance if you can't log in to it?

You can log into it, but not for that price. It'll count against your data.

For what price? $5? The bare server is either 25 or 183 cents cheaper, and the amount of bandwidth used by a month of ssh should be less than 1 cent.

Worry about the disk, worry about things that use gigabytes of bandwidth. But don't worry about it costing extra to log in.

Logging in uses bandwidth

I thought each instance came with 1GB of outgoing traffic per month?

There's a reason they are posting the profit margins they have been. AWS != cheap.

Why not just get two dedicated servers and have one be used as a backup?

For $60/mo you can get 4Tb space free, basically unlimited bandwidth and a permanent 4XL-equivalent instance that makes scaling up to that point extremely easy with no need to configure autoscaling, s3, inter-service communication, cloud formations or complicated fail-over strategies. Unlike AWS instances that die frequently, a dedicated server is much less likely to malfunction.

Assume you are talking about OVH or Hetzer? Those companies have their issues too.

Yeah, those two. There are some Latvian ones as well. What are some of the issues you've ran into with them? Genuinely curious.

I used Hetzner a few years ago, in the space of 6 weeks I had 3-4 times I couldn't login because the network was saturated under a "DDoS", I had random and sporadic network drops even when they said they weren't having problems.

Latency was an absolute disaster even typing characters via SSH was horribly slow.

The management platform was hilariously bad (they where still sending me notification emails about 9mths after I cancelled).

We went back to linode for everything after that, for our needs it hits the sweet spot between cost and reliability (back then DO was new and had it's own issues, largely due to a flaky network).

Now I'd possibly consider DO for compute or job servers but I think I'd still use Linode for anything that wasn't running on my own hardware, I've been a customer with them for something like 6 years and they've never let me down.

Makes sense.

I've used Hetzner for 2 years now for a lot for personal projects, and I do get slowish SSH sometimes, but that's what Mosh is for. Also, it's much better for European customers than for US ones.

They are also the only service I can afford for personal projects that requires a lot of RAM. Basically, any machine learning with a ton of features.

Here's a test with latency:

    URL tested:	http://rerecommender.com
    Test performed from:	New York, NY
    Test performed at:	2015-12-18 04:36:44 (GMT +00:00)
    Resolved As:	188.40.128.87
    Status:	OK
    Response Time:	0.306 sec
    DNS:	0.075 sec
    Connect:	0.115 sec
    Redirect:	0.000 sec
    First byte:	0.116 sec
    Last byte:	0.001 sec
    Size:	7688 bytes

Are there other well known dedicated hosters beside OVH in North America?

reliablesite.net, 100tb.com (bit expensive, unless you need the bandwidth) are two good ones.

OVH puts a ban on business from mainland China. It`s kind of weird. You maybe hate those harassment, and you want to shame mainland Chinese(because almost all other countries, even Hong Kong SAR and Taiwan are accepted). But a) this is business, you should not be that emotional, there are good people in China too. Doing business in this way may only humiliate yourselves. b) you can not just ignore China market, it`s a market no company should overlook.

Could it be because of a lot of spam originates in China?

So some Chinese residents buy the servers to spam others and then they have to clear blacklists for their server IPs?

No offence, fighting spam in that way is either an IQ problem or an EQ problem. Banning mainland China business completely for couple of years to clear blacklists? Brilliant. Actually you can buy any servers from EC2, linode, DO freely from China anytime. How can they deal with this issue?

Totally unrelated to this post, but THANK YOU for Userify. I don't know how you guys are making any money yet (or how you plan to), but you make my life so much effing easier!

Thank you, and you're totally welcome!

We're racing against time and 4,000% growth... Userify[1] is also offering an enterprise[2] version for customers that require it for compliance reasons.. and working on some really helpful upcoming features.

1. https://userify.com

2. enterprise[at]userify

I just checked userify (haven't heard of it before), looks like a useful tool. What I found a bit odd is that pricing page is nowhere to be seen and it doesn't seem like userify is one of those enterprise 'request a demo' kind of services.

It might be ok if I really really need it like right now but if I'm just 'oh looks cool, might be useful' and have no clue how expensive it might get (and not motivated enough to sign up and find it out then) then it's not great.

Not to speak for Userify, but assuming the prices are 'relatively similar' to other DevOps hosted service type tools, I'd be happy to pay.

The ability to arrange user by projects, and then roles in projects has been awesome. You could have devs who have full access to development servers for project #1, limited access to staging, and no access to production, while having totally different levels of access on a different project, etc. You don't have to organize that way, but you basically get an org structure of a top level group that contains any number of child level groups (so, two levels of organization).

The total number of options are limited, I would say Userify is focusing on doing their one thing (or few things) really well.

Paired with Puppet scripts, AWS, Deploy Bot, and Laravel Forge, I'm able to deploy PHP, Java, Python, and Node.js apps to all kinds of funky server configs and access management is not even thought about any more. Just put the right keys in the provisioning scripts and the server gets the correct SSH user list and user keys.

Also, being able to centrally revoke privileges is also very awesome as contractors come into and out of projects.

Wow, thank you for that awesome review!

The parent was correct that a pricing page was missing; fixed! https://userify.com/

And to use 5TB in a month you have to sustain over 15 megabit/sec 24 hours a day. DO is betting on you not doing that (and in aggregate nobody does), because in any other context sustaining 15 megabit is about as expensive as Amazon. Quote a 100 megabit drop some time if you don't believe me.

If every VPS on Digital Ocean and Linode actually used their quota, they wouldn't have uplink capacity to support it and the network would fail. That's overselling. Numbers that high are extensive overselling. Linode has 40 Gbit links (at least they used to), and give 2 TB (6 Mbps) to each small Linode, meaning about 7,000 Linodes actually using the quota would saturate the link. They have a few more than that. Do the math.

The bandwidth quotas are sales stuff so that you will say exactly this in threads like these, and it's amazing how well it works.

The oversubscription model actually works, just like insurance does. We consume massive amounts of bandwidth on a continuous basis due to thousands of servers checking in. (We're not spikey in our bandwidth or CPU -- we're surprisingly rigid at 87% to 90% continuous.) We're anomalous, which kinda proves the point.

What is ironic about your statement is that t2 CPU allocations are massively oversubscribed in the same way. (but far more aggressively!)

A former Linode employee, actually (why did you edit out the ad hominem accusing me of working for Amazon?), but my employment is irrelevant to pointing out that over subscription means you are not actually getting what you pay for. If everybody used it you wouldn't have it any more. It's not yours. You are deciding entirely on perceived instead of actual value.

And yes, AWS's CPU over subscription on these instances is important if your workload is CPU intensive. They're very quickly outgrown due to that but useful in a pinch.

over subscription means you are not actually getting what you pay for. If everybody used it you wouldn't have it any more. It's not yours. You are deciding entirely on perceived instead of actual value.

I don't agree; as long as you can always use it when you needed, you are getting what you paid for - that's actual. What you're talking about is the value in an hypothetical situation. Now, if you are actually impacted by the oversubscription, then the actual value is below what was promised, but that doesn't seem to be the case for jamiesonbecker's use of DO.

Like the parent said, it's just like insurance. You can get what you pay for, because almost everybody else is paying for something they don't use, just in case they need it—and there's no situation where everybody's "just in case" will happen at the same time. Aggregated traffic is predictable in a way that individual flows aren't; you can have enough surety to stake your business on giving some of your customers a throughput/latency SLA, without reserving any circuits for them—because you know that even the peak of your traffic's froth is predictably below your uplink's capacity, and that there's no way for this fact to change on a timescale small-enough that your staff can't cope with it.

To answer your question, 100mbit isn't much these days. This article says you can get transit at $0.63/mbit. http://drpeering.net/white-papers/Internet-Transit-Pricing-H...

Explain why despite wholesale internet prices collapsing, AWS continues to charge the same with virtually no yoy decline? Azure et al also do this fwiw. It's just gouging.

What? They decrease prices all the time... There have been something like 3 price decreases just this year.

AWS has not decreased bandwidth pricing in a long time

You can go even lower depending on the commit. I've seen $0.35/Mbit from the low-end carriers, $0.50 from mid-range.

Sure, 100Mbit/s circuits aren't very cost effective -- also, no one buys them. If your needs are a little larger, you can do a 1 gig commit on a 10 gig circuit in a well connected datacenter in a first tier region (e.g. Silicon Valley), and figure on paying between $1.50 - $2.00 per megabit at 95th percentile. Much better prices are available with some elbow grease applied. Add $350 per month for the cross connect and $500 for the installation.

That comes to $22,700 for a year of service if you don't exceed 1 gigabit/s at the 95th percentile.

In Amazon's bandwidth terms, that's 331 terabytes transferred, or $17,000 per month if all of that bandwidth is outbound. You would have to limit your average transfer to 110mbit/s in order to achieve cost parity.

So yes, bandwidth is incredibly expensive in AWS and you can easily do better in datacenter-land (or with cloud providers who bill differently), provided that you are operating at sufficient scale.

"provided that you are operating at sufficient scale"

Where "sufficient scale" probably means you're spending 100k+ a year on infrastructure. Let alone staff costs to support it. Also in your example you have no datacenter redundancy so it's 44k if you want to be in 2 DCs, which means "sufficient scale" is probably more like 200k+.

"if you don't exceed 1 gigabit/s at the 95th percentile"

That's a big if, the internet, startups and business in general are uncertain and it's often very difficult to know what your 95th percentile will look like in the future.

Also, it's unlikely that your network utilization is constant. A utilization rate of 50% would be ambitious.

So a better comparison is 44k for 1gig in 2 datacenters vs 102k for AWS (17/2*12), assuming you can predict your network utilization pretty well.

> "provided that you are operating at sufficient scale"

Nothing says you have to do it all on your own. Let the DC operator do the heavy lifting, all you have to do is rent a dedicated server, VPS or colocate. Total spend to sufficient scale: a few dollars - a few grand per month.

> That's a big if, the internet, startups and business in > general are uncertain and it's often very difficult to > know what your 95th percentile will look like in the > future.

It's really not that hard. Look at historical usage and assume it continues as before. If you are worried have a spare dedicated server or two with 1G/10G interfaces.

> Also, it's unlikely that your network utilization is > constant. A utilization rate of 50% would be ambitious.

This really is a moot point as you effectively only pay for peak usage with 95th percentile billing. Alternatively you can pay per terabyte of traffic, which is effectively average usage. Average usage also correlates strongly with peak usage, so in the end it's all the same, give or take a constant.

> So a better comparison is 44k for 1gig in 2 datacenters > vs 102k for AWS (17/2*12), assuming you can predict your > network utilization pretty well.

Nope. A proper comparison is about $1000 per month for two dedicated servers in two DCs versus $102k per year for AWS.

You think if every ec2 instance used their available upload capacity, simultaneously, that AWS regions wouldn't fail?

Isn't that the whole point of VPS pay a fraction because you you'll only use a fraction.

Even internal clouds, people migrate their data centres to VMware, xen, kvm etc... because most servers only use a fraction of CPU and network allocated to them.

When you do need full cpu or bandwidth, you know it and you build for it.