Amazon EC2 Dedicated Instances | Hacker News Reader

Amazon EC2 Dedicated Instances

aws.typepad.com

66 pointsjeffbarr37 comments

There are some features which read, essentially, "Check this box if you're an enterprise." This is one of them -- it addresses a very narrow slice of security concerns of the sort that people with a checklist need to address and most of us are (by implication) OK with ignoring. You should generally expect checking boxes like this to be very expensive, because anyone who checks it is not primarily motivated by money. (Similarly, Amazon isn't implementing this so they can get $10 per hour. That is absence of whee. Whee is landing the kind of client who requires this sort of thing to consider EC2 as an option. Their needs laugh in the general direction of $70k a year.)

See also: compliance, auditing, etc, etc.

Some people actually need the performance guarantees, too. I've heard a lot of people whine about the unpredictable performance of I/O caused by shared tenancy, specifically people running decent sized cassandra/hbase installations or multiple database hosts. If you're willing to pay for a higher guaranteed minimum baseline, you could be in this market without a butthead ISO 9000 checklist.

But Amazon can, and still will put multiple (of your) instances onto the same box. So performance is still not as predictable as a dedicated machine.

I understand a fractional machine being wasted, but I'm a little confused as to the pricing. $10/hour => $10 x 24 x 30 = $7200/month?

I understand its useful if your running thousands of instances but I don't see the value in it as a public service. Wouldn't it be much cheaper for small clients to simply take your business elsewhere where you can get high-end servers in the $2000 range?

This looks like a fixed cost per region, rather than a cost per instance. If so, it might actually end up being quite cost-efficient.

"we add a $10/hour charge whenever you have at least one Dedicated Instance running in a Region. When figured as a per-instance cost, this charge will asymptotically approach $0 (per instance) for customers that run hundreds or thousands of instances"

The point here is customers concerned about security, like the government. It's a good call to make the public aware of the feature for when these big customers are evaluating AWS.

Yes, nobody with less than 50 servers will want to even think about this, but that's not who it's targeted at. Plus, anyone who cares about security to the extent of isolating VMs physically won't care too much about the cost.

I was all excited until I saw that $10/hour charge. I guess they're targeting high end users, but it would have been nice if they had priced this within the budget of a 'normal' person. This doesn't seem to be in line with their pricing strategy for all the other AWS services.

Really exciting. By far my biggest hope for EC2 is scriptable on demand dedicated hardware. I even flirted with the idea of launching a company the did just this.

The pricing seems designed to make it a feature only usable by really big customers though, which seems like a big mistake that hopefully they can do something about. Maybe by adjusting their server configs or something.

Still keep getting impressed by the AWS team.

I believe Softlayer offers an API where you can script some aspects of ordering and managing machines. I haven't spent much time on it though, but I suspect it isn't scriptable in the sense that EC2 stuff is.

You can order and provision (with hardware config, OS installation/image) via their API.

I just find that the time to provision is much slower than EC2 (10-20 minutes vs. 1-5 minutes).

We built a system for doing exactly this type of provisioning of physical hardware at Digg:

http://clusto.org/

The pricing model seems like it could be a big win for heroku, phpfog, and anyone else arbitraging EC2. Edit: My reasoning is that currently EC2 performance sucks and is reliably inconsistent, and if you can spend a fixed price to ensure that you're actually getting what you're paying for, as an ec2 arbitrator, your clients are going to be a lot more satisfied

Edit 2: Ah i see, i misunderstood

I think you're misunderstanding the pricing model. AWS is charging $10 per hour on top of the regular per instance charges, because if you have a small number of instances they may have to dedicate entire machines to your small instances. As you increase the number of dedicated instances they can put more of your instances on your dedicated machines, recovering the wasted resources.

Heroku and other multitenant PaaS services have no reason to use dedicated instances... customers concerned about running on multitenant platforms wouldn't even consider them in the first place.

How would this be useful to them? I don't think they care if their instances are on the same box or not. Also, how is what Heroku and PHP Fog arbitrage? They're just providing a service on top of another service; they don't even work with multiple markets.

My main concern with AWS is still EBS and the network latency. While this would prevent others on the same box from starving you of resources, EBS could still be affected by a neighboring box hammering the network.

The biggest advantage to this (other than security) is that you won't need to share I/O or CPU time with other people, resulting in much more consistent and predictable performance across the board.

Now, if only they could provide EBS this way...

But you have to share it with yourself once you launch multiple instances, which you will if you are paying $10/hour.

I don't see how this makes sense for Amazon.

The $10/hr charge seems to be a flat rate no matter how many dedicated instances you have in the region. Thus I can start 10000 "small" instances and pay pennies per hour + $10/10000. Surely a small instance does not cover the cost of an entire physical machine.

This seems like an economical way to DoS EC2

The crucial point is that you can still have multiple VPS's on a single machine, but you're guaranteed they'll all be yours. That is, the only difference with the dedicated instances is that there won't be an instance of someone else running on a machine you're also using.

I think most people wouldn't want to enable this unless you're required to; chances are that when one instance needs all the resources, the others are busy too, and then suddenly you have all your instances competing for resources on the same machine.

I think you missed this part: "launching a set of instances with dedicated tenancy does not in any way guarantee that they'll share the same hardware (they might, but you have no control over it)". In other words, when launching many small/medium instances they will make sure to stuff them onto only a few physical machines. So dedicated tenancy does not mean no virtualization - it just means that you don't share hardware with other users. So this is really just a way to accommodate whatever privacy policies that might require such thing.

> Surely a small instance does not cover the cost of an entire physical machine.

Who said you'd get a machine per instance? All they're promising is that no one else's instances will be on the servers hosting your instances.

I'm pretty sure the $10/hr ensures no-one _else's_ VMs run on the hardware your VM is running on. If you spin up 10 small instances, I think they'd happily put them all on the same hardware, just like that might have done anyway, and pocket the $10/hr for no cost to Amazon (except the minor administrative cost associated with getting that all right).

Sounds like this might reduce redundancy - if all your instances tend to be on the same machine, a single failure could take them all out...

It doesn't work like that. There's some intelligence built in to the instance placement algorithm to guard against this.

The $10/hour charge is to make up for the fractional machine you are wasting. That doesn't change as you add more instances. There is always just one fractional machine being wasted.

They only let you launch 20 EC2 instances unless you request an increase[1]. I assume you need a damn good use case before they're up your limit to 10,000.

[1] - http://aws.amazon.com/contact-us/ec2-request/

I think it is more of a credit check than a use case you need.

Ya, it's just this. And also a check against spam farms.

Even if you started up 10,000 Dedicate Instances you would not tie up 10,000 physical machines. A Dedicated Instance won't share hardware with another customer's VM but it could share hardware with your other VMs. It would probably be beneficial to you if they did, as the post says:

"It is important to note that launching a set of instances with dedicated tenancy does not in any way guarantee that they'll share the same hardware (they might, but you have no control over it)."

It sounds to me like they are saying they really don't want to sell these Dedicated Instances.

10$ / hour is HUGE

Amazon must be assuming that anyone seriously interested in this service spends well over $88,000/year on servers, which is likely the case.

As someone else explained, that's per-region. So if you have a lot of servers in a region, that cost is amortized over all of them.

That's $86,700 per year, not cheap to say the least. Is that kind of setup good enough for some level of PCI compliance? I think they still require that your servers are physically secure.

Yes Amazon has PCI compliance now, well the parts they need. The physical security is fine too, their datacentres are well secured. So this type of workload is possible.

Not if your HUGE

What about IPv6?

Reply on news.ycombinator.com