Dropping Prices Again-- EC2, RDS, EMR and ElastiCache | Hacker News Reader

Dropping Prices Again-- EC2, RDS, EMR and ElastiCache

aws.typepad.com

192 pointsjeffbarr71 comments

Does anybody have a history of AWS price decreases?

I have seen a lot of startup internal docs where they compare the costs of AWS with colo, other cloud etc. and not one of them took into account the discounts that Amazon applies. Chances are that over 3-5 years you will see a number of price discounts - one of the only providers that does this for existing purchased infrastructure, keeping your costs in line with savings from Moore's law rather than locking in for years.

I would like to know the complete discount history because there would be enough data points now to project estimated discounts forward

If you rent servers or buy colo from anyone else, you can ALWAYS renegotiate all prices based on a new level (if you've got a cage and want to expand to a floor, it becomes a negotiating point; you don't get stuck paying the higher price on that cage and a lower price on the floor.

Until very recently, the only Amazon discounts I've ever heard of were the Reserved Instances (which really only makes sense once you're familiar with your needs). Inability to negotiate pricing is one of the main reasons a lot of consultants still go with non-Amazon options -- the client sees the list pricing for EC2 and knows what markup you're applying, vs. another option (or your own cage/suite/floor/facility) where you can consolidate all the charges.

In case anyone believes this comment (that "until recently" there have even no price drops), I will throw in an alternative anecdote (as I'm at a restaurant and am not interested anyway in digging up alternative evidence: readers can do that themselves), Amazon has decreased their prices for all kinds of things many times; bandwidth quite often, servers a few times... I actually find myself misquoting Amazon's prices in conversations often due to having lost track of random price drops.

The one thing that hasn't dropped is where 99% of my costs are: the price of a PUT to S3.

We're talking about two different things -- price drops vs. discounts.

They drop prices across the board all the time. They don't do individual negotiated prices at time T which are different for one customer vs. another customer. (they just started doing this a few months ago, in the form of offering special services to some customers but not others, and now in terms of volume pricing discounts).

All hosting providers drop prices over time (mainly by increasing the amount of RAM per server at the same price, vs. dropping the pricing of packages) for some things (and tend to raise for power, cooling, and physical space). But everyone except Amazon usually also does individual negotiation of prices.

Amazon do negotiate individual prices in some circumstances.

I'm not entirely sure what circumstances it takes to make them negotiate, but the discounts I've heard about are significant. It doesn't seem to be based on the size of the deal though.

The only times I've heard of this were GovCloud (which costs more, but includes a lot of extra service) and the recent announcement of volume based pricing (which is sort of individual in a way).

Dropbox, Heroku, and Netflix would be the authoritative data points, I think. I've heard "won't do anything custom", but I never really pushed to find out about discounts. Even if they do exist, they're nowhere near as common as with every other provider (where either you get a marketing deal discount, or can just get a discount for the asking, especially early on, or as you grow).

If consultants can't do business by adding a line item for their costs and have to mark up what you buy then don't hire them.

There are a lot of resellers of various services with different levels of value-add. The reality of the hosting business is that different providers give different levels of service, and in a lot of industries you do not want your customers to know your costs, so breaking it out as a separate line item isn't always optimal. Businesses actively try to avoid their product being viewed as a commodity.

There are also purchasing people in big companies who are paid to negotiate down quotes. If you tell them there is only one price for EC2, and they can't get a 5% discount for a large order, they will in some cases have a perverse incentive to go to a more expensive (list price) provider who will give them a discount (so they appear to be doing their job), even if the total price ends up higher, or the service ends up worse. This is one of the many problems with enterprises.

I'm not saying which model is better, but it's definitely a factor with some clients.

I don't think there were price decreases of the hourly prices for EC2 instances - previous decreases were for traffic, or introducing reserved instances where you get lower price in exchange for commitment.

I think trying to factor in future price decreases in an ec2 cost model doesn't make sense - it is too little and unpredictable to worry about. Don't forget that if you reserve an instance for 3 years and Amazon drops prices, you'd get only partial savings on the hourly fraction of the cost. Even as I use Amazon for hosting (instances running full time) I stay away from the 3 yrs reserved instances because I want to be able to reevaluate my options more frequently.

They definitely dropped hourly instance prices. They started at $0.10/hr for a small instance.

Probably not for the Asia Pacific Singapore region.

Few months ago, the on demand hourly price for a small instance is 0.085, now it's 0.09.

There's definitely a significant decrease for reserved small instances.

Edit: Reserved micro instances also went up.

I am pretty certain that the price was 0.095, not 0.085.

But did those "internal docs" take the increasing power of colo machines (or rent) into account? The colo you buy next year probably is more powerful than the one this year. If you take AWS price decreases into account, you'd need to take increasing power (and decreasing prices) for colo into account.

Hetzner where I have servers increases the machine power (and sometimes decreases prices) every year.

http://www.hetzner.de/hosting/produktmatrix/rootserver-produ...

Most dedicated hosting solutions offer some bandwidth in their pricing and not a trivial amount either. 3TB a month is not uncommon which would cost an extra $350 a month on Amazon. If they want to compete vs dedicated hosts and it seems they are with all these price drops and reserved instances, they should take that into account, because right now they are just way too expensive.

We actually get 10 TB included... :)

Hetzner also has 10TB per month.

You get 100TB at 100tb.com... :)

I dropped my account with them because their network could never push data faster than a few MB/s. Just personal experience.

Bandwidth =! total data transfer

Amazon would only cost you an extra $350/month if you actually use that 3TB.

Not dropping the hourly rate for existing reserved instances leaves me feeling very short-changed.

Yes, I feel like an idiot now for just buying a bunch of reserved instances. By my calculations if you just bought a reserved instance you'd actually save money by rebuying it if you plan to run it heavily over the three year period!

I've been holding off on buying reserved instances because it looked like they were overdue for a price cut. Even now I'm not sure if I should go ahead and buy -- sure, I'd save money compared to the current on-demand instance prices, but is the 2012 reserved pricing cheaper than the 2014 on-demand pricing will be?

I really hope Amazon adds more transparency into the pricing so I can stop worrying about things like this.

I think your in luck. Looks like your existing reserved instances will see the savings as well.

"This customer was in the process of switching to 3-year Heavy Utilization Reserved Instances (seeing most of their instances are running steady state) for a whopping savings of 55%. Now, with the new EC2 price drop we're announcing today, this customer will save another 37% on these Reserved Instances."

I think that was carefully worded - note the "in the process of switching". The email sent to existing owners this morning says the new pricing will only apply to reserved instances purchased since March 6.

Came here to say the same thing. I sent an annoyed message to customer service. It can't hurt, maybe they'll get enough complaints to change their minds.

Slightly off topic, but...

I launched a little web app about a year ago to play around with AWS and to take advantage of a free trial. It has grown organically with absolutely no marketing and even got covered by Lifehacker. It's a filesharing app for groups and is used heavily by the education sector.

My free trial period expired a few months ago and I don't know if I can justify supporting the app. I've put it maybe 10 minutes of work since I launched it a year ago, and have not yet attempted to monetize. Anybody interested in taking this off my hands? Cash or something else? I'm open to ideas.

How much traffic?

You could try selling it on flippa.com

So i could just be dumb, but there is something about this pricing that doesn't make sense to me. All of this is based on micros in the Virginia region, and having them for 12 months.

On demand for 12 months: $170 = (0.02 24 * 365)

Spot instance for 12 months: $52.56 = (0.006 24 * 365)

Light utilization - reserved for 1 year: $128 = (($0.012 24 * 365) + $23)

Medium utilization - reserved for 1 year: $115.32 = (($0.007 24 * 365) + $54)

Heavy utilization - reserved for 1 year: $105.80 = (($0.005 24 * 365) + $54)

I get that the spot instance price may change, but it would likely always be lower than on demand and given that the others prices change, are they just the best option?

So...maybe the marketing words are getting me confused, but i thought that heavy would be more expensive? Or is this telling me that i'm located with other heavy users, therefore i'm getting less and i pay less?

Today we use a lot of micros, i'm starting to realize how much we could save if we switch...however whats a "better server" a heavy or a light?

From Amazon:

Light Utilization:

"Light Utilization RIs offer the lowest upfront payment of all of the Reserved Instance types. Along with this low upfront payment, you’ll receive a significantly discounted hourly usage fee. Light Utilization RIs allow you to turn off your instance at any point and not pay the hourly fee."

Medium Utilization:

"Medium Utilization RIs are the exact same Reserved Instances that EC2 has offered for the last several years. They have a higher upfront payment than Light Utilization RIs, but a much lower hourly usage fee. Medium Utilization RIs allow you to turn off your instance at any point and not pay the hourly fee."

Heavy Utilization:

"With this RI, you pay a little higher upfront payment than Medium Utilization RIs, a significantly lower hourly usage fee, and you’re charged that lower hourly rate for every hour in the Reserved Instance term you purchase."

Source: http://aws.amazon.com/ec2/reserved-instances/

Yeah i get the pricing of light and heavy etc....but it doesnt actually state if they are different in performance or if its just a pricing thing?

If its just a pricing thing, the difference between $23 - $54 to add an extra $15 / savings over a year seems reasonable to me...but if there was a performance hit i'd think again.

No, there's no performance difference. Utilization is "time the instance is turned on", not "level of CPU usage".

Low utilization, for example, would be an instance you only turn on once a day for 30 minutes to do a scheduled task. Most of the time, it isn't running.

It's only a pricing thing. (Think of it as a variable capex/opex tradeoff)

Source: AWS guys that I met a couple of weeks ago were repeating this like a mantra (I think they get a lot of support requests for it)

ah ok then. Well i guess thats good news. That will make our business a lot cheaper to run...

There isn't a difference in performance. It's a price difference because you're guaranteeing Amazon that you will use the instance (and pay for it even if you aren't using it).

With light utilization, you could pay the up-front fee and then never turn on the instance. Yes, Amazon gets your reservation fee, but gets no hourly usage money from you. By choosing heavy utilization, you are saying "Amazon, I'm going to guarantee you that I will pay you $X for this instance over the next (1|3) years." In response, Amazon says "well, if you'll guarantee me that you aren't going to up and move to Linode or shut down for the next (1|3) years, I will give you a discount."

Amazon needs to make sure that it has enough capacity that when a random person wants to launch an instance, there's a machine with the capacity to handle it. That means that Amazon needs to keep a lot of unused capacity around. If you're willing to say, "I guarantee that I will pay for every hour over the next three years" Amazon can give you a discount.

Think of it like a CD at a bank. You're telling the bank, "you don't have to worry about your reserves when it comes to this account since I'm committing to not withdrawing this money for a year." That has advantages to the bank because they don't have to include that account in their planning of how much excess money (capacity) they need to keep around for when people want to withdraw.

Amazon can plan their capacity differently (and more cheaply to them) when you will guarantee that you will keep using/paying for an instance. As such, they're passing the savings on to you. This is a great benefit to a company who has a web server or two and knows that they won't be eliminating their presence. It's also a help to someone like Heroku. While the elastic nature of EC2 allows Heroku to always meet demand for dynos, Heroku could probably plan that they would need at least X instances and then just pay for on-demand instances for spikes in usage.

Heroku is actually a wonderful case for how it helps Amazon. Let's say that Heroku uses 5,000 instances. That's decently sizable. If Heroku left Amazon for whatever reason, Amazon could find itself with hundreds of boxes of excess capacity that they were planning would be used by Heroku. By committing to "heavy utilization", Heroku is telling Amazon "don't worry. We will be on EC2 for the next (1|3) years. When you calculate how many instances might be created/destroyed, those 5,000 instances are safe for the next (1|3) years since we've committed to paying the hourly fee on all of them for that period of time."

Sorry, the point I was trying to show was:

Light - You pay very little up front, but more per hour. You can decommission your instance and not pay hourly.

Medium - You pay more up front, less per hour, and can still decommission your instance and not pay the hourly fee when it is decommissioned.

Heavy - You pay the same up front, but in exchange for a lower hourly fee, you agree to pay for every hour that is in the reserved timeframe. E.g. even if you decommission, you are still charged the hourly rate.

We're almost done updating over 2,000 prices from AWS clouds: http://blog.shopforcloud.com/2012/03/were-updating-over-2000...

You might prefer to use ShopForCloud.com to get cost estimates rather than use Excel spreadsheets. They have a lot prices to update, for example, EC2 has 7 clouds x 13 types of instances x at least 3 OS choices (Linux, Windows, Windows with SQL Server) x at least 4 purchase options (on-demand, light/medium/heavy-utilization reserved) => 1,092 individual prices! Just goes to show that they have pretty complicated pricing schemes.

It's fair to mention that maintenance of your infrastructure on AWS is far from negligible. This is not unique for private clouds.

Especially given certain unpredictability in instance and EBS performance.

EBS is like the Matrix part III and Star Wars Episode I... It is best if you ignore the rumors that it exists.

Must be kind of funny to look at their accounting reports.

"Thanks everyone for your hard work this week. We're making substantially less money than last week. Let's keep it up people!"

AWS would be making no money if they cost more than everybody else.

You can either sell one can of soda for $5 profit or you can sell ten for $1 profit.

Or you can sell 1 for $1 profit, and lose $4 dollars over one sell at $5.

Lower prices != more volume guaranteed, you have to find an equilibrium.

Of course. But I'm sure they have thought of that though before lowering their prices and their intention wasn't losing money.

Something tells me Amazon understands price elasticity pretty well.

AWS is a shared service, so as they add more customers, the overhead costs of administering the service continue to drop. Their priority is to keep growth going at a steady clip.

AWS is also a very granular service, so Amazon gets near real-time metrics re: usage, demand, etc for a variety of services. That allows them to analyze changes in usage patterns and correlate price signals to actual consumption of services.

My guess is they adjust their pricing to control growth. Too cheap, and demand will spike, requiring Amazon to grow the environment too quickly. Too expensive, and they'll have trouble making the continuous capital investments that they need to make while still making money!

EC2 pricing was so absurd that everyone was using spot pricing for 50-80% off. This adjustment just gets more people back on the standard pricing scheme and makes tracking financial easier for Amazon.

Off-topic but relevant: does anyone have a good up-to-date tutorial on how to get started with cloud computing? Ie different alternatives, and pros/cons of each. And some info on how to get started doing common tasks.

Cloud computing could mean a lot of things but there was recently an article on HN that did a pretty good job at describing the different types of hosting: http://blog.pinboard.in/2012/01/the_five_stages_of_hosting/

I wrote a basic primer for django on EC2 to take advantage of the free tier last week here: http://eddychan.com/post/18484749431/minimum-viable-ops-depl...

Thanks, I'll need this in a few weeks.

https://www.coursera.org/saas/class/index might be worth looking into.

If only mobile phone service providers followed the same path...

I'm anxiously waiting for that moment where cloud storage becomes cheaper than buying a hard drive. Only at that point will consider I going all-cloud.

Google Docs storage is $0.25 / GB * year. A random 1TB drive on Newegg was $0.17 / GB. Depending on you utilization and amortization schedule, it may be worth it.

$170? That's a damned expensive 1TB HDD. Most are under $130, and that's still inflated due to Thailand.

Also, 1TB hasn't been the sweet spot for a while. 2TB drives are coming in under $140 now, $0.07/GB.

3TB squeaking in just under that too, $0.65/GB:

http://ncix.com/products/?sku=67447&vpn=WD30EZRS&man...

Before the Thailand floods, I was eyeing that drive for $130 ($0.04/GB) and thinking "can't go wrong waiting until I really need it."

And obviously, comparing to cloud strictly on price, you can divide that cost by the number of years you expect to keep that drive... so then it likely drops to less than 1 cent per GB per year.

*$0.065/GB

> $170? That's a damned expensive 1TB HDD.

Google's price includes the electricity to run the disk, any enclosures, network connectivity, whatever CPU is needed to front end it, dealing with any warranty & maintenance issues, backups and backup media and who knows how many other items in addition to the bare metal.

If you want to do an apples to apples price comparison then some of those factors should also be factored in to the NewEgg price!

Not to mention redundancy - Google certainly do not store your docs on one hard drive.

I'm waiting for the moment where a cloud machine with extremely modest specs and traffic becomes as cheap as running a junkyard PC from my basement.

Junkyard PC's are always cheap, but I keep thinking with economies of scale they'd be able to carve out 64MB RAM/2GB disk/100MHz for $10-20/year.

lowendbox.com is your friend for that kind of stuff

As a heavy EC2 user, you have got to think about the performance ramifications. EC2 EBS disks are abysmal in terms of performance.

I keep wondering why Amazon doesn't fix this. The problem isn't so much the baseline performance which isn't stellar but can be designed to, but rather the massive variance in performance that arbitrarily changes moment to moment.

Arbitrary configuration networked disk is Hard, probably harder than anything else AWS does. Enterprise still spends a lot on storage (generally big hardware RAIDs, dedicated storage area networks, etc.). 1 GB of enterprise storage might cost 50-100x the base per-GB cost of the cheapest SATA disk, by the time you include all the costs. No AWS customer wants to pay that much of a premium.

You can often build a storage system for a very specific purpose more more cheaply (maybe 3x the base drive cost?), like the Backblaze pods, or just figure the cost of adding 4-8 x SATA drives and software RAID 10 to each of your servers, but then you get to try to solve the filesystem problem, and have weird performance characteristics. Plus, if you think EBS reliability is bad (it is), wait for the joy of statistically significant number of spinning rust drives in anything not designed with enough redundancy, and where some failures are correlated (trays failing at the same time, controller bugs, ...).

One of Google's big advantages was always their Google File System. And GFS is a lot better for the specific loads in search than for the kind of loads applications like gmail impose.

Local instance storage is really the only performant and cost effective thing. You can hang a lot more direct attached disk off your own server than off any AWS instance (1.7TB).

The other area where AWS kind of falls down is "mad RAM" servers. I can buy 288GB RAM servers (18-slot) cheaply now (well, 144GB is cheap, 288GB is less so), and 2TB RAM is available (64-slot). AWS tops out at 68GB for about 5x the lease payments on these servers (so maybe 2x the loaded cost after Reserved Instance discounts)

Moving from MySQL on EBS to a Cassandra cluster on locally-attached ephemeral storage was the first time I was happy with a database on EC2.

EBS is convenient as a permanent root for instances, but for anything that actually uses the disk, it's a pain train. I think of it as a large "pen-drive" for the cloud these days.

It's a hard problem to solve, because you're selling disk based on capacity. Problem is, capacity is easy to figure out, but capacity at a specific performance profile is hard.

In an enterprise datacenter, a "fast" Tier-1 or Tier-2 SAN runs anywhere from $10-40/GB/Year to the end user. A "cheap" (and much slower) Tier-3 SAN runs anywhere from $1-15/GB, depending on utilization, availability, etc.

I don't think even those rates are possible with Amazon (with the exception of S3 object storage), because they don't know anything about their users. In the enterprise, you segment your users based on anticipated peak performance requirements. Amazon has no idea what your performance profile is -- and no idea whether it will change tomorrow.

Wondering that too. I really like the idea of having hosting in the cloud. But I'm not going to move anything critical there until the storage gets better and more predictable. All these stories you read about having to stripe several EBS volumes together to get even half-decent performance seems to add a lot of complexity/extra engineering for something that ought to just work out of the box. Seems to me that its unlikely that EBS will be any good until their entire infrastructure gets upgraded to 10gbit ethernet.

Meanwhile we still have 1 1/2 year left on our current 3 year datacenter contract(we run on 6 server vmware setup hooked up to a hitachi SAN on which we rent storage space). So Amazon have around 1 year left to come up with some good reasons for us to switch. Otherwise we will probably just grab another 3 years with the current provider.

From hearsay (I don't have direct knowledge) EBS is an EMC SAN solution. So Amazon can't really do anything directly to fix this. They probably have put pressure on EMC for years.

Well, it certainly won't come from minor price reductions on Amazon. A dedicated server from here [1] costs €299/month for 45TB. The same on Amazon would cost ~€3400/month for storage alone. That isn't even factoring in the fact that the dedicated server also comes with CPU and traffic which you have to pay for separately on Amazon which would make it much more expensive still. Amazon's storage is also notoriously slow. Amazon would have to reduce their prices by more than a factor of 10 before they come even close.

[1] http://www.hetzner.de/en/hosting/produkte_rootserver/xs29

So long as the cloud-storage provider also needs to buy hard drives, that day will probably never come.

Even if the provider can get better discounts buying in bulk (which won't be very much, storage is a commodity with thin margins), he's paying for a lot more than just the storage hardware.

Reply on news.ycombinator.com