Both are going to compete fiercely on prices, so my money goes to whoever doesn't make it easier for CIA and NSA to assassinate people, much like how IBM made it easier for the Nazi to gas people.
In the longer term, I would expect both of rthem to be afraid of the many smaller local players in each country when they start to up their game in terms of hosting. People are starting to become wary of putting all your data in one cloud - whether it is a Google or amazon shaped one.
Well for now AWS still charges a large premium over discounted hosting companies such as Hetzner. Over the past years it doesn't seem like anybody is catching up though. AWS is still by far the most advanced cloud hosting solution. Some great features:
- ELB
- Cloudformation
- RDS
- ElastiCache
- The BOTO api client
- Snapshots (Also see snaptastic btw, https://github.com/tschellenbach/snaptastic)
OP is asking whether Amazon offers public addresses that will respond to DNS requests for end users, such as 208.67.222.222 and 208.67.220.220 (OpenDNS).
To my knowledge, they do not - Route 53 is made for domain administrators to use as their nameservers instead of their domain registrar's.
I honestly hate this type of response, there must now be one in any thread where any company is discussed in anything that could be construed as a positive light.
The problem with AWS reserved instances is their limitations on modification later http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ri-modify... -- if you grow and need more capacity, you're kinda stuck unless you have Amazon Linux AMI. It would be nice if you could upgrade RDS instance, for example -- but I don't think that is possible.
Cept it can be hard to plan for smaller companies. If your startup is a fail, you don't need anything. If it is bootstrapping, you might need a medium or large for a DB. If you are hitting it huge, you need all you can get.
The pricing system with reserved instances is fine for a stable business with 10% yearly growth, not good for a startup.
As a user I would totally just prefer a discount for using more, but I can also see why amazon doesn't want to have a bunch of hardware that is 1 year old and no one wants to use it (but couldn't they just adjust the price to make it valuable still? )
Sometimes. I've had poor luck selling reservations, especially if the instance type is no longer current (as often happens). I've had instances sit on the market for the better part of a year, at less than half "list price", and not sell.
If this is Amazon trying to compete with Google's pricing scheme, they missed. Google's sustained use discounts just happen, and give you a discount on whatever time you used. With Amazon's new pricing for reserved instances, any reserved instance you buy (upfront fee or not) commits you to pay an hourly fee regardless of whether the instance is running now.
They removed the light/medium reserved instance options where you paid a discounted usage rate, and now just have 3 ways of purchasing a heavy reserved instance.
This is going to raise prices for us on AWS because we won't be able to use reservations for the instances we autoscale so the current medium reservations we have are going to be on-demand now.
The author of that article (and their strange quotes) doesn't understand what Amazon just changed or doesn't understand Google's sustained usage discounts.
If anything Amazon just moved in the opposite direction of what Google is doing, not to copy it as the article implies.
What's it cost to run a website with a MySQL database on app engine? Starting with a free App Engine instance and a data store always had a propriety lockin.
Just go get a VPS or small dedi. Cloud services aren't really ideal for dedicated site hosting, they're really for people growing their own services. The main killer for anyone wanting to host infrastructure is that bandwidth is something like 20x dearer.
The price-wars between amazon, ms, and google are constant. They always tweak pricing to match the tweaks the other two make. Given this is no change to the norm, i don't see this as a referendum on google's position in the market. If revenue is changing, that may be an indication. But price changing by itself has little meaning at this point.
i'm absolutely in the aws camp. in my opinion (having tried both), aws is far superior to google. but the biggest difference in pricing, which no one has mentioned, is that AWS offers a free 1 year trial for developers, while google only offers 60 days. That's a big difference for indie developers prototyping a MVP.
This article read like a paid Google App engine advert.
They point to the announcement about RI pricing simplification as proof that "Google has finally arrived." No justification at all, and it isn't obvious how one relates to the other (in particular as the RI changes weren't a direct price reduction).
> “Google is making a dent and AWS is starting to feel the pain from a pricing perspective,”
But they didn't announce a price reduction this week!
> says one ex-Amazon employee who asked not to be identified because his current employer works with the cloud company.
Because they work for Google. Because this is a paid Google advert.
> But this spring, Google introduced a product called Sustained Use Discounts, and this made things much simpler.
I'd argue that the sustained use model makes it harder to calculate your ultimate bill, not easier. It has the same issues Amazon's now retired Light and Medium usage tiers had, how do you track if you're at 25%, 50%, 75%, etc utilisation? Keeping track of that is annoying and harder than you'd think in a real production environment with dynamic instances responding to load.
Amazon's new pricing model actually just scrapped something akin to sustained use billing because it was unpopular. You wouldn't know that reading the article.
> Maybe they didn’t want to draw attention to what’s obviously a reaction to Google
Maybe it was a reaction to customer feedback? I see no evidence that Google caused this change, and the fact that Amazon moved AWAY from Google's model is very telling. Light and medium utilisation RI are effectively Google's model, they just do it dynamically. Amazon hasn't added that or indicated that they will.
Do these people even understand Amazon's RI pricing structure, or Google's sustained usage, or anything at all?
> Byrne says that until Amazon changed its pricing, his company was toying with the idea of switching to Google, a move that would have saved Copper.io hundreds of thousands of dollars—until Monday, at least.
That literally makes zero sense. Heavy utilisation RIs have been available since forever, and are extremely price competitive with every other major competitor in the cloud infrastructure space (namely Google, and Azure).
If it would have saved you "hundreds of thousands" legitimately then you are mismanaging your account heavily.
> “When I saw the Amazon announcement, I said: ‘Great. We don’t have to consider moving.'”
Because, why? I don't understand what it is they think Amazon changed. Because that statement makes no sense. If you were mismanaging your account before, you'd still be mismanaging it now. Amazon moving payments will save you a little money (due to the interest on that money sitting in your accounts) but nothing like what is described here.
> When Google first got into the cloud business back in 2008, the company bet that people would want to run cloud applications on its infrastructure in highly specialized ways and not mess around with operating systems and virtual machines. That proved to be a bad bet.
That is an under-statement. Google thought that people wanted to be locked into their cloud APIs and initially at least have to completely re-write all of their software for Google's app engine. It was supremely arrogant on Google's part.
App engine now supports more standardised APIs and databases, but it remains more expensive to do than it should be. Azure's "web-sites" product is what Google App Engine should have been when it first got released, and even today I find "web-sites" more compelling as a bottled solution than Google App Engine (although VMs trump both).
> And now the company is fast catching up.
Is that a fact? Because both Google and Azure topped $1B in 2013, and from what I've been seeing Azure, not Google, has been growing to compete with Amazon in 2014. Microsoft has been leveragi...
Fine "Google Cloud Platform" then. Happy? They've changed the name a few times. If nitpicking is all you can criticise my post for them I must be doing something right.
People call Amazon's entire offering "EC2" all the time. People call Google's offering "App Engine" (because it existed first, and because initially it included more stuff than it does today).
It is pretty obvious that I was talking about Google's VM product ("Google Compute Engine") because this entire discussion makes no sense otherwise. You're being intentionally obtuse just to have something to whine about.
I'm pretty sure you went into this article looking for excuses to call it a paid Google placement. If I were reading your comment through a similar lens, I would call you an Amazon shill but I don't think either of these posts were paid for by Amazon or Google.
>> “Google is making a dent and AWS is starting to feel the pain from a pricing perspective,”
>But they didn't announce a price reduction this week!
My biggest issue with the article is no mention of MSFT or Azure. MSFT is a public company and by all accounts their hosted business appears to be growing very rapidly. So to not even mention what they're doing and how AMZN would also be concerned about them as competition makes the article kind of weak.
Azure is growing like mad, and in most corporations it's listed as 'AWS', 'Azure' and 'everyone else' when discussing the cloud. I've had multiple meetings where I was the only person who'd even heard that Google had an option...RackSpace comes up more often than Google. This article is pretty clearly paid for by Google, in a battle that they are really late in arriving to.
And given Google's reputation for support (fully warranted in my experience), it's going to take a lot for corporations to shift away from AWS. It's actually astounding how much market share Azure has gotten already in spite of AWS.
Your experience is not universal. Every single person I've talked to their company is deciding between amazon and google. Neither MS nor rackspace even come up.
I read that as saying that they work for a company that works with Amazon. Admittedly "the cloud company" is ambiguous, but I'd identify those three words with Amazon, not Google.
> That literally makes zero sense. Heavy utilisation RIs have been available since forever, and are extremely price competitive with every other major competitor in the cloud infrastructure space (namely Google, and Azure).
The big difference is that with AWS, I am probably "stuck" with an instance for 1-3 years, unless I want to take my chances with the hit-or-miss reservation marketplace. This is a really bad fit for some businesses that are still vertically scaling, but don't have deep pockets. AWS can still make sense for companies on a budget because of their great portfolio of services, but EC2 is by far the most inconvenient billing-wise.
For example, we reserved an m3.large RDS instance, which at the time left us more than enough breathing room. About four months went by, and traffic grew far faster than projections. We weren't hurting, but we now wanted to bump to a m3.xlarge to feel comfortable, even after a round of optimizations and profiling. We're stuck with that 1-year reservation's sunk cost, and have no use for the old m3.large reservation if we switch to m3.xlarge.
The other fun part was buying a 3-year m1.medium reservation, only to see the new m3 series instances come out before those three years were up. These new instance types have been landing pretty frequently.
Reservations shift a lot of "risk" on to the customer and away from Amazon. GCE's pricing model is harder to estimate your bill with, but I actually like the idea of not being "stuck" if I want to be economical. Automatically give me discounts for instances that I run often, but don't lock me down.
I can't see what's wrong on Amazon's side. It's like a business telling you "if you buy four coffes you'll get one for free"; and then you complain "but I would have to buy four coffees there!".
> I can't see what's wrong on Amazon's side. It's like a business telling you "if you buy four coffes you'll get one for free"; and then you complain "but I would have to buy four coffees there!".
To make that analogy more correct, you've got a finite amount of coffee money, and you'd be agreeing to drink the same coffee (size, flavor, quantity) for 1-3 years. If a better blend or different flavor arrived at the coffee shop, you'd be out of luck until the end of your contract. Unless you wanted to go outside of your contracted discount and pay the full rate (On Demand), while eating the sunk cost of the reservation fee. If you found your tolerance rising and wanted a bigger cup or more cups, you'd lose your discount.
It's not as simple as "give me an instance and give me a discount". It's a shifting of risk from AWS to the customer.
The risk to AWS is that you'll go the fancy coffee shop down the street and leave them holding 1-3 years of bulk coffee. You are getting discount for saying "No risk, I'll stick around until the coffee runs out". There is no "shifting the risk", it's explicitly the reason for the discount.
> There is no "shifting the risk", it's explicitly the reason for the discount.
Sure there is. I'm not going to get into potential motivations and logistics (since I as a customer don't care), but there is no doubt that reservations are for shifting risk to the customer and away from Amazon. It doesn't matter if they're offering a discount, that's really not the point I am trying to make. I'm more concerned with the "risk" I take on each time I reserve an instance.
The big gripe I have is that a reservation is made with the assumption that everything (quality of service, the needs of your application, customer service) stays roughly the same for the duration of the reservation. If the customer service stinks, or there are some really bad outages, I've sunk the cost and end up throwing the money away if I switch (we eat the expense instead of Amazon). If I want to upgrade my instance size and have mixed duration reservations, I've got that sunk cost to swallow. If I want to scale down, ditto. I risk throwing away money if I change plans.
In a perfect world, I'd like the option to either pay an upfront reservation fee for the best rates, or run On Demand and get a graduated discount curve like Google Compute. They'd still get have an incentive to stay put, and people or companies with lower budgets could avoid the large reservation fees. At the same time, you could still pay the upfront reservation fee if you wanted the best rate and didn't mind committing for a year+.
Actually, you can upgrade reservations as long as they are on the same machine type, so m3.large -> m3.xlarge -> m3.large is possible mid-year. m1 -> m3 is certainly a problem, but that shouldn't happen too often.
So as long as you stuck to m3 and you know you will be using some machines, you should be fine. A bigger problem I see is that most startups don't actually know where they will be in a year (or 3).
One solution would be to reserve compute units, instead of machines. You could reserve 16 m3 "units" and could use that as 4 x m3.large or 2 x m3.xlarge. Those are equivalent in hardware anyway (minus some fragmentation).
> Actually, you can upgrade reservations as long as they are on the same machine type, so m3.large -> m3.xlarge -> m3.large is possible mid-year. m1 -> m3 is certainly a problem, but that shouldn't happen too often.
There are a lot of caveats to this. Most importantly, you had to have purchased the reservations that you are going to combine at the same time. You can't mix and match durations. If you have one with 8 months left and another wix 6 months left, you're out of luck. So if you want this kind of thing, you've got to shell out a lot more at the same time to buy a group of reservations. That's kind of a pain to me.
> One solution would be to reserve compute units, instead of machines. You could reserve 16 m3 "units" and could use that as 4 x m3.large or 2 x m3.xlarge. Those are equivalent in hardware anyway (minus some fragmentation).
The fact that we're sitting here talking about making a huge investment that we may or may not end up using effectively is the thing I dislike about this reservation system. I'm going to have to shell out a good bit of coin upfront for all of this, and I am effectively locked in for at least a year. Unless I want to gamble on the reservation market.
This is my biggest gripe, as an AWS customer. Give me more simple billing and get out of my way. Don't lock me into this thing that resembles a contract with a cancellation fee.
> This article read like a paid Google App engine advert.
Yeah, that was my impression too. The fact is: Google is really afraid of AWS. Google knows that it's several quarters behind on cloud infrastructure deployment, and that the lack of presence in places like Australia is a problem.
That is why they're working like crazy to rush the 'miniclusters' project out of the door to try to capture market share in those regions before Amazon and Microsoft take everything and laugh on their way to the bank.
On top of that, Google's cloud offering is also behind in the tools and support department. Most of their projects are so behind schedule it's not even funny and will most likely miss their Q4 objectives.
Amazon and Microsoft have, today, much better offerings then Google will by the end of Q2 2015.
Disclaimer: I work for TAGA (The Arrogant Google Assholes)
So, you are saying that you work for Google, and this is your view of Google Cloud Platform?
Pretty strange to "bash" the company you work for - even if, in my humble view, this is honorable and PR should never give you a hard time for being honest like this (for me, Google just won points by having someone like you be so honest about his view).
If the article is a paid Google advert (which might be), then "Someone1234" can also be a newly created anonymous HN account used by an Amazon employee.
So Amazon has "simplified" billing by removing a number of complicated options, but simultaneously adding a bunch of new options.
I much prefer Google's approach where the discounts "just happen". Planning for reserved instances, even before AWS added the crazy light/medium/heavy workloads, was always stressful (do I take 3 years? do I take 1?).
You can try your luck on the reservation marketplace, but depending on your region, instance size, reservation duration, reservation age, you may not end up being able to get out.
My decision on which provider to use is based on trust. How can any developer trust that Google won't pull the rug out from under them. This can happen with AWS too but Google has a history of doing this. No matter how cheap I'll stick with AWS and the safer bet for my business, thanks!
54 comments
[ 3.5 ms ] story [ 124 ms ] threadTo my knowledge, they do not - Route 53 is made for domain administrators to use as their nameservers instead of their domain registrar's.
Reads too much like a Google marketing response to an Amazon announcement.
The pricing system with reserved instances is fine for a stable business with 10% yearly growth, not good for a startup.
As a user I would totally just prefer a discount for using more, but I can also see why amazon doesn't want to have a bunch of hardware that is 1 year old and no one wants to use it (but couldn't they just adjust the price to make it valuable still? )
They removed the light/medium reserved instance options where you paid a discounted usage rate, and now just have 3 ways of purchasing a heavy reserved instance.
This is going to raise prices for us on AWS because we won't be able to use reservations for the instances we autoscale so the current medium reservations we have are going to be on-demand now.
The author of that article (and their strange quotes) doesn't understand what Amazon just changed or doesn't understand Google's sustained usage discounts.
If anything Amazon just moved in the opposite direction of what Google is doing, not to copy it as the article implies.
They point to the announcement about RI pricing simplification as proof that "Google has finally arrived." No justification at all, and it isn't obvious how one relates to the other (in particular as the RI changes weren't a direct price reduction).
> “Google is making a dent and AWS is starting to feel the pain from a pricing perspective,”
But they didn't announce a price reduction this week!
> says one ex-Amazon employee who asked not to be identified because his current employer works with the cloud company.
Because they work for Google. Because this is a paid Google advert.
> But this spring, Google introduced a product called Sustained Use Discounts, and this made things much simpler.
I'd argue that the sustained use model makes it harder to calculate your ultimate bill, not easier. It has the same issues Amazon's now retired Light and Medium usage tiers had, how do you track if you're at 25%, 50%, 75%, etc utilisation? Keeping track of that is annoying and harder than you'd think in a real production environment with dynamic instances responding to load.
Amazon's new pricing model actually just scrapped something akin to sustained use billing because it was unpopular. You wouldn't know that reading the article.
> Maybe they didn’t want to draw attention to what’s obviously a reaction to Google
Maybe it was a reaction to customer feedback? I see no evidence that Google caused this change, and the fact that Amazon moved AWAY from Google's model is very telling. Light and medium utilisation RI are effectively Google's model, they just do it dynamically. Amazon hasn't added that or indicated that they will.
Do these people even understand Amazon's RI pricing structure, or Google's sustained usage, or anything at all?
> Byrne says that until Amazon changed its pricing, his company was toying with the idea of switching to Google, a move that would have saved Copper.io hundreds of thousands of dollars—until Monday, at least.
That literally makes zero sense. Heavy utilisation RIs have been available since forever, and are extremely price competitive with every other major competitor in the cloud infrastructure space (namely Google, and Azure).
If it would have saved you "hundreds of thousands" legitimately then you are mismanaging your account heavily.
> “When I saw the Amazon announcement, I said: ‘Great. We don’t have to consider moving.'”
Because, why? I don't understand what it is they think Amazon changed. Because that statement makes no sense. If you were mismanaging your account before, you'd still be mismanaging it now. Amazon moving payments will save you a little money (due to the interest on that money sitting in your accounts) but nothing like what is described here.
> When Google first got into the cloud business back in 2008, the company bet that people would want to run cloud applications on its infrastructure in highly specialized ways and not mess around with operating systems and virtual machines. That proved to be a bad bet.
That is an under-statement. Google thought that people wanted to be locked into their cloud APIs and initially at least have to completely re-write all of their software for Google's app engine. It was supremely arrogant on Google's part.
App engine now supports more standardised APIs and databases, but it remains more expensive to do than it should be. Azure's "web-sites" product is what Google App Engine should have been when it first got released, and even today I find "web-sites" more compelling as a bottled solution than Google App Engine (although VMs trump both).
> And now the company is fast catching up.
Is that a fact? Because both Google and Azure topped $1B in 2013, and from what I've been seeing Azure, not Google, has been growing to compete with Amazon in 2014. Microsoft has been leveragi...
Except for the fact it doesn't mention App Engine at all.
And Google does have a lot of other offerings besides App Engine.
Please go on about App Engine. Perhaps you should take a better look at Googles other offerings before you criticize.
People call Amazon's entire offering "EC2" all the time. People call Google's offering "App Engine" (because it existed first, and because initially it included more stuff than it does today).
It is pretty obvious that I was talking about Google's VM product ("Google Compute Engine") because this entire discussion makes no sense otherwise. You're being intentionally obtuse just to have something to whine about.
>> “Google is making a dent and AWS is starting to feel the pain from a pricing perspective,”
>But they didn't announce a price reduction this week!
Uh, they announced a change related to pricing, not a reduction: http://aws.amazon.com/blogs/aws/simplified-reserved-instance...
>> says one ex-Amazon employee who asked not to be identified because his current employer works with the cloud company.
>Because they work for Google. Because this is a paid Google advert.
They don't work for Google. They work for a company that works with Google. That could be one of thousands (?) that are hosted on Compute Engine.
And given Google's reputation for support (fully warranted in my experience), it's going to take a lot for corporations to shift away from AWS. It's actually astounding how much market share Azure has gotten already in spite of AWS.
In my experience the only times I've heard Google mentioned by companies as an option are small start ups.
Dang - are Wired beating your voting ring detection or are they paid promoted content?
The big difference is that with AWS, I am probably "stuck" with an instance for 1-3 years, unless I want to take my chances with the hit-or-miss reservation marketplace. This is a really bad fit for some businesses that are still vertically scaling, but don't have deep pockets. AWS can still make sense for companies on a budget because of their great portfolio of services, but EC2 is by far the most inconvenient billing-wise.
For example, we reserved an m3.large RDS instance, which at the time left us more than enough breathing room. About four months went by, and traffic grew far faster than projections. We weren't hurting, but we now wanted to bump to a m3.xlarge to feel comfortable, even after a round of optimizations and profiling. We're stuck with that 1-year reservation's sunk cost, and have no use for the old m3.large reservation if we switch to m3.xlarge.
The other fun part was buying a 3-year m1.medium reservation, only to see the new m3 series instances come out before those three years were up. These new instance types have been landing pretty frequently.
Reservations shift a lot of "risk" on to the customer and away from Amazon. GCE's pricing model is harder to estimate your bill with, but I actually like the idea of not being "stuck" if I want to be economical. Automatically give me discounts for instances that I run often, but don't lock me down.
To make that analogy more correct, you've got a finite amount of coffee money, and you'd be agreeing to drink the same coffee (size, flavor, quantity) for 1-3 years. If a better blend or different flavor arrived at the coffee shop, you'd be out of luck until the end of your contract. Unless you wanted to go outside of your contracted discount and pay the full rate (On Demand), while eating the sunk cost of the reservation fee. If you found your tolerance rising and wanted a bigger cup or more cups, you'd lose your discount.
It's not as simple as "give me an instance and give me a discount". It's a shifting of risk from AWS to the customer.
Sure there is. I'm not going to get into potential motivations and logistics (since I as a customer don't care), but there is no doubt that reservations are for shifting risk to the customer and away from Amazon. It doesn't matter if they're offering a discount, that's really not the point I am trying to make. I'm more concerned with the "risk" I take on each time I reserve an instance.
The big gripe I have is that a reservation is made with the assumption that everything (quality of service, the needs of your application, customer service) stays roughly the same for the duration of the reservation. If the customer service stinks, or there are some really bad outages, I've sunk the cost and end up throwing the money away if I switch (we eat the expense instead of Amazon). If I want to upgrade my instance size and have mixed duration reservations, I've got that sunk cost to swallow. If I want to scale down, ditto. I risk throwing away money if I change plans.
In a perfect world, I'd like the option to either pay an upfront reservation fee for the best rates, or run On Demand and get a graduated discount curve like Google Compute. They'd still get have an incentive to stay put, and people or companies with lower budgets could avoid the large reservation fees. At the same time, you could still pay the upfront reservation fee if you wanted the best rate and didn't mind committing for a year+.
So as long as you stuck to m3 and you know you will be using some machines, you should be fine. A bigger problem I see is that most startups don't actually know where they will be in a year (or 3).
One solution would be to reserve compute units, instead of machines. You could reserve 16 m3 "units" and could use that as 4 x m3.large or 2 x m3.xlarge. Those are equivalent in hardware anyway (minus some fragmentation).
There are a lot of caveats to this. Most importantly, you had to have purchased the reservations that you are going to combine at the same time. You can't mix and match durations. If you have one with 8 months left and another wix 6 months left, you're out of luck. So if you want this kind of thing, you've got to shell out a lot more at the same time to buy a group of reservations. That's kind of a pain to me.
> One solution would be to reserve compute units, instead of machines. You could reserve 16 m3 "units" and could use that as 4 x m3.large or 2 x m3.xlarge. Those are equivalent in hardware anyway (minus some fragmentation).
The fact that we're sitting here talking about making a huge investment that we may or may not end up using effectively is the thing I dislike about this reservation system. I'm going to have to shell out a good bit of coin upfront for all of this, and I am effectively locked in for at least a year. Unless I want to gamble on the reservation market.
This is my biggest gripe, as an AWS customer. Give me more simple billing and get out of my way. Don't lock me into this thing that resembles a contract with a cancellation fee.
Yeah, that was my impression too. The fact is: Google is really afraid of AWS. Google knows that it's several quarters behind on cloud infrastructure deployment, and that the lack of presence in places like Australia is a problem.
That is why they're working like crazy to rush the 'miniclusters' project out of the door to try to capture market share in those regions before Amazon and Microsoft take everything and laugh on their way to the bank.
On top of that, Google's cloud offering is also behind in the tools and support department. Most of their projects are so behind schedule it's not even funny and will most likely miss their Q4 objectives.
Amazon and Microsoft have, today, much better offerings then Google will by the end of Q2 2015.
Disclaimer: I work for TAGA (The Arrogant Google Assholes)
I much prefer Google's approach where the discounts "just happen". Planning for reserved instances, even before AWS added the crazy light/medium/heavy workloads, was always stressful (do I take 3 years? do I take 1?).